ISEC24 Information & Abstracts
- When: 10:00am - 1:00pm
- Where: G014
- Instructors: Bob O’Hara, Philip Mostert, Ron Tugonov & Kwaku Adjei
- Overview: The workshop will introduce integrated distribution modelling and the PointedSDMs software. This will include how to format the data, and fit and adapt models with different observation processes. The workshop will be interactive, so attendees will be able to practice fitting the models and get help working with their own data.
- When: 2:00pm - 5:00pm
- Where: G014
- Instructors: Daniel Turek & Wei Zhang
- Overview: Those who want to learn a general and extensible platform for hierarchical statistical modelling, including custom distributions for ecological models and the new features of HMC and Laplace approximation. Attendees should be familiar with writing models in the BUGS language (JAGS, BUGS or NIMBLE).
- When: 10:00am - 5:00pm
- Where: G011
- Instructors: Eric Pedersen, Fonya Irvine, Natasha Klappstein & Théo Michelot
- Overview: This one-day workshop will focus on the application of generalised additive models (GAMs) to animal movement data, including theoretical background, implementation, and interpretation. Participants are expected to be familiar with simpler regression models and with the statistical software R.
- When: 10:00am - 5:00pm
- Where: G022
- Instructors: Fanny Dupont, Marco Gallegos Herrada, Marie Auger-Méthé & Vianey Leos Barajas
- Overview: he workshop aims to demonstrate how hidden Markov models (HMMs) can be used to classify behaviours and identify behaviour-specific habitat associations using a range of movement and biologging data.
runMCMCbtadjust: using Bayesian MCMCs
- When: 10:00am - 1:00pm
- Where: G037
- Instructors: Frédéric Gosselin
- Overview: Presentation of a new R package called
runMCMCbtadjustthat can help R users of MCMC models having a more efficient, quality oriented use of these types of models while saving analyst’s and potentially computer time.
Hmsc
- When: 10:00am - 5:00pm
- Where: G014
- Instructors: Otso Ovaskainen & Gleb Tikhonov
- Overview: In this workshop, the participants learn how to apply the Joint Species Distribution Modelling Framework of Hierarchical Modelling of Species Communities (HMSC) with the R-package Hmsc-R. HMSC can be used to model multispecies data on species occurrences/abundances as a function of environmental, spatial and temporal predictors, species traits and taxonomies/phylogenies. The workshop includes brief lectures introducing the conceptual and statistical background of HMSC. The main part of the workshop consists of computer demonstrations showing how to apply Hmsc-R to various types of data, such as hierarchical, spatial and temporal sampling designs, as well as dataset including traits and taxonomical/phylogenetic information. Compared to earlier HMSC courses organized in 2020 and 2022, this course includes as new elements instructions on how to fit HMSC up to 1000 times faster with the newly released high-performance computing module Hmsc-HPC, as well as new methods on how to link HMSC outputs to niche theory by summarizing them into shared and idiosyncratic responses of the species to measured and latent predictors. Participants working already with HMSC on their own data are provided one-to-one support.
RTMB – Automatic differentiation in R
- When: 10:00am - 5:00pm
- Where: G018
- Instructors: Anders Nielsen, Ben Bolker
- Overview: This workshop will introduce participants to RTMB. RTMB is the latest iteration of TMB (Template Model Builder), a widely used tool for flexible statistical modeling. RTMB’s novel feature is that models are specified in plain R code (rather than C++ code or another special modelling language). Once the model is specified it evaluates function values and derivatives as efficiently as if the model were written in C++. Calculations in RTMB are supported by automatic differentiation, sparse matrices, delta-method calculations, and efficient Laplace approximation, which makes RTMB ideal for maximum likelihood inference in small and large models with and without random effects. Models specified in RTMB can also sampled with MCMC by linking them to Stan. In the course examples and exercises will span non-linear regression, generalized linear models, random effects, state-space, and spatio-temporal models with flexible spatial mesh structure (as known from INLA).
Plenary 1
- Prof. Rachel McCrae (Lancaster University)
- Conservation translocations: a collection of statistical developments
- Room G043
- 11:00-12:00am
14:00: Estimating hotspots using an automated pipeline and large-scale integrated species distribution models [Talk ID: 123]
Philip S Mostert (Norwegian University of Science and Technology)+; Kwaku Peprah Adjei (Norwegian Institute for Nature Research); Ron R. Togunov (Norwegian University of Science and Technology); Sam Perrin (Norwegian University of Science and Technology); Robert B O’Hara (NTNU); Anders Finstad (Norwegian University of Science and Technology); Joseph Chipperfield (Norwegian Institute for Nature Research)
14:15: Can integrated species distribution models improve spatial model transfer? Antarctic ‘islands’ as a case study [Talk ID: 160]
Charlotte R Patterson (Queensland University of Technology)+; Xiaotian Zheng (University of Woollongong); Kate Helmstedt (Queensland University of Technology ); Justine Shaw (Queensland University of Technology)
14:30: An integrated metapopulation model to quantify environmental drivers of demographic rates in a migratory bird of conservation concern [Talk ID: 256]
Alexander Schindler (University of Saskatchewan)+; Anthony Fox (Aarhus University); Alyn Walsh (National Parks and Wildlife Service); Larry Griffin (Wildfowl and Wetlands Trust, ECO-LG Limited); Mitch Weegman (University of Saskatchewan)
14:45: Using two-species Integrated Population Models to estimate competitive outcomes and coexistence mechanisms in stage-structured systems [Talk ID: 88]
Matthieu Paquet (SETE CNRS)+; Frédéric Barraquand (Institute of Mathematics of Bordeaux, University of Bordeaux, CNRS, Bordeaux INP)
14:00: Natal dispersal timed to the lunar cycle in a nocturnal bird of prey [Talk ID: 112]
Ying-Chi Chan (Swiss Ornithological Institute)+; Matthias Tschumi (Swiss Ornithological Institute); Fränzi Korner-Nievergelt (Swiss Ornithological Institute); Martin Gruebler (Swiss Ornithological Institute)
14:15: Developing statistical inference methods for animal movement modelling [Talk ID: 76]
Eloise Bray (University of Sheffield)+
14:30: The Efficient Modelling of Individual Animal Movement in Continuous Time [Talk ID: 215]
Dominic P.D. Grainger (University of Sheffield)+; Paul Blackwell (University of Sheffield)
14:45: Using a stochastic movement simulator to estimate wild bee’s pollination contribution in heterogeneous agricultural landscapes [Talk ID: 99]
Anouk Glad (INP-ENSAT, UMR 1201 Dynafor, Auzeville-Tolosan)+; Sylvain Moulherat (OïkoLab, TerrOïko, 2 place Dom Devic, Sorèze,); Emilie Andrieu (INRAE, UMR 1201 Dynafor, Castanet-Tolosan); David Sheeren (INP-ENSAT, UMR 1201 Dynafor, Auzeville-Tolosan); Annie Ouin (INP-ENSAT, UMR 1201 Dynafor, Auzeville-Tolosan)
14:00: Beavers increase the dissolved organic carbon in the water by enhancing terrestrial litter subsidies and primary producers abundance [Talk ID: 55]
Leonardo Capitani (WSL (Swiss Federal Institute for Forest, Snow and Landscape Research) & Eawag (Swiss Federal Institute of Aquatic Science))+; Valentin Moser (WSL (Swiss Federal Institute for Forest, Snow and Landscape Research) & Eawag (Swiss Federal Institute of Aquatic Science)); Francesco Pomati (Eawag (Swiss Federal Institute of Aquatic Science)); Anita Risch (WSL (Swiss Federal Institute for Forest, Snow and Landscape Research))
14:15: Conservation Planning for Nesting Grassland Birds in Agricultural Landscapes [Talk ID: 58]
Allison Binley (Cornell Lab of Ornithology)+; Scott Wilson (ECCC); Sara Barker (Cornell Lab of Ornithology); Amy Johnson (Smithsonian’s National Zoo & Conservation Biology Institute ); Joseph Lawrence (Cornell University); Daryl Nydham (Cornell University); Justin Proctor (Smithsonian’s National Zoo & Conservation Biology Institute); Kristan Reed (Cornell University); Bernadette Rigley ( Smithsonian’s National Zoo & Conservation Biology Institute); Joseph Waddell (Cornell University); Amanda Rodewald (Cornell Lab of Ornithology)
14:30: Expert elicitation on the effectiveness of potential measures to compensate for impacts of offshore renewables [Talk ID: 241]
Adam Butler (Biomathematics and Statistics Scotland)+; Anastasia Frantsuzova (BioSS); Kate Searle (UK Centre for Ecology and Hydrology); Esther L Jones (Biomathematics & Statistics Scotland); Francis Daunt (UK Centre for Ecology and Hydrology); Eleanor Skeate (ABPmer); Bob Furness (MacArthur Green); Annette Fayet (NINA); Maria Bogdanova (UK Centre for Ecology and Hydrology); Tone Reiertsen (NINA); Ana Couto (BioSS); Charlotte Regan (UK Centre for Ecology and Hydrology); Oliver Leedham (UK Centre for Ecology and Hydrology)
14:00: Evaluating mark-resight survey design performance using simulation: endangered Steller sea lions as a case study [Talk ID: 33]
Amanda J Warlick (NOAA Fisheries AFSC)+; Brian Fadely (NOAA Fisheries AFSC); Peter Mahoney (NOAA Fisheries AFSC); Sharon Melin (NOAA Fisheries AFSC); Tom Gelatt (NOAA Fisheries AFSC); Kim Raum-Suryan (NOAA Fisheries AFSC); Sarah Converse (U.S. Geological Survey, WCFWRU, School of Aquatic and Fishery Sciences & School of Environmental and Forest Sciences, University of WA)
14:15: Designing an effective close-kin mark-recapture survey for Pacific walrus [Talk ID: 104]
Eiren K Jacobson (Centre for Research into Ecological and Environmental Modelling, University of St Andrews)+; Mark Bravington (Estimark Research); David Miller (NA); Irina Trukhanova (North Pacific Wildlife Consulting, LLC, Seattle, Washington and US Fish and Wildlife Service, Marine Mammals Management, Anchorage, Alaska); Rebecca Taylor (US Geological Survey, Alaska Science Center, Anchorage, Alaska); William Beatty (US Geological Survey, Alaska Science Center, Anchorage, Alaska and US Geological Survey, Upper Midwest Environmental Sciences Center, La Crosse, Wisconsin)
14:30: Data fusion models for combining sparse data at different scales [Talk ID: 224]
Ana Couto (BioSS)+; Fergus J Chadwick (University of St Andrews); David L Miller (BioSS/UKCEH); Thomas Cornulier (Biomathematics and Statistics Scotland); Janice Scheffler (UKCEH); Peter Levy (UKCEH); Jackie Potts (BioSS)
15:30: Survival of rings and birds at high elevations [Talk ID: 50]
Fränzi Korner-Nievergelt (Swiss Ornithological Institute)+; Sebastian Dirren (Swiss Ornithological Institute); Anne-Cathérine Gutzwiller (Swiss Ornithological Institute); Carole Niffenegger (Swiss Ornithological Institute); Elisenda Peris Morente (Catalanian Institute of Ornithology); Claire Pernollet (Swi); Christian Schano (Swiss Ornithological Institute); Irmi Zwahlen (Swiss Ornithological Institute)
15:45: A conditional approach to estimate and account for correlations between long-term pairs in mark-recapture studies [Talk ID: 131]
Simon Bonner (University of Western Ontario)+; Alex Draghici (University of Western Ontario)
16:00: Continuous-time Models of Avian Survival Utilizing Staggered-entry Encounter Data [Talk ID: 204]
Todd Arnold (University of Minnesota)+
15:30: A novel method to census king penguins in UAV survey data via a Digital Elevation Model (DEM) [Talk ID: 43]
Tara Cunningham (University of Edinburgh)+; Stuart King (University of Edinburgh); Ruth King (University of Edinburgh); Norman Ratcliffe (British Antarctic Survey); Peter Fretwell (British Antarctic Survey)
15:45: Marginal negative effect of GPS tag equipment on survival and breeding success of Bonelli’s eagle [Talk ID: 62]
Lise Viollat (CEFE)+; Roger Pradel (CEFE); Alexandre Millon (IMBE); Cécile Ponchon (CEN PACA); Alain Ravaryol (La salsepareille); Aurélien Besnard (CEFE)
16:00: Comparing trend estimates from the North American Breeding Bird Survey and eBird Status and Trends [Talk ID: 141]
Orin J Robinson (Cornell Lab of Ornithology)+; Tom Auer (Cornell Lab of Ornithology); Wesley M. Hochachka (Cornell Lab of Ornithology); Alison Johnston (University of St Andrews); Daniel Fink (Cornell University)
16:15: Using Bayesian removal models to estimate residual population dynamics of an invasive predator over 20 years of control. [Talk ID: 163]
Albert Bonet Bigata (University of Aberdeen)+
15:30: Dealing with area-to-point spatial misalignment in species distribution models [Talk ID: 132]
Bastien Mourguiart (IFREMER)+
15:45: Forecasting global-scale marine species distributions: Combining Marine Ecosystem Models and Bayesian additive regression trees [Talk ID: 66]
Alba Fuster Alonso (Institute of Marine Sciences (ICM) - CSIC)+; Jeroen Steenbeek (Ecopath International Initiative (EII)); Jorge Mestre Tomás (Institute of Marine Sciences (ICM) - CSIC); M. Grazia Pennino (Spanish Institute of Oceanography (IEO) - CSIC); Xavier Barber (Operations Research Center, Miguel Hernández University (UMH)); Jose M. Bellido (Spanish Institute of Oceanography (IEO) - CSIC); David Conesa (Department of Statistics and Operations Research (VaBar), University of Valencia (UV)); Antonio López-Quílez (Universitat de Valencia); Villy Christensen (Institute of the Oceans and Fisheries, University of British Columbia); Marta Coll (Institute of Marine Sciences (ICM) - CSIC)
16:00: Spatio-Temporal Species Distribution Modelling [Talk ID: 7]
Sam Mason (UNSW)+
16:15: Species Distribution Modelling with Expert Elicitation and Bayesian Calibration [Talk ID: 95]
Karel J Kaurila (University of Helsinki)+; Sanna Kuningas (Natural Resources Institute Finland); Antti Lappalainen (Natural Resources Institute Finland); Jarno Vanhatalo (University of Helsinki)
15:30: Directly from data to statistical models for stock assessment: optimizing data input processes for more accurate and efficient modeling [Talk ID: 272]
Margarita M. Rincón (Spanish Institute of Oceanography, National Spanish Research Council (IEO-CSIC))+; Jamie Lentin (Shuttlethread); María José Zúñiga (Spanish Institute of Oceanography, National Spanish Research Council (IEO-CSIC)); Alfonso Pérez (Spanish Institute of Oceanography, National Spanish Research Council (IEO-CSIC))
15:45: Stochastic block models and network metrics to assess patterns of forced labor connectivity in between vessels and ports [Talk ID: 206]
Rocio Joo (Global Fishing Watch)+; Laura Osborne (Global Fishing Watch)
16:00: A single framework for synthesising forecasts seamlessly across time scales [Talk ID: 115]
Michael A Spence (Cefas)+
Plenary 2
- Dr. Marie Auger Methé (University of British Columbia, Canada)
- State-space models for animal movement
- Room G043
- 9:00-10:00am
Roundtable 1
- Developing pipelines for integrated distribution models
- Bob O’Hara (Norwegian University of Science and Technology, Norway)
- Room G014
- 1:00-2:00pm
10:30: The Motion Picture: Leveraging Movement to enhance AI Object Detection in Ecology [Talk ID: 15]
Ben J Maslen (University of New South Wales)+
10:45: Utilizing capture-recapture methods to generate wildlife counts with AI observers on single capture UAV imagery [Talk ID: 30]
Rebecca Wilks (Edinburgh University)+; Ruth King (University of Edinburgh); Stuart King (Edinburgh University); David Williams (University of Leeds); Murray Collins (Space Intelligence); Niall McCann (National Park Rescue); Mike Chase (Elephants Without Borders)
11:00: From sonar to species: using wideband acoustics and machine learning to classify fish species [Talk ID: 192]
Jessica A Leivesley (University of Toronto)+; Hunter Chen (University of Toronto); Simone Collier (Ontario Ministry of Natural Resources and Forestry); Lewei Er (University of Toronto); Henrique Giacomini (Ontario Ministry of Natural Resources and Forestry); Ryan Grow (Lakehead University); Jeremy Holden (Ontario Ministry of Natural Resources and Forestry); Victoria Kopf (Ontario Ministry of Natural Resources and Forestry); Yihang Luo (University of Toronto); Scott Milne (Milne Technologies); Michael Rennie (Lakehead University); Alex Ross (Lakehead University); Christine Wang (University of Toronto); Alice Zhang (University of Toronto); Miley Zhang (University of Toronto); Dak de Kerckhove (Ontario Ministry of Natural Resources and Forestry); Vianey Leos Barajas (University of Toronto)
11:15: Accounting for AI Classification Errors in Downstream Ecological Analysis [Talk ID: 70]
Aimée Freiberg (University of Fribourg)+; Madleina Caduff (University of Fribourg); Daniel Wegmann (University of Fribourg)
11:45: Avoiding confusion: modelling image identification with classification errors [Talk ID: 230]
Thomas Bartos (Centre for Environment, Fisheries and Aquaculture Science (Cefas))+; Michael A Spence (Cefas); Jon Barry (Centre for Environment, Fisheries and Aquaculture Science (Cefas)); Robert Blackwell (Alan Turing Institute); James Scott (Centre for Environment, Fisheries and Aquaculture Science (Cefas)); Sophie Pitois (Centre for Environment, Fisheries and Aquaculture Science (Cefas))
10:30: Clustering home ranges using Bhattacharyya distance [Talk ID: 252]
Miranda Tilberg (Travelers Insurance); Philip M Dixon (Iowa State University)+
10:45: Sample-size estimation for approximating the habitat availability for Resource-Selection function [Talk ID: 247]
Nilanjan Chatterjee (senckenberg biodiversity and climate research centre)+; Christen Fleming (University of Central Florida); Jesse Alston (University of Arizona); Justin Calabrese (CASUS); John Fieberg (University of Minnesota)
11:00: The interplay between scale and positioning error upon step selection function accuracy [Talk ID: 212]
Rachel Mawer (Ghent University)+; James Campbell (Leibniz-Institute of Freshwater Ecology and Inland Fisheries); Jelger Elings (Ghent University); Ine Pauwels (Research Institute for Nature and Forest (INBO)); Peter Goethals (Ghent University); Stijn Bruneel (Research Institute for Nature and Forest (INBO))
11:15: Step selection analysis with non-linear and random effects in mgcv [Talk ID: 232]
Natasha J Klappstein (Dalhousie University)+
11:45: Scaling up from movement decisions to broad-scale space use patterns: an approach via step selection [Talk ID: 79]
Jonathan Potts (University of Sheffield)+
12:00: Multiscale models of animal movement with irreversible dynamics [Talk ID: 51]
Théo Michelot (Dalhousie University)+
12:15: Come together: Joining species distribution and connectivity models for improved mapping of functional habitat for Norwegian wood beetles. [Talk ID: 235]
Ron R. Togunov (Norwegian University of Science and Technology)+; Sam Perrin (Norwegian University of Science and Technology ); Philip S Mostert (Norwegian University of Science and Technology); Bram Van Moorter (The Norwegian Institute for Nature Research); Manuela Panzacchi (The Norwegian Institute for Nature Research); Anders Finstad (Norwegian University of Science and Technology ); Rannveig Margrete Jacobsen (The Norwegian Institute for Nature Research ); Joseph Chipperfield (The Norwegian Institute for Nature Research ); Kwaku Peprah Adjei (Norwegian Institute for Nature Research); Robert B O’Hara (NTNU)
10:30: Closed-form likelihood functions for spatial capture-recapture [Talk ID: 164]
Jing Liu (University of Auckland)+; Rachel Fewster (University of Auckland); Ben Stevenson (University of Auckland)
10:45: Extending spatial capture-recapture with the Hawkes process [Talk ID: 83]
Alec BM van Helsdingen (University of Auckland)+
11:00: Spatial Capture-Recapture for detectors moving in continuous space and time [Talk ID: 207]
Savannah A Rogers (University of St Andrews)+
11:15: Framework for integrating telemetry data in spatial capture recapture [Talk ID: 242]
Abinand Reddy Kodi (CREEM, University of St Andrews)+
11:45: Are acoustic spatial capture-recapture models also robust to misspecifed detection functions? [Talk ID: 13]
David K. E. Chan (The University of New South Wales)+; Janice Seo (The University of Auckland); Ben Stevenson (The University of Auckland)
12:00: Dealing with the badness of goodness-of-fit [Talk ID: 44]
Rishika Chopara (University of Auckland)+; Ben Stevenson (University of Auckland); Rachel Fewster (University of Auckland)
12:15: A Monte Carlo resampling framework for implementing goodness-of-fit tests in spatial capture-recapture models [Talk ID: 122]
Yan Ru Choo (University of St Andrews)+; Alison Johnston (University of St Andrews); Chris Sutherland (University of St Andrews)
10:30: clustglm and clustord: R packages for clustering with covariates for binary, count, and ordinal data [Talk ID: 140]
Louise McMillan (Victoria University of Wellington)+; Daniel Fernández (Universitat Politècnica de Catalunya–BarcelonaTech); Shirley Pledger (Victoria University of Wellington); Richard Arnold (Victoria University of Wellington); Ivy Liu (Victoria University of Wellington); Murray Efford (Otago University)
10:45: Operationalising Ensemble Modelling for Ecology [Talk ID: 237]
Michael J Thomson (Centre for Environment, Fisheries and Aquaculture Science)+; Michael A Spence (Cefas)
11:00: momentuHMM 2.0 – recent software advances for the analysis of biotelemetry data using hidden Markov models of animal movement [Talk ID: 135]
Brett T McClintock (NOAA-NMFS AFSC Marine Mammal Laboratory)+
11:15: One model, many methods: NIMBLE with automatic differentiation, Hamiltonian Monte Carlo, Laplace approximation, quadrature, posterior approximations, and your latest method [Talk ID: 151]
Perry de Valpine (UC Berkeley)+; Daniel Turek (Lafayette Collete); Wei Zhang (University of Glasgow); Paul van Dam-Bates (Fisheries and Oceans Canada); Benjamin R Goldstein (North Carolina State University); Christopher Paciorek (UC Berkeley)
11:45: Shape-constrained smooths as a bridge between theoretical and statistical ecology [Talk ID: 255]
Benjamin M Bolker (McMaster University)+
12:00: The Fisheries Integrated Modeling System: A Case Study in Open Source Statistical Software Development [Talk ID: 265]
Andrea Havron (NOAA Fisheries Office of Science & Technology)+; Christine Stawitz (NOAA Fisheries Office of Science & Technology); Matthew Supernaw (NOAA Fisheries Office of Science & Technology); Bai Li (ECS Federal in support of NOAA Fisheries Office of Science & Technology); Kathryn Doering (NOAA Fisheries Office of Science & Technology); Patrick Lynch (NOAA Fisheries Office of Science & Technology); Richard Methot (NOAA Fisheries Office of the Assistant Administrator); FIMS Implementation Team (NOAA Fisheries)
12:15: Smoothers and phylogenetically structured random effects in glmmTMB via a proportional random effect structure [Talk ID: 157]
Maeve McGillycuddy (UNSW Sydney)+; Gordana Popovic (UNSW Sydney); David Warton (UNSW Sydney)
14:30: Quantifying the rewidling: comparative movement analysis of wild and translocated predators [Talk ID: 142]
Mohammad MF Farhadinia (DICE, University of Kent)+; Luciano Atzeni (Oxforrd); Jose Hernandez (Severtzov); Anna Yachmennikova (Severtzov); Viatcheslav Rozhnov (Severtzov); Maria Chistopolova (Severtzov); Minaev Alexander (Severtzov); Natalie Dronova (Severtsov Institute of Ecology and Evolution); Alim Pkhitikov (Institute of Ecology of Mountain Territories of the Russian Academy of Sciences); Paul Johnson (WildCRU, University of Oxford); David Macdonald (WildCRU, University of Oxford)
14:45: Integrating biologging-derived data on denning status in a multi-event capture-recapture model for Svalbard polar bears [Talk ID: 245]
Marwan Naciri (CEFE)+; Jon Aars (Norwegian Polar Institute); Sarah Cubaynes (EPHE)
15:00: Particle filters for animal movement modelling in autonomous receiver networks [Talk ID: 91]
Edward Lavender (Eawag)+; Andreas Scheidegger (Eawag); Carlo Albert (Eawag); Stanisław Biber (University of Bristol); Janine Illian (University of Glasgow); James Thorburn (Edinburgh Napier University); Helen Moor (Eawag - Swiss Federal Institute of Aquatic Science and Technology)
15:15: Analyzing whale calls through novel multivariate Hawkes processes [Talk ID: 153]
Bokgyeong Kang (Duke University); Erin Schliep (North Carolina State University); Alan Gelfand (Duke University); Tina Yack (Duke University); Christopher Clark (Cornell University); Robert S Schick (Southall Environmental Associates, Inc)+
15:45: Acoustic monitoring and abundance estimation: the challenges of sparsely distributed instruments [Talk ID: 268]
Danielle Harris (University of St Andrews)+; Kerri Seger (Applied Ocean Sciences, LLC); David Mellinger (Oregon State University); Jennifer Miksis-Olds (University of New Hampshire)
16:00: Lessons learnt from estimating acoustic cue production rates for places and times without data [Talk ID: 273]
Tiago Marques (University of St Andrews / CEAUL / DBA / FCUL)+
14:30: Integrating data from different taxonomic resolutions to better estimate alpha diversity. [Talk ID: 96]
Kwaku Peprah Adjei (Norwegian Institute for Nature Research)+; Robert B O’Hara (NTNU); Nick Isaac (UK Centre for Ecology & Hydrology ); Francesca Mancini (UK Centre for Ecology & Hydrology); Claire Carvell (UK Centre for Ecology & Hydrology)
14:45: Evaluating large-scale population reinforcement efforts for houbara bustards [Talk ID: 146]
Kylee D Dunham (Cornell University Lab of Ornithology)+; Stephanie Harris (Bangor University); Orin J Robinson (Cornell Lab of Ornithology); Yves Hingrat (Reneco International Wildlife Consultants); Viviana Ruiz-Gutierrez (Cornell Lab of Ornithology)
15:00: Combining diverse information sources through integrated models and consensus approaches [Talk ID: 149]
Mario Figueira (University of Valencia)+; David Conesa (University of Valencia); Antonio López-Quílez (Universitat de Valencia); Iosu Paradinas (AZTI)
15:15: Integrating telemetry and count data to estimate animal distribution [Talk ID: 100]
Valentin Lauret (CEFE - University of Montpellier)+; Nicolas Courbin (CEFE - CNRS); Aurélien Besnard (CEFE - EPHE )
15:45: Spatio-temporal data integration for species distribution modelling in R-INLA [Talk ID: 107]
Fiona M Seaton (UK Centre for Ecology & Hydrology)+; Susan Jarvis (UK Centre for Ecology & Hydrology); Pete Henrys (UK Centre for Ecology & Hydrology)
16:00: Integrated movement models for individual tracking and species distribution data [Talk ID: 134]
Frances Buderman (Pennsylvania State University)+; Ephraim M Hanks (Penn State University); Viviana Ruiz Gutierrez (Cornell Lab of Ornithology); Michael Schull (Pennsylvania State University); Robert Murphy (Eagle Environmental, Inc.); David Miller (Pennsylvania State University)
16:15: Nonparametric models for individual movement and population dynamics of migratory species [Talk ID: 200]
Ephraim M Hanks (Penn State University)+; Frances Buderman (Pennsylvania State University); Viviana Ruiz Gutierrez (Cornell Lab of Ornithology); Jim Russell (Muhlenberg College)
14:30: How best to ordinate spatio-temporal data? [Talk ID: 25]
Baptiste Alglave (Université Bretagne Sud)+; Benjamin Dufée (Université Bretagne Sud); Said Obakrim (University of Rennes 1); James Thorson (NOAA)
14:45: Bayesian point pattern models of badger distribution and abundance as tools in bovine TB Eradication in Ireland [Talk ID: 5]
Virginia Morera-Pujol (University College Dublin)+; Damien Barrett (Department of Agriculture, Food, and the Marine); Andrew W. Byrne (Department of Agriculture, Food, and the Marine); Guy McGrath (University College Dublin); David J. Quinn (Department of Agriculture, Food, and the Marine); Simone Ciuti (University College Dublin)
15:00: Quantifying human socioeconomic predictors of land cover change and habitat loss for migratory birds [Talk ID: 138]
Simon English (University of British Columbia)+; Peter Arcese (University of British Columbia); Amanda Rodewald (Cornell Lab of Ornithology); Scott Wilson (Environment and Climate Change Canada)
15:15: Low-rank factor models for modelling dynamic spatiotemporal dependency [Talk ID: 103]
Karunarathna K. A. N. K. (UQ Spatial Epidemiology Laboratory, School of Veterinary Science, Faculty of Science, The University of Queensland, Queensland, 4343, Australia)+
15:45: Clustering in an Agent Based Model [Talk ID: 136]
Lena L Payne (University of Kent)+
16:00: From Landscape to Hotspot: A Wildfire Case Study [Talk ID: 106]
Oscar Rodriguez de Rivera Ortega (University of Exeter)+
16:15: Lagging behind and sideways: estimating the spatial and temporal zone of influence of environmental predictors on ecological processes [Talk ID: 236]
Thomas Cornulier (Biomathematics and Statistics Scotland)+; David L Miller (BioSS/UKCEH); Kate R. Searle (UKCEH); Charlotte Regan (UKCEH); Maria Bogdanova (UKCEH)
14:30: Modelling microbiota composition across the host phylogeny [Talk ID: 209]
Guilhem Sommeria-Klein (University of Turku)+; Benoît Pérez-Lamarque (Institut de Biologie de l’ENS (IBENS), ENS-PSL); Santiago Rosas-Plaza (Universidad Nacional Autónoma de México (UNAM)); Leo M Lahti (University of Turku); Hélène Morlon (Institut de Biologie de l’ENS (IBENS), ENS-PSL)
14:45: The R package occumb for Bayesian inference of multispecies site occupancy models for eDNA metabarcoding [Talk ID: 19]
Keiichi Fukaya (Institute for Environmental Studies)+; Yuta Hasebe (Kanagawa Environmental Research Center)
15:00: Phenologically weighted corrections for sampling effort in hierarchical species community modelling: an analysis of five decades of opportunistically sampled data on ground-dwelling invertebrate communities [Talk ID: 37]
Thomas Neyens (Hasselt University & KU Leuven)+; Maxime Fajgenblat (KU Leuven)
15:15: A hierarchical modeling framework for estimating animal activity patterns: the effect of human presence on diel activity [Talk ID: 54]
Fabiola Iannarilli (Max Planck Institute of Animal Behavior)+; Brian Gerber (USGS, Colorado Cooperative Fish and Wildlife Research Unit, Colorado State University); John Erb (Minnesota Department of Natural Resources); Martin Wikelski (Max Planck Institute of Animal Behavior); John Fieberg (University of Minnesota)
15:45: Predicting service provision to agroecosystems with joint species models [Talk ID: 102]
Teresa Morán López (University of Oviedo)+; Roquer-Beni Laura (CREAF, E08193 Bellaterra (Cerdanyola del Vallés) Spain); Jordi Bosch (CREAF, E08193 Bellaterra (Cerdanyola del Vallés) Spain); Peter Hambäck (Department of Ecology, Environment and Plant Sciences, Stockholm University); Alexandra-Maria Klein (Chair of Nature Conservation and Landscape Ecology, University of Freiburg); Marcos Miñarro (Servico Regional de Investigación y Desarrollo Agroalimentario. Villaviciosa); Ulrika Samnegard (Department of Ecology, Environment and Plant Sciences, Stockholm University); Otso Ovaskainen (University of Jyväskylä); Daniel García (University of Oviedo)
16:00: Issues of convergence of latent-factor Joint Species Distribution Models and ways forward [Talk ID: 194]
Frédéric Gosselin (INRAE)+; Ghislain Vieilledent (CIRAD); Clément Vallé (MNHN)
16:15: Sample size considerations for detecting species co-occurrence with multispecies occupancy models [Talk ID: 216]
Amber Cowans (University of St Andrews)+; Albert Bonet Bigata (University of Aberdeen); Chris Sutherland (University of St Andrews)
Plenary 3
- Dr. Matthew Schofield (University of Otago, New Zealand)
- Back to basics: a closer look at some simple statistical models
- Room G043
- 9:00-10:00am
10:30: Predator-prey models and the hares-eat-lynx paradox [Talk ID: 266]
Eduard Campillo-Funollet (Lancaster University)+; James Van Yperen (University of Sussex)
10:45: Investigating the link between life history and population synchrony through a variety of statistical approaches [Talk ID: 193]
Aline M Lee (Norwegian University of Science and Technology)+; Ellen Martin (Vogelwarte); Ragnhild Bjørkås (Norwegian University of Science and Technology); Jonatan Marquez (Institute of Marine Research); Ivar Herfindal (Norwegian University of Science and Technology); Brage Hansen (Norwegian Institute of Nature Research); Marlene Gamelon (Laboratoire de Biométrie et Biologie Évolutive, CNRS); Sondre Aanes (Norwegian Computing Center); Steinar Engen (Norwegian University of Science and Technology); Are Salthaug (Institute of Marine Research); Bernt-Erik Sæther (Norwegian University of Science and Technology)
11:00: Evaluation of prediction skills for population dynamics models [Talk ID: 81]
Toshihide Kitakado (Tokyo University of Marine Science and Technology)+
11:15: Forecast evaluation of fish weight using LMMs, GAMMs, and mechanistic models [Talk ID: 218]
Mollie E Brooks (DTU Aqua)+
11:45: Gaussian process emulation for an individual-based model simulation of offshore wind development impact on breeding seabirds. [Talk ID: 183]
Anastasia Frantsuzova (BioSS)+
12:00: Design and analysis of feeding experiments for multi-species functional responses [Talk ID: 108]
Benjamin Rosenbaum (iDiv)+
12:15: Strings better left afray? - Investigating Accuracy of Ecological Network Inference using Demographic Simulations [Talk ID: 10]
Erik Kusch (University of Oslo Natural History Museum)+
10:30: Individual heterogeneity capture-recapture models: Accounting for survivorship bias [Talk ID: 21]
Blanca Sarzo (Foundation for the Promotion of Health and Biomedical Research of Valencian Region, FISABIO); Ruth King (University of Edinburgh)+; Rachel McCrea (Lancaster University)
10:45: A Capture-Recapture-Recovery Model with Temporary Emigration for Estimating Population Size from Incomplete Registers [Talk ID: 67]
Lucy Y Brown (University of Kent)+; Eleni Matechou (University of Kent); Bruno Santos (Stockholm University); Eleonora Mussino (Stockholm University)
11:00: On the Identifiability of the Trinomial Model for Mark-Recapture-Recovery Studies [Talk ID: 133]
Simon Bonner (University of Western Ontario)+; Wei Zhang (University of Glasgow); Jiaqi Mu (University of Western Ontario)
11:15: Extension of Pradel model accounting for transients opens a new avenue for exploring continental-scale data and to study transience [Talk ID: 270]
Tomáš Telenský (Center for Theoretical Study, Charles University in Prague)+; David Storch (Center for Theoretical Study, Charles University in Prague); Petr Klvana (Bird Ringing Centre, National Museum, Prague); Jiří Reif (Institute for Environmental Studies, Faculty of Science, Charles University, Prague)
11:45: Close-kin mark-recapture: Assessing the assumption of a large, sparsely sampled population [Talk ID: 203]
Brandon D Merriell (Trent University)+; Micheline Manseau (Environment & Climate Change Canada); Paul Wilson (Trent University)
12:00: Application of close-kin mark–recapture models to estimate American black bear population size [Talk ID: 90]
Anthony Seveque (Senckenberg Research Institude)+; Robert Lonsinger (USGS OK Cooperative Fish & Wildlife Research Unit ); Lisette Waits (University of Idaho); Kristin Brzeski (Michigan Technological University); Caitlin Ott-Conn (Michigan Department of Natural Resources); Sarah Mayhews (Michigan Department of Natural Resources); Cody Norton (Michigan Department of Natural Resources); Tyler Petroeljet (Michigan Department of Natural Resources); Anaïs Tallon (Mississippi State University); Dana Morin (Mississippi State University)
12:15: A multispecies capture–recapture model to estimate biodiversity metrics from coordinated monitoring programs [Talk ID: 18]
Neil Gilbert (Michigan State University)+; Graziella V DiRenzo (U. S. Geological Survey); Elise Zipkin (Michigan State University)
10:30: Greater than the sum of its parts: Computationally flexible Bayesian hierarchical modeling [Talk ID: 85]
Devin Johnson (National Oceanic and Atmospheric Administration)+; Mevin Hooten (University of Texas at Austin); Brian Brost (National Oceanic and Atmospheric Administration )
10:45: Integrating presence-only data into spatio-temporal models to predict taxon encounter probability and density in freshwater and marine environments [Talk ID: 53]
Anthony Rafael Charsley (National Institute of Water and Atmospheric (NIWA) Research)+
11:00: The scale-dependency of biotic interactions in the eyes of integrated species distribution models [Talk ID: 189]
Florencia Grattarola (Czech University of Life Sciences Prague)+; Gurutzeta Guillera-Arroita (Pyrenean Institute of Ecology, Spanish National Research Council); José J. Lahoz Monfort (Pyrenean Institute of Ecology, Spanish National Research Council); Petr Keil (Czech University of Life Sciences Prague)
11:15: Using multi-stage integrated models to quantify global change impacts on North American bird communities [Talk ID: 143]
Jeffrey Doser (Michigan State University)+; Sarah Saunders (National Audubon Society); Shannon Reault (National Audubon Society); Brooke Bateman (National Audubon Society); Joanna Grand (National Audubon Society); Elise Zipkin (Michigan State University)
11:45: Understanding causes of spatially varying environmental effects in integrated hierarchical models of North American mammals [Talk ID: 82]
Benjamin R Goldstein (North Carolina State University)+; Alex Jensen (North Carolina Museum of Natural Sciences); Roland Kays (North Carolina Museum of Natural Sciences); Elizabeth Kierepka (North Carolina Museum of Natural Sciences); Michael Cove (North Carolina Museum of Natural Sciences); William McShea (Smithsonian’s National Zoo and Conservation Biology Institute); Brigit Rooney (Smithsonian’s National Zoo and Conservation Biology Institute); Krishna Pacifici (North Carolina State University)
12:00: Combining data sources for predicting seed retention times in birds [Talk ID: 257]
Juan M Morales (UNIVERSITY OF GLASGOW)+; Claudio Bracho Estévanez (Universidad de Cádiz); Juan Pedro González-Varo (Universidad de Cádiz)
12:15: Data integration for spatially explicit abundance modeling of a keystone species: the case of European rabbit [Talk ID: 166]
Javier Fernández-López (Institute for Game and Wildlife Research IREC (CSIC-UCLM))+; Olivier Gimenez (CNRS); Pelayo Acevedo (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); José Antonio Blanco-Aguiar (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); Joaquín Vicente (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); Ana Santamaría (WWF-Spain); Tamara Burgos (WWF-Spain); Sonia Illanas (Institute for Game and Wildlife Research (IREC-CSIC-UCLM)); Davide Carniato (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); Fernando Silvestre (Fundación CBD-Hábitat); Sergio Ovidio (Junta de Comunidades de Castilla-La Mancha); Ángeles Sánchez (Junta de Comunidades de Castilla-La Mancha); Llanos Gabaldón (Junta de Comunidades de Castilla-La Mancha); Ramón Pérez de Ayala (WWF-Spain)
10:30: Causal mediation approaches for quantifying ecological mechanisms [Talk ID: 20]
Hannah Correia (Johns Hopkins University)+; Paul Ferraro (Johns Hopkins University); Laura Dee (University of Colorado Boulder)
10:45: Estimating extinction time from the fossil record using regression inversion [Talk ID: 12]
David Warton (UNSW Sydney)+; Victor Tsang (UNSW Sydney)
11:00: Partially observable DBN to model the dynamics of partially observable metapopulations : advantages and open challenges [Talk ID: 47]
Hanna BACAVE (INRAE)+; Pierre-Olivier CHEPTOU (CNRS); Nathalie PEYRARD (INRAE)
11:15: A new method to select the number of states in Hidden Markov Model, with application to Movement Ecology. [Talk ID: 63]
Marie-Pierre Etienne (Institut Agro)+; Marie Du Roy De Chaumary (Mathematical Research Institute of Rennes IRMAR); Salima El Koleil (Univ. Rennes, Ensai, CNRS, CREST); Matthieu Marbac (Univ. Rennes, Ensai, CNRS, CREST)
11:45: Modelling Fish Movements and Determining Vulnerability to Fishing Effort in Lake Winnipeg Using Bayesian State-space Models [Talk ID: 4]
Saman Muthukumarana (University of Manitoba)+
12:00: Turning-points identification in High-Resolution Animal Movement Data [Talk ID: 118]
Abdulmajeed Alharbi (Sheffield University)+
12:15: A general likelihood-based method for the inferential analysis of agent-space reactant-catalyst-product models. [Talk ID: 173]
Niklas Moser (University of Jyväskylä)+; Dmitri Finkelshtein (Swansea University); Georgy Chargaziya (Swansea University); Sara Hamis (University of Uppsala); Dagim Tadele (University of Oslo); Otso Ovaskainen (University of Jyväskylä)
Plenary 4
- Dr. Elise Zipkin (Michigan State University, USA)
- Integrated Community Models: A framework combining multispecies data sources to estimate biodiversity dynamics
- Room G043
- 9:00-10:00am
Roundtable 2
- What key statistical methods & concepts should all [undergraduate] ecologists learn?
- Will Kay (Cardiff University, UK)
- Room G014
- 1:00-2:00pm
10:30: Multivariate Zero-Inflated Tobit Regression Modeling [Talk ID: 8]
Becky Tang (Middlebury College)+
10:45: Bayesian Identifiability in Ecological Models [Talk ID: 92]
Diana Cole (University of Kent)+; Daniel Bearup (University of Leicester)
11:00: Approximate Gaussian conjugate sampling for efficient MCMC for complicated hierarchical models [Talk ID: 114]
Wei Zhang (University of Glasgow)+; Oliver Stoner (University of Glasgow)
11:15: Hidden Markov models with an unknown number of states and a repulsive prior on the state parameters [Talk ID: 27]
Ioannis Rotous (University of Kent); Alex Diana (University of Essex); Alessio Farcomeni (Tor Vergata University of Rome); Eleni Matechou (University of Kent)+
11:45: Seasonal survival of the gray mouse lemur under changing climate [Talk ID: 56]
Dilsad Dagtekin (Swiss Federal Institute of Aquatic Science and Technology (Eawag))+; Dilsad Dagtekin (Department of Evolutionary Biology and Environmental Studies, University of Zurich); Dominik Behr (Department of Evolutionary Biology and Environmental Studies, University of Zurich); Claudia Fichtel (Behavioral Ecology and Sociobiology Unit, German Primate Center, Leibniz Institute for Primate Research); Peter M. Kappeler (Behavioral Ecology and Sociobiology Unit, German Primate Center, Leibniz Institute for Primate Research; Department of Sociobiology/Anthropology, University of Göttingen); Arpat Ozgul (Department of Evolutionary Biology and Environmental Studies, University of Zurich)
12:00: A product-multinomial likelihood for multievent mark-recapture models [Talk ID: 165]
Murray B Christian (Centre for Statistics in Ecology, Environment and Conservation, University of Cape Town)+
12:15: Incorporating Memory into Spatially-Explicit Capture-Recapture Models [Talk ID: 145]
Clara Panchaud (University of Edinburgh)+; Ruth King (University of Edinburgh); David Borchers (University of St Andrews); Hannah Worthington (University of St Andrews); Ian Durbach (University of St Andrews); Paul Van Dam-Bates (Fisheries and Oceans Canada, Pacific Biological Station)
10:30: Indirect impacts of a highway on movement behavioral states of a threatened tortoise and implications for landscape connectivity [Talk ID: 150]
Seth Harju (Heron Ecological)+; Scott Cambrin (Clark County Desert Conservation Program); Jodi Berg (Alta Science and Engineering)
10:45: Assessing the impact of temporal resolution on animal behavioural inference [Talk ID: 65]
Rebecca Akeresola (University of Edinburgh)+; Victor Elvira (University of Edinburgh); Ruth King (University of Edinburgh); Adam Butler (Biomathematics and Statistics Scotland); Esther L Jones (Biomathematics & Statistics Scotland); Gail Robertson (Biomathematics and Statistics Scotland)
11:00: Selecting the number of states in hidden Markov models: a double penalty likelihood approach. [Talk ID: 158]
Fanny R Dupont (UBC)+; Marie Auger Methe (UBC); Marianne Marcoux (DFO); Nigel Hussey (Windsor University)
11:15: Inference on the state process of periodically inhomogeneous hidden Markov models for animal behavior [Talk ID: 26]
Jan-Ole Koslik (Bielefeld University)+; Carlina C Feldmann (Bielefeld University); Sina Mews (Bielefeld University); Rouven Michels (Bielefeld University); Roland Langrock (Bielefeld University)
11:45: Inferring individual variation in behavioural patterns using mixtures of hidden Markov models [Talk ID: 190]
Carlina C Feldmann (Bielefeld University)+
12:00: Using movement modelling to understand predator-prey interactions at multiple scales [Talk ID: 184]
Katherine Whyte (Biomathematics and Statistics Scotland (BioSS))+; Ana Couto (BioSS); Charlie Cooper (Scottish Government Marine Directorate); James Dunning (Scottish Government Marine Directorate); Christopher Pollock (UK Centre for Ecology and Hydrology (UKCEH)); Thomas Cornulier (Biomathematics and Statistics Scotland); Adam Butler (Biomathematics and Statistics Scotland); Thomas Regnier (Scottish Government Marine Directorate); Kate Searle (UK Centre for Ecology and Hydrology (UKCEH)); Francis Daunt (UK Centre for Ecology and Hydrology (UKCEH)); Esther L Jones (Biomathematics & Statistics Scotland)
12:15: Combining step selection functions and hidden Markov models to understand and predict animal movement [Talk ID: 211]
Rachel Mawer (Ghent University)+; Ine Pauwels (Research Institute for Nature and Forest (INBO)); Jelger Elings (Ghent University); Stijn Bruneel (Research Institute for Nature and Forest (INBO)); Johan Coeck (Research Institute for Nature and Forest (INBO)); Peter Goethals (Ghent University)
10:30: How many young are too many? - modelling under-dispersed count data [Talk ID: 72]
James A. Clarke (British Trust for Ornithology)+; Philipp Boersch-Supan (British Trust for Ornithology); Jeremy Smith (British Trust for Ornithology); Robert Robinson (British Trust for Ornithology)
10:45: Using hierarchical occupancy and abundance models to choose better indicator species. [Talk ID: 105]
Sanet Hugo (University of Venda)+
11:00: Using informative priors to account for identifiability issues in occupancy models with identification errors [Talk ID: 110]
Célian Monchy (CEFE-CNRS)+; Marie-Pierre Etienne (Institut de recherche mathématique de Rennes); Olivier Gimenez (CNRS)
11:15: Integrated model for dependent spatial capture recapture and presence-absence data [Talk ID: 126]
Mehnaz Jahid (University of Victoria)+
11:45: Precision ecology for targeted action [Talk ID: 223]
Eleanor Jackson (University of Reading); James Bullock (UK Centre for Ecology & Hydrology); Emma Gardner (UK Centre for Ecology & Hydrology); Tord Snäll (Swedish University of Agricultural Sciences); Rebecca Spake (University of Reading)+
12:00: Integrating citizen science into surveillance schemes for invasive plant pests [Talk ID: 182]
Bob Douma (Wageningen University)+; Eveline van Woensel (Wageningen University); Stephen Parnell (University of Warwick); Arnold van Vliet (Wageningen University); Wopke van der werf (Wageningen University)
12:15: Teaching Study Design and Analysis - Comparisons between Life Sciences [Talk ID: 121]
William P Kay (Cardiff University)+; Crispin Jordan (University of Edinburgh); Nicola Romano (University of Edinburgh); Kasia Banas (University of Edinburgh); Vanessa Armstrong (Newcastle University); Jenny Terry (University of Sussex)
10:30: Exploring the use of the South African Nest Record Scheme to detect changes in phenology: A case study using four well represented species [Talk ID: 49]
Rebecca Muller (University of Cape Town)+
10:45: Occupancy Modelling for Rare Species Using Large Datasets: A Subsampling Approach [Talk ID: 219]
Johanna de Haan-Ward (University of Western Ontario)+; Simon Bonner (University of Western Ontario); Danielle Ethier (Birds Canada); Douglas Woolford (University of Western Ontario)
11:00: Identifying drivers of population change with the R-learner and participatory science data [Talk ID: 234]
Daniel Fink (Cornell University)+; Courtney L Davis (Cornell Lab of Ornithology); Tom Auer (Cornell Lab of Ornithology); Matt Strimas-Mackey (Cornell Lab of Ornithology); Alison Johnston (University of St Andrews); Cynthia Crowley (Cornell Lab of Ornithology); Wesley M. Hochachka (Cornell Lab of Ornithology); Shawn Ligocki (Cornell Lab of Ornithology)
11:15: Correcting for sampling bias when modelling with citizen science data [Talk ID: 41]
Guillaume Blanchet (Université de Sherbrooke)+
11:45: Using identification guides as informative priors for species misclassification by citizen scientists [Talk ID: 240]
Fergus J Chadwick (University of St Andrews)+; Daniel Haydon (University of Glasgowe); Dirk Husmeier (University of Glasgow); Jason Matthiopoulos (University of Glasgow); Otso Ovaskainen (University of Jyväskylä)
12:00: What Are They Even Looking For? The Role of Taxonomic Scope in Inferring Effort for Occupancy Models [Talk ID: 6]
Sara Stoudt (Bucknell University)+; Laura Melissa Guzman (University of Southern California); Benjamin R Goldstein (North Carolina State University); Jayme Lewthwaite (University of Southern California); Vaughn Shirey (University of Southern California)
12:15: Spatiotemporal ordination of historical species communities using sparse, opportunistically collected data [Talk ID: 267]
Maxime Fajgenblat (KU Leuven & Hasselt University)+; Thomas Neyens (Hasselt University & KU Leuven)
14:30: Rank abundance curves redux [Talk ID: 186]
Gavin L Simpson (Aarhus University)+
14:45: Advancing spatio-temporal modelling in ecology: a framework to validate model predictions [Talk ID: 162]
Julie Vercelloni (Australian Institute of Marine Science)+
15:00: Modelling Patterns of Functional Similarity in Environmental Responses in Communities Using Markov Random Field Smoothers [Talk ID: 34]
Fonya Irvine (Concordia University)+; Eric Pedersen (Concordia University)
15:15: Advancing restoration ecology through new multivariate time-series models [Talk ID: 98]
Audun Rugstad (NTNU)+; Robert B O’Hara (NTNU); Bert van der Veen (Norwegian University of Science and Technology)
15:45: Attributing space use at-sea for guillemot and razorbill in winter using geolocator tracking data [Talk ID: 251]
Esther L Jones (Biomathematics & Statistics Scotland)+; Ana Couto (BioSS); Lila Buckingham (NINA); Maria Bogdanova (UKCEH); Kate Searle (UKCEH); Francis Daunt (UKCEH); Adam Butler (Biomathematics and Statistics Scotland)
16:00: Updating Projections of Ecological Networks by Integrating Co-Extinction Resilience Mechanisms [Talk ID: 9]
Erik Kusch (University of Oslo Natural History Museum)+
14:30: Assessing the vulnerability of Brazilian primate species to yellow fever in the context of global changes [Talk ID: 46]
Maxime Pierron (CNRS-Université de Strasbourg)+; Carlos Ruiz-Miranda (Universidade Estadual do Norte Fluminense); Cédric Sueur (Université de Strasbourg); Valéria Romano (Institut Méditerranéen de Biodiversité et d’Ecologie marine et continentale)
14:45: Mapping the recolonization of Eurasian otters in France: Coping with heterogeneous data using an integrated species distribution model [Talk ID: 74]
Simon Lacombe (CEFE)+; Sébastien Devillard (Laboratoire de biométrie et de biologie évolutive); Cécile Kauffmann (Société Francaise pour l’Etude et la Protection des Mammifères ); Mélanie Aznar (Groupe Mammalogique d’Auvergne); Xavier Birot-Colomb (LPO Auvergne-Rhône-Alpes); Ondine Dupuis (LPO Bourgogne-Franche-Comté); Christine Fournier-Chambrillon (Groupe de Recherche et d’Etude pour la Gestion de l’Environnmement); Pascal Fournier (Groupe de Recherche et d’Etude pour la Gestion de l’Environnmement); Camille Fraissard (LPO Occitanie); Nicolas Fuento (LPO PACA); Tiphaine Heugas (LPO Vendée); Alexandre Martin (LPO Pays de la Loire); Meggane Ramos (Groupe Mammalogique Breton); Antoine Roche (Groupe Mammalogique et Herpétologique du Limousin); Thomas Ruys (Groupe de Recherche et d’Investigation sur la Faune Sauvage); Franck Simonnet (Groupe Mammalogique Breton); Daniel Sirugue (Société d’histoire naturelle d’Autun); Bastien Thomas (Groupe Mammalogique Normand); Angélique Villéger (Sologne Nature Environnement); Olivier Gimenez (CNRS)
15:00: Interpretable machine learning reveals range-wide variation in fire impacts on birds [Talk ID: 196]
Andrew Stillman (Cornell Lab of Ornithology)+; Gavin Jones (USDA Forest Service); Matt Strimas-Mackey (Cornell Lab of Ornithology); Guillermo Duran (Cornell Lab of Ornithology); Caitlin Andrews (USDA Forest Service); Shawn Ligocki (Cornell Lab of Ornithology); Tom Auer (Cornell Lab of Ornithology); Viviana Ruiz-Gutierrez (Cornell Lab of Ornithology); Sarah Sawyer (USDA Forest Service); Daniel Fink (Cornell University)
15:15: Continental-scale integrated species distribution models for pest species [Talk ID: 156]
Matthew W Rees (CSIRO)+; Jens Froese (CSIRO)
15:45: Challenging the predator-prey overlap metrics in spatial density: a simulation approach [Talk ID: 120]
Marine Ballutaud (La Rochelle Université)+; Mathieu Doray (Ifremer Nantes); Maxime Olmos (Ifremer Brest); Matthieu Authier (La Rochelle Université)
16:00: Do we predict right? How past landscape change affects species distribution models [Talk ID: 137]
Yacob Haddou (University of Glasgow)+; Rebecca Mancy (University of Glasgow); Sofie Spatharis (University of Glasgow); Davide Dominoni (University of Glasgow); Jason Matthiopoulos (University of Glasgow)
16:15: Generalized linear model based on latent factors and supervised components [Talk ID: 35]
Julien Gibaud (University of Monpellier)+
14:30: Trends in plant cover derived from longitudinal vegetation-plot data using ordinal zero-augmented beta regression [Talk ID: 195]
Cas Retel (Statistics Netherlands)+; Kathryn M. Irvine (United States Geological Survey); Arco J. van Strien (Statistics Netherlands)
14:45: Density and abundance estimation of unmarked ungulates using camera traps in the Mudumu National Park, Namibia [Talk ID: 84]
LineekelaOmwene Tuhafeni Nauyoma (University Of Namibia)+; Camille H Warbington (University of Alberta); Fernanda C Azevedo (Universidade Federal de Catalão); Frederico G Lemos (Universidade Federal de Catalão,); Fernando Sequeira (Universidade Do Porto,); Ezequiel C Fabiano (University Of Namibia)
15:00: Stopping rule sampling to monitor and protect endangered species [Talk ID: 39]
Ken B Newman (Biomathematics & Statistics Scotland)+; Lara Mitchell (US Fish and Wildlife Service); Leo Polansky (US Fish and Wildlife Service)
15:15: Extended Batch Marking Models: Improving tractability for very large populations [Talk ID: 129]
Laura Cowen (University of Victoria)+; Kehinde Olobatuyi (University of Victoria); Matthew Parker (Simon Fraser University)
15:45: Unraveling Dolly Varden Survival Patterns: A Bayesian Multi-state Approach with Covariate Considerations [Talk ID: 130]
Arjun Banik (University of Victoria)+; Laura Cowen (University of Victoria); Saman Muthukumarana (University of Manitoba)
14:30: Exploring Cutting-Edge Modeling Techniques for Riparian Bird Species Distribution Prediction in Madjerda River, North-Eastern Algeria [Talk ID: 210]
Mohcen Menaa (University of Souk Ahras)+; Abdelkader Djouamaa (University of Souk Ahras); Moundji Touarfia (University of Souk Ahras); Kaouther Guellati (University of Souk Ahras); Mohamed Cherif Maazi (University of Souk Ahras)
14:45: Disentangling biodiversity concepts and measurements [Talk ID: 86]
Hideyasu Shimadzu (Kitasato University)+
15:00: Disentangling the landscape and local-scale effects of farming on moths: a causal pathway approach [Talk ID: 147]
Rochelle Kennedy (SRUC)+; Nick Littlewood (SRUC); Elisa Fuentes-Montemayor (University of Stirling); Kirsty Park (University of Stirling); Sarah Marley (SRUC)
15:15: Inferring the effects of land use change on unstudied species reveals more serious biodiversity loss [Talk ID: 77]
Jenny A Hodgson (University of Liverpool)+; Claudia Gutierrez-Arellano (University of Liverpool)
15:45: Using Integrated Community Models to Estimate Long-Term Species and Community Trends of Butterflies in the Midwestern United States [Talk ID: 260]
Wendy Leuenberger (Michigan State University)+; Jeff Doser (Michigan State University); Mike Belitz (Michigan State Univerisity); Leslie Ries (Georgetown University); Nick Haddad (Michigan State University); Wayne Thogmartin (USGS); Elise Zipkin (Michigan State University)
16:00: Measuring niche overlap with joint species distribution models: shared and idiosyncratic responses of the species to measured and latent predictors [Talk ID: 60]
Otso Ovaskainen (University of Jyväskylä)+; Nerea Abrego (University of Jyväskylä)
Plenary 5
- Prof. John Fieberg (University of Minnesota, USA)
- Playing in the backyard: The role of statistical ecologists in the evolution of step-selection analysis
- Room G043
- 9:00-10:00am
11:00: How high do birds fly, really? [Talk ID: 197]
Philipp H Boersch-Supan (British Trust for Ornithology)+
11:15: A hierarchical model to evaluate energetic trade-offs in migration decision-making, reproductive effort, and subsequent parental care in a long-distance migratory bird [Talk ID: 254]
Alexander Schindler (University of Saskatchewan)+; Anthony Fox (Aarhus University); Christopher K Wikle (University of Missouri); Bart Ballard (Caesar Kleberg Wildlife Research Institute; Texas A&M University-Kingsville); Alyn Walsh (National Parks and Wildlife Service); Seán Kelly (National Parks and Wildlife Service); Larry Griffin (Wildfowl and Wetlands Trust; ECO-LG Limited); Mitch Weegman (University of Saskatchewan)
11:30: Bayesian models for complex animal movement data [Talk ID: 75]
Toryn Schafer (Texas A&M University)+
11:45: Predicting animal movement with a deep learning step selection framework [Talk ID: 22]
Scott W Forrest (Queensland University of Technology)+; Daniel E Pagendam (CSIRO Data61); Chris Drovandi (Queensland University of Technology); Michael Bode (Queensland University of Technology); Jonathan Potts (University of Sheffield); Maryam Goldchin (CSIRO Health and Biosecurity); Andrew Hoskins (CSIRO Environment)
12:15: Movement-based models for abundance data [Talk ID: 170]
Ricardo Carrizo Vergara (Schweizerische Vogelwarte)+; Marc Kéry (Schweizerische Vogelwarte); Trevor Hefley (Kansas State University)
12:30: The “permeability” R package: A maximum-likelihood based tool to quantify the permeability of linear barriers to animal movement [Talk ID: 275]
Nicole Barbour+, Allicia Kelly, Eliezer Gurarie
11:00: Hierarchical Ordination, an efficient (if not necessarily fast) fitting with INLA [Talk ID: 239]
Robert B O’Hara (NTNU)+; Bert van der Veen (Norwegian University of Science and Technology)
11:15: Fast fitting of phylogenetic random effect models [Talk ID: 222]
Bert van der Veen (Norwegian University of Science and Technology)+; Robert B O’Hara (NTNU)
11:30: Joint taxa distribution models: Generalizing JSDMs while better accommodating rare taxa Joint taxa distribution models: Generalizing JSDMs while better accommodating rare taxa [Talk ID: 213]
Niccolo Anceschi (Duke University)+; Federica Stolf (University of Padua); Gleb Tikhonov (University of Helsinki); Otso Ovaskainen (University of Jyväskylä); David Dunson (Duke University)
12:15: Biplots for multi-species presence-only data via multivariate log-Gaussian Cox processes [Talk ID: 93]
Elliot Dovers (UNSW Sydney)+
12:30: Asymmetric species interactions cannot be inferred without accounting for priority effects [Talk ID: 87]
Francisca Powell-Romero (The University of Queensland)+; Konstans Wells (Swansea University); Nicholas Clark (The University of Queensland)
11:00: Evaluation of an amphibian conservation program with dynamic occupancy models [Talk ID: 57]
Helen Moor (Eawag - Swiss Federal Institute of Aquatic Science and Technology)+
11:15: Spatial stream network occupancy models [Talk ID: 17]
Olivier Gimenez (CNRS)+
11:30: Bringing circuit theory into spatial occupancy models to assess landscape connectivity [Talk ID: 59]
Maëlis Kervellec (University of Montpellier)+; Thibaut Couturier (CEFE); Sarah Bauduin (OFB); Delphine Chenesseau (OFB); Pierre Defos du Rau (OFB); Nolwenn Drouet Hoguet (OFB); Christophe Duchamp (OFB); Julien Steinmetz (OFB); Jean-Michel Vandel (OFB); Olivier Gimenez (CNRS)
11:45: Line Transect data in Occupancy Studies: data collection methods and model identifiability. [Talk ID: 116]
Milly Jones (University of Kent)+
12:15: Hidden Markov models for efficient fitting of dynamic occupancy models to large citizen science occurence datasets [Talk ID: 261]
Hannah Worthington (University of St Andrews)+; Emily Dennis (Butterfly Conservation); Byron Morgan (University of Kent); Takis Besbeas (University of Kent)
12:30: Relative contributions of local and neighbourhood colony size and breeding success to the dynamics of a metapopulation of Black-Headed Gulls (Chroicocephalus ridibundus) [Talk ID: 191]
Killian GREGORY (CEFE, Univ Montpellier, CNRS, EPHE, IRD, Montpellier)+; Charlotte FRANCESIAZ (OFB, DRAS, Juvignac); Jean-Yves BARNAGAUD (CEFE, Univ Montpellier, CNRS, EPHE-PSL University, IRD, Montpellier); Jean-Dominique LEBRETON (CEFE, Univ Montpellier, CNRS, EPHE, IRD, Montpellier); Pierre-André CROCHET (CEFE, Univ Montpellier, CNRS, EPHE, IRD, Montpellier); Aurélien BESNARD (CEFE, Univ Montpellier, CNRS, EPHE-PSL University, IRD, Montpellier); Julien PAPAÏX (INRAE, Avignon)
11:00: Wildlife by Numbers: Data Collection on the Safari Path [Talk ID: 263]
Sougata Sadhukhan (Bharati Vidyapeeth (Deemed to be University) Institute of Environment Education and Research)+
11:15: FLYING LOW AND SLOW: ESTIMATING BROWN BEAR (URSUS ARCTOS) DENSITY USING AERIAL DISTANCE SAMPLING IN KATMAI NATIONAL PRESERVE, ALASKA. [Talk ID: 161]
Leslie Skora (University of Massachusetts )+; Tammy Wilson (U.S Geological Survey)
11:30: Predicting detection probabilities for rare and undersampled North American landbirds [Talk ID: 139]
Brandon PM Edwards (Carleton University)+; Marcel Gahbauer (Environment and Climate Change Canada); Alexis Grinde (University of Minnesota, Duluth); David Hope (Environment and Climate Change Canada); Elly Knight (Boreal Avian Modelling Project); Nicole Michel (National Audubon Society); Barry Robinson (Environment and Climate Change Canada); Péter Sólymos (University of Alberta); Joseph Bennett (Carleton University); Adam Smith (Environment and Climate Change Canada)
11:45: Distance sampling with directional animal movement [Talk ID: 228]
Alison Johnston (University of St Andrews)+; Michael Schrimpf (Cornell University); Wesley M. Hochachka (Cornell Lab of Ornithology); Charles Eaton (University of Manchester); Steve Buckland (University of St Andrews); Charlie Wright (Independent)
12:15: Improved abundance trajectories with Bayesian population dynamics models: case study with a Hawaiian honeycreeper [Talk ID: 32]
Richard J Camp (U.S. Geological Survey)+; Len Thomas (University of St Andrews); David Miller (University of St Andrews); Steve Buckland (University of St Andrews); Steve Kendall (U.S. Fish and Wildlife Service)
12:30: Reproducible workflows for large-scale integrated population analyses [Talk ID: 36]
Chloé R Nater (Norwegian Institute for Nature Research)+; James Martin (University of Georgia); Erlend Nilsen (Norwegian Institute for Nature Research)
- Session: P1 & P3
- Time: Monday & Thursday (4:30-6:30pm)
- Room: Great Hall
Authors: Lukmann Haqeem Alen (WWF-Malaysia)+.
Abstract: Accurate measurement is essential for effective management. The bearded pig (Sus barbatus) holds ecological and socio-economic significance in Sarawak, yet its population faced a sharp decline due to African Swine Fever disease between 2020 and 2022. The first step towards species recovery is knowing the current population size. However, acquiring accurate abundance information for species like the bearded pig poses challenges due to the need for identifying unique individuals or additional species data, such as reproduction and survival rates, in commonly used count-based models. To overcome these hurdles, this study explored statistical models such as N-mixtures and Royle-Nichols. These models estimate animal abundance based solely on unmarked individuals. We simulated the true geographic distribution of bearded pig abundance across Sarawak using five site covariates: elevation, forest cover, water occurrence, distance to the road, and distance to human settlements and then sample with various sets using simple random sampling. We employed four statistical models, including three variants of N-mixture models: Poisson, Negative Binomial, and Zero-Inflated Poisson, as well as the Royle-Nichols model. This study demonstrates the reliability of these models in estimating bearded pig abundance, with model estimates closely matching true abundance up to a certain point and revealing a strong spatial structure in predicted abundance distribution. While substantial sampling efforts are required, it seems these models could offer a promising approach for cost-effective surveys and the establishment of a sustainable, long-term bearded pig monitoring programme in Sarawak. Future efforts will concentrate on validating model performance and refining methodologies through experimentation with real-world data..
Authors: Erik Kusch (University of Oslo Natural History Museum)+.
Abstract: Advances in climate science have rendered obsolete gridded observation data sets commonly used in macroecological analyses. Novel climate reanalysis products outperform legacy data products in accuracy, temporal resolution, and provision of uncertainty metrics. Consequently, there is an urgent need to develop a workflow through which to integrate these improved data into analyses. The ERA5 product family are the latest and most advanced global reanalysis products created by the ECMWF. These data products offer up to 83 essential climate variables at hourly intervals for the time-period of 1950 to today. Spatial resolutions range from 30km (ERA5) to 9km (ERA5-Land) and can be statistically downscaled to study-requirements at finer spatial resolutions. Kriging is one such method to interpolate data to finer resolutions and has the advantages that one can leverage additional covariate information and obtain the uncertainty associated with the downscaling.
The KrigR R Package enables users to (1) download ERA5(-Land) climate reanalysis data for a user-specified region, and time-period, (2) aggregate these climate products to desired temporal resolutions and metrics, (3) acquire topographical co-variates, and (4) statistically downscale spatial data to a user-specified resolution using co-variate data via kriging. This R package provides a workflow for implementation of state-of-the-art climate data into biological analyses while avoiding issues of storage limitations at high temporal and spatial resolutions by providing data according to user-needs rather than in global data sets. Consequently, KrigR provides a toolbox to obtain a wide range of tailored climate data at unprecedented combinations of high temporal and spatial resolutions thus enabling the use of world-leading climate data in the R-environment..
Authors: Kathleen Gundermann (Pennsylvania State University)+; Sean Murphy (Pennsylvania Game Commission); Clayton Lutz (Pennsyvlania Game Commission); Lisa Williams (Pennsylvania Game Commission); Frances Buderman (Pennsylvania State University).
Abstract: Understanding distributions and habitat associations of species of greatest conservation need (SGCN) is critical for making appropriate management recommendations. However, some SGCN, such as marsh birds, are difficult to monitor due to their secretive nature. In Pennsylvania, many marsh birds are considered species of conservation need and are primarily monitored through systematic, repeated detection/non-detection surveys. However, additional data sources on marsh birds are available through semi-structured surveys conducted by volunteers using guidelines provided by the PGC and unstructured community science data (e.g., eBird). Our objective was to compare the effect of supplemental data sources on the precision of site-level occupancy estimates using a multi-species, integrated data occupancy model in a Bayesian framework. We first developed this modeling framework and compared the precision of site-level estimates of occupancy using different combinations of data types. In addition, we used spatial splines to account for unexplained spatial autocorrelation of occupancy estimates. This modeling framework allowed us to share information about detection probability from the structured surveys with the semi- and unstructured surveys. Our results indicate that supplemental data, in the form of semi-structured or community science detections, increased precision of occupancy estimates, specifically in areas where structured surveys were not conducted. We found that habitat associations were the largest drivers of probability of occupancy in all models, especially for ground nesting species. Through this work, we have highlighted how the use of supplemental data sources, such as unstructured community science data, can increase the precision of occupancy estimates for low-density and elusive species..
Authors: Scott W Forrest (Queensland University of Technology)+; Daniel E Pagendam (CSIRO Data61); Andrew Hoskins (CSIRO Environment); Chris Drovandi (Queensland University of Technology); Jonathan Potts (University of Sheffield); Justin Perry (North Australian Indigenous Land and Sea Management Alliance); Eric Vanderduys (CSIRO Environment); Michael Bode (Queensland University of Technology).
Abstract: There has been increasing interest in simulating stochastic movement paths from step selection models. Hitherto, models used to simulate trajectories have not included temporally dynamic coefficients on both the movement and external selection processes, despite animals having temporally dynamic behaviour over daily or seasonal timescales. Here, we focused on simulating stochastic trajectories from step selection functions (SSFs) that include temporal dynamics using harmonic terms, focusing on dynamic behaviour on a daily timescale. The models also incorporated home ranging behaviour through a decaying memory process, which was also interacted with the harmonic terms. We simulated trajectories of individual animals and assessed how they compared to observed data through animal-movement-informed summary statistics. We applied our methods to GPS-tracked water buffalo (Bubalus bubalis), which are an invasive species in Northern Australia’s tropical savannas. The simulations generated from the temporally dynamic models reproduced the movement and habitat selection behaviour that we observed in the GPS data, and allowed for more informative interpretation of animal behaviour, particularly relating to the buffalo’s crepuscular movement behaviour and diurnal habitat selection for thermoregulation. When assessed over longer time scales, models both with and without daily temporal dynamics generated simulations that performed similarly according to the summary statistics. We recommend fitting temporally dynamic SSFs when generating simulated trajectories to most closely represent the animal’s dynamic behaviour. We considered daily behavioural dynamics here, although any timescale of interest can be incorporated. Including temporal dynamics in SSFs for the purpose of simulating new data can address ecological and behavioural questions and provide valuable information for conservation management, particularly for species with clear daily or seasonal behaviour patterns..
Authors: Léa Pautrel (TerrOïko)+.
Abstract: With the increasing use of sensors such as camera traps and autonomous recording units for biodiversity monitoring, there is a growing trend towards collecting fauna observations in continuous-time. This questions traditional analysis with discrete-time models, which, when data is collected in continuous-time, require data discretisation. Discretisation is an aggregation of data into arbitrarily chosen non-independent discrete time intervals, producing information loss. To overcome the limitations of discretisation, ecologists are increasingly turning to continuous-time models. Focusing on occupancy models, a type of species distribution models, we asked ourselves: Should we dedicate time and effort to learning and using these continuous-time models, or can we go on using discrete-time models?
We conducted a comparative simulation study, using detection data generated within a continuous-time framework. We aimed to evaluate the ability of different models to retrieve the simulated occupancy probability. We compared five static occupancy models with varying detection processes: discrete detection/non-detection process, discrete count process, continuous Poisson process, and two types of modulated Poisson processes. All models accurately estimated occupancy when we simulated detection of easily detectable animals, however all models struggled with highly elusive ones. Variations in discretisation intervals had minimal impact on the discrete models’ capacity to estimate occupancy accurately. Our findings suggest that opting for models with a complex detection process may not be advantageous over simpler models when the sole aim is to accurately estimate occupancy. However, such models can offer valuable insights into specific species behaviour and broader ecological inquiries..
Authors: Andrew Houldcroft (University of Exeter)+.
Abstract: Shared landscapes in which humans and wildlife coexist, are increasingly recognized as integral to the long-term persistence of many threatened species. However, these landscapes are heterogeneous and expose wildlife to spatially complex threats including hunting, bidirectional zoonosis and infrastructural mortalities. Effective conservation therefore necessitates population density and distribution data at finer spatial resolutions than offered by conventional approaches. Using the novel R package inlabru, we develop a full-likelihood joint log-Gaussian Cox process model to simultaneously perform spatial distance sampling and model a spatially varying cluster size distribution, which we condition upon detection probability to account for cluster-size bias. We accommodate spatial dependencies by incorporating a non-stationary Gaussian Markov random field for the first time, enabling the explicit inclusion of geographical barriers to wildlife dispersal. We demonstrate this model using 136 georeferenced detections of Campbell’s monkey (Cercopithecus campbelli) clusters, collected with 398.56-km of line transects across a shared agroforest landscape mosaic (1067-km2) in Guinea-Bissau. We assess a suite of anthropogenic and environmental spatial covariates, finding that normalized difference vegetation index (NDVI) and proximity to mangroves are both powerful spatial predictors of densities, suggesting that mangroves are providing hunting refuge benefits at the landscape scale. We estimate a population of 10,730 (95% CI [7608-15,330]) individuals and produce a fine-resolution predictive density map, revealing the importance of mangrove-habitat interfaces for the persistence of this heavily hunted primate. This work demonstrates a powerful, widely applicable approach for monitoring wildlife and informing evidence-based conservation in complex, heterogeneous landscapes moving forward..
Authors: Yiran Shao (Western University)+; Danielle Ethier (Birds Canada); Simon Bonner (Western University).
Abstract: The World Conservation Union lists invasive species as the second largest threat to global biodiversity. Understanding the dispersal patterns of invasive species is crucial for prediction and management efforts. Observational studies of some species have concluded that they possess an innate preference to disperse in a certain direction, but this effect is rarely modelled and quantified independently of the effects of movement between connected areas (the neighbourhood effect) and the species’ habitat preference. In this study, we established a hierarchical dynamic occupancy modelling framework to quantify the effect of factors including habitat, climate, environmental conditions, presence in neighbouring areas and directionality preference on the dispersion patterns of an invasive species. We obtained data on the spread of the Eurasian Collared-Dove (Streptopelia decaocto, EUCO) in North America from the citizen-science study, Project FeederWatch, and estimated the effect of each factor on the three dynamic components: initial colonization, persistence, and recolonization. In particular, our model considers the strength of directional influence separately from the neighbourhood effect. Our results indicate that the previously suggested directional preference of EUCO can largely be explained by the effects of habitat, climate, and environmental conditions. More specifically, EUCO tends to colonize warm and wet grassland habitats and persist in urban areas. Our model enriches the toolkit for ecologists with a way to independently assess various effects in modelling the spread of invasive species, and demonstrates that historical data on invasive species can identify the drivers of invasions, aiding in predictions vital for resource management..
Authors: Thomas Cheale (University of Kent)+; Rachel McCrea (Lancaster University); Eduard Campillo-Funollet (Lancaster University); David Roberts (University of Kent).
Abstract: In conservation contexts, understanding human behaviour and its underlying drivers is paramount. Surveys serve as crucial instruments in this endeavour, offering valuable insights and, when no biases are present, yield reliable population-level estimates. However, when asking sensitive questions, parameters of interest may be appreciably underestimated due to social desirability biases or fear of consequences if answers are leaked. To address this challenge, Randomized Response Techniques (RRTs) (for example Warner’s model), ensure individual privacy by adding an element of randomness into the survey designs. However, the more privacy that a survey offers, the higher the variance, and hence required sample size.
Consequently, survey design requires careful consideration: inadequate privacy risks bias, while excessive privacy escalates sample size requirements for inferential power. This poster presents a novel framework to describe multiple randomized response survey methods. The framework allows for tailored survey design based upon both privacy and variance considerations.
The poster will outline how the framework works, what parameters can be chosen, and the resultant survey types. The framework utilizes a maximum likelihood approach and shows that no matter the choice of parameter, an MLE always exists, though their practical utility may vary. Moreover, the poster offers an explicit estimation of the covariance matrix based on Fisher’s information, allowing for informed decisions in sample size determination..
Authors: Alba Fuster Alonso (Institute of Marine Sciences (ICM) - CSIC)+; Haavard Rue (King Abdullah University of Science and Technology (KAUST)); Finn Lindgren (University of Edinburgh (UoE)); Elias Krainski (King Abdullah University of Science and Technology (KAUST)).
Abstract: Climate change is causing alterations in marine ecosystems on a global scale. Therefore, accurately estimating and predicting both present and future spatial and temporal distributions of marine species and key drivers has become crucial for effective management and conservation efforts. For this reason, we present a tutorial for the implementation of the barrier model proposed by Bakka et al., 2019 and model fitting with INLA (Rue et al., 2009). This model could be useful in the context of Species Distribution Models (SDMs), especially on a global scale as shown in this tutorial.
The tutorial’s framework addresses the challenge of spatial domain modeling by considering physical barriers and the Earth’s spherical geometry (https://github.com/hrue/r-inla/tree/devel/rinla/vignettes/barrier-global). Given our focus on global oceans, continents serve as significant physical barriers, thereby close to the barrier correlation tends to zero. Furthermore, we consider the Earth’s spherical shape, as using a planar approximation can introduce biases, particularly when analyzing phenomena on a global scale.
The tutorial illustrates a simulation and Bayesian modeling framework where we simulate two different spatial and temporal scenarios from different likelihoods: Gaussian and Bernoulli. Then, we sample from both scenarios to fit spatial and spatio temporal models with the selected observations and verify if we retrieve the parameters. Results show that INLA and inlabru can be powerful tools for analyzing changes in the distribution of marine species on a global scale.
Bakka, H., Vanhatalo, J., Illian, J. B., Simpson, D., and Rue, H. (2019). Non-stationary Gaussian models with physical barriers. Spatial statistics, 29, 268-288.
Rue, H., Martino, S., and Chopin, N. (2009). Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. Journal of the Royal Statistical Society Series B: Statistical Methodology, 71(2), 319 392..
Authors: Gleb Tikhonov (University of Helsinki)+; Anis Rahman (University of Jyväskylä); Jari Oksanen (University of Helsinki); Tuomas Rossi (CSC – IT Center for Science); Otso Ovaskainen (University of Jyväskylä).
Abstract: Joint Species Distribution Modelling (JSDM) is a powerful and increasingly widely used statistical methodology in biodiversity modelling, enabling researchers to assess and predict the joint distribution of species across space and time. However, JSDM can be computationally intensive and even prohibitive, especially for large datasets and sophisticated model structures. Several recent works specifically targeted JSDM scalability through leveraging machine learning techniques designed for big data analysis and demonstrated capability to scale-up for largest existing species community datasets. Yet, so far these approaches have been adapted for rather limited simple model structures only. On the other hand, the much-desired opportunity to disentangle more intricate ecological patterns from increasingly larger datasets additionally calls for adequately flexible JSDM structures that can reflect the characteristics of data collection design and modelling assumptions. To address computational limitations of flexible JSDM, we expanded one widely used JSDM framework, Hmsc-R, by developing a Graphical Processing Unit (GPU) -compatible implementation of its model fitting algorithm. While our augmented framework retains the original user interface in R, its new computational core is coded in Python and dominantly uses TensorFlow library. This enhancement primarily targets to enable leveraging high-performance computing resources effectively, though it also accelerates model fitting with consumer-level machines. This upgrade is designed to leverage high-performance computing resources more effectively. We evaluated the performance of the proposed implementation across diverse model configurations and dataset sizes. Our results indicate significant model fitting speed-up compared to the existing Hmsc-R package across most models. Notably, for the largest datasets, we achieved >1000 times speed-ups. This GPU-compatible enhancement boosts the scalability of Hmsc-R package by several orders of magnitude, reaching a significantly higher level. It opens promising opportunities for modelling extensive and intricate datasets, enabling better-informed conservation strategies, environmental management, and climate change adaptation planning. .
Authors: Thomas F Bayliss White (University of Exeter)+; Ashok Krishnamurthy (Mount Royal University).
Abstract: Introduction and Objective(s): Mathematical modelling of infectious diseases is an interdisciplinary area of increasing interest. We present a spatial variant of the common SVEIRD (Susceptible-Vaccinated-Exposed-Infectious-Recovered-Dead) model of epidemiology to capture the transmission dynamics of the spread of Ebola in the 2018-2020 outbreak in the Democratic Republic of the Congo (DRC). The challenge is that population mobility is low in this area of Africa, and distances are long, so spatial epidemics tend to burn out, or they expand only slowly. As such, predicting the transmission dynamics of Ebola is challenging and comes with a lot of uncertainty. The goal of this research is to minimise this uncertainty using Bayesian data assimilation methods and provide insight that would support public health officials towards informed, data-driven decision making. Method(s) and Results: The ensemble optimal statistical interpolation data assimilation method has been shown to produce optimal Bayesian statistical tracking of emerging epidemics[1]. Our simulations show good correspondences between the model and the available sparse empirical data. A comparison between the weekly incidence data set and our compartmental model coupled with Bayesian data assimilation highlights the role of a realisation conditioned on all prior data and newly arrived data. In general, the compartmental model with data assimilation gives a better fit than the model without data assimilation for the same time period. We present spatio-temporal disease maps for the infectious variable for the progress of Ebola in the North-Kivu and Ituri provinces of DRC during 2018-2020. This case study was conducted using real-world data from the WHO and practical simulation exercises using free and open-source software. References: [1] L. Cobb, A. Krishnamurthy, J. Mandel, and J. Beezley, Bayesian Tracking of Emerging Epidemics using Data Assimilation Methods, Spatial and Spatio-Temporal Epidemiology, vol. 10, 2014, 39-48..
Authors: Ilaria Pia (University of Helsinki)+; Jarno Vanhatalo (University of Helsinki); Lari Veneranta (Luke).
Abstract: Population growth models are essential tools for natural resources management and conservation. However, we still do not properly understand how the processes underlying reproduction of natural animal populations are affected by environment at the spatial scale at which reproduction actually happens. A particular challenge is that observations from different life cycle stages are often collected at different scales and there is a lack of statistical methods to link local and spatially aggregated information. We address this challenge by developing spatially explicit population growth models for annually reproducing fish. Our approach integrates mechanistic Ricker and Beverton-Holt population growth models with a zero-inflated species distribution model and utilizes the hierarchical Bayesian approach to estimate the model parameters from data with varying spatial support: local scale count data on offspring and environment, and areal data from commercial fisheries informing about a spawning stock size. We used the model to analyze the drivers of distribution of whitefish (Coregonus laveratus (L.) s.l.) reproduction areas, spawner density and maximum proliferation rate along the Finnish coast of the Gulf of Bothnia in the Baltic Sea, both in current climatic conditions and in different predictive scenarios corresponding to future climate conditions. To identify the environmental factors that mainly impact the processes underlying whitefish reproduction, we gave covariate effects shrinkage priors that shrink unnecessary covariate effects towards zero. We also define informative priors for the maximum proliferation rate, and the effect of fish weight on it, based on prior information on density-independent survival and fish egg production. Our model showed good predictive performance as measured by the Bayesian posterior predictive checks and they outperformed a traditional zero-inflated, generalized linear model based, species distribution model in their predictive skill..
Authors: Matthew Tarnowski (Swansea University)+; Eva Sonnenschein (Swansea University); Andy Stawowy (Swansea University).
Abstract: Globally, produced and discarded plastics now exceed the combined weight of all terrestrial and marine animals. In the environment, plastics abiotically fragment into smaller pieces, so-called microplastics, polluting soils, oceans and the atmosphere. The overall lifecycle and ultimate fate of microplastic in Nature is poorly understood, however, the recent discovery of microbial plastic-degrading enzymes indicate that microbes have the potential to bioremediate plastic. To identify novel plastic-degrading enzymes and microorganisms from various habitats, we are employing microplate-based high-throughput assays and amplicon sequencing to enrich and monitor microplastic-associated microbial communities growing on plastic monomers. High-throughput microbial enrichment and characterisation can provide large datasets which reveal how microbial communities adapt to novel substrates. Following the enrichment, the most active microbial community will then be analysed using metagenomics and metatranscriptomics to elucidate its plastic-degrading capabilities. Any enzyme candidates that are identified will be assessed for microplastic bioremediation in wastewater within a novel reactor technology developed by the BMRex consortium (www.bmrex-project.eu/) and may offer solutions to plastic pollution. .
Authors: Megan R. Morton (University of Glasgow)+.
Abstract: Animal movement studies vary widely in spatiotemporal resolutions, model types, and data collection methods, resulting in numerous windows of insight into multiscale habitat preferences and movement processes. For example, telemetry tagging provides insight into local habitat preferences of individuals, whereas survey data can uncover global-scale preferences of the population as a whole. Different data types and modelling methodologies have their own pros and cons, as well as varying sources of bias and perspectives on the underlying processes of interest. Using data integration and joint modelling, we can strengthen estimation of selection parameters, improving overall understanding of animal habitat preferences.
This poster presents an ongoing project in which a joint model of two animal movement modelling perspectives (resource selection functions and step selection functions) is implemented using the R package inlabru. Survey and telemetry data are jointly modelled, combining point process methodology with Langevin diffusion-based movement modelling. The method makes use of the Integrated Nested Laplace Approximation (INLA) methodology for computationally efficient spatiotemporal modelling and The Stochastic Partial Differential Equation (SPDE) approach to incorporate a Gaussian Random Field (GRF) into the model. By accounting for spatial autocorrelation, the risk of spurious significance in estimation of selection parameters is reduced. The joint modelling framework is implemented in inlabru with user-friendly and concise code..
Authors: Moritz Klaassen (MARE-Madeira)+; Tiago Marques (Centre for Research into Environmental and Ecological Modelling, University of St Andrews); Filipe Alves (MARE-Madeira); Marc Fernandez (MARE-Madeira/ARDITI).
Abstract: Correlative ecological niche models (ENMs) and species distribution models (SDMs) are quantitative tools in biogeography and macroecology. Building upon the ecological niche concept, they correlate environmental covariates to species presence to model habitat suitability and predict species distributions. Since their development, ENMs and SDMs have undergone substantial advances in their predictive accuracy, benefiting from increased data availability, advanced machine learning algorithms, novel data integration procedures, refined model validation techniques and incorporation of species interactions. Although initially applied in terrestrial systems, these models are now also being widely used in the marine environment, recognized for their value in marine conservation planning, fisheries management, and understanding species responses to climate change. Despite their increased application, ENMs and SDMs face unique challenges when applied in the marine environment. These include the three-dimensional complexity of marine ecosystems, availability of environmental covariates across suitable spatiotemporal scales, the dynamic properties of these covariates, and unique dispersal patterns and mobility traits of marine species. Here, we review recent methodological advances and emerging trends in marine ENMs and SDMs. We highlight three-dimensional modelling approaches that capture species distributions below the sea surface and assess the importance of temporal resolution, particularly for modelling highly mobile marine species. Further, we discuss the expansion in the types of occurrence data being used, such as citizen science contributions and satellite tracking data. We also explore novel methodologies for environmental data collection, including remote-sensing technologies and ocean models. Together, our review synthesizes methodological innovations, highlights ongoing challenges and discusses emerging trends within the extensive literature of marine ENMs and SDMs..
Authors: Eleonora Gatti (MPI of Animal Behavior)+.
Abstract: Animal movement is essential for survival, as it allows animals to find food, mates, and avoid predators. Efficient movement is particularly important when resources are scattered and animals must choose their paths wisely. This is the case with soaring birds, where individuals are highly dependent on environmental energy (updraughts) as a resource to reduce movement costs. Here, we use soaring flight as a model system and present an Agent-Based Model to investigate the influence of social information on the movement performances of animals within dynamic environments. Here we assess how changes in the perceptual radius - the spatial range over which an individual can acquire social cues - may influence movement decisions and thus the overall cost of movement in a dynamic environment of otherwise invisible updraughts. Our results reveal that an increase in the perceptual radius enhances movement efficiency up to a certain threshold, beyond which no further improvements are observed. This suggests the existence of an optimal perception radius for maximizing movement efficiency in dynamic environments, highlighting the critical role of social information in animal movement strategies and a focus for future work regarding the trade-off between social cue acquisition and competition for the resources they reveal..
Authors: Philip T Patton (Hawaiʻi Institute of Marine Biology)+; Krishna Pacifici (North Carolina State University); Robin Baird (Cascadia Research Collective); Erin Oleson (NOAA Pacific Islands Fisheries Science Center); Jason Allen (Chicago Zoological Society’s Sarasota Dolphin Research Program, c/o Mote Marine Laboratory); Erin Ashe (Oceans Initiative); Alinea Athayde (Projeto Baleia à Vista (ProBaV)); Charla Basran (Research Center in Húsavík); Elsa Cabrera (Centro de Conservación Cetacea (CCC)); John Calambokidis (Cascadia Research Collective); Júlio Cardoso (Projeto Baleia à Vista (ProBaV)); Emma Carroll (University of Auckland - Waipapa Taumata Rau); Amina Cesario (Tethys Research Institute); Barbara J Cheney (University of Aberdeen); Ted Cheeseman (Happywhale.com); Enrico Corsi (Cascadia Research Collective); Jens Currie (Pacific Whale Foundation); John Durban (Oregon State University); Erin Falcone (Marine Ecology and Telemetry Research); Holly Fearnbach (SR3, SeaLife Response, Rehabilitation and Research); Kiirsten Flynn (Cascadia Research Collective); Wally Franklin (The Oceania Project); Trish Franklin (The Oceania Project); Bárbara Galletti Vernazzani (Centro de Conservación Cetacea (CCC)); Tilen Genov (Morigenos - Slovenian Marine Mammal Society); Marianne Rasmussen (Research Center in Húsavík); Marie Hill (NOAA Pacific Islands Fisheries Science Center); David Johnston (University of Otago); Erin Keene (Marine Ecology and Telemetry Research); Claire Lacey (Hawaiʻi Institute of Marine Biology); Sabre Mahaffy (Cascadia Research Collective); Tamara McGuire (The Cook Inlet Beluga Whale Photo-ID Project); Liah McPherson (Hawaiʻi Institute of Marine Biology); Catherine Meyer (University of Auckland); Robert Michaud (Groupe de recherche et d’education sur les mammiferes marins); Anastatsia Miliou (Archipelagos Institute of Marine Conservation); Dara Orbach (Texas A&M University - Corpus Christi); Grace Olson (Pacific Whale Foundation); Heidi Pearson (University of Alaska Southeast); William Rayment (University of Otago); Caroline Rinaldi (L’association Evasion Tropicale); Salvatore Siciliano (Escola Nacional de Saúde Pública/Fiocruz); Stephanie Stack (Pacific Whale Foundation); Beatriz Tintore (Archipelagos Institute of Marine Conservation); Leigh Torres (Oregon State University); Jared Towers (Bay Cetology); Reny Tyson Moore (Chicago Zoological Society’s Sarasota Dolphin Research Program, c/o Mote Marine Laboratory); Caroline Weir (Falklands Conservation); Bec Wellard (Curtin University); Randall Wells (Chicago Zoological Society’s Sarasota Dolphin Research Program, c/o Mote Marine Laboratory); Kym Yano (NOAA Pacific Islands Fisheries Science Center); Jochen Zaeschmar (Far Out Ocean Research Collective); Lars Bejder (Hawaiʻi Institute of Marine Biology).
Abstract: Marine mammal stocks assessments, particularly for cetaceans, often include an abundance estimate from photographic identification (photo-id) surveys, which involve processing considerable photographic data. To ease this burden, agencies increasingly make use of automated matching algorithms. These algorithms confront agencies with an opportunity, reducing the cost of population assessments, and a challenge, propagating photo-id errors into abundance estimators at a large scale. In this study, we explore the tradeoffs between labor costs and estimator performance in population assessments relying on automated photo-id. We developed a general optimization tool for finding the optimal action amongst a set, which included true automation—where the model generated capture-histories without user input—to five degrees of partial automation—where users selected the correct match from 5, 10, …, 25 suggested matches. We developed the tool via simulation, generating 100 Jolly-Seber datasets that we corrupted with the real-life error rates from a multispecies matching algorithm evaluated on 39 catalogs representing 24 cetacean species. This allowed us to evaluate estimator performance and cost across a wide range of species, actions, and catalog sizes. We found that true automation was optimal for catalogs where the algorithm matched images well. As matching performance declined, the tool tended to recommend that users evaluate more suggested matches from the algorithm, particularly for smaller catalogs. We found that false negative errors strongly predicted estimator performance, with a 2% increase in the false negative rate translating to a 5% increase in the relative bias in the superpopulation size. Agencies can use our tool to estimate expected performance of the abundance estimator, project labor effort, and find the optimal degree of automation for their catalog and algorithm. At any rate, we recommend that agencies evaluate their algorithm’s performance, particularly by estimating the false negative rate, before deploying it in a stock assessment..
Authors: Marc Fernandez (MARE-Madeira/ARDITI)+; Mieke Weyn (MARE/Madeira/ARDITI); Filipe Alves (MARE-Madeira/ARDITI).
Abstract: When studying highly mobile marine species, like cetaceans, several challenges arise due to their oceanic range and elusive behaviour. This makes traditional data collection methods, such as visual surveys, labour-intensive and limited in scope. Understanding their habitat preferences and use is essential to promote adequate conservation and management measures. One of the main challenges remaining is to better characterize residency patterns, which is usually done using limited photo-identification datasets. This study presents a novel approach to evaluating short-finned pilot whales’ residency patterns by leveraging satellite telemetry data in the Macaronesia region. We deployed 14 SPOT ARGOS satellite tags on short-finned pilot whales, resulting in a comprehensive 5-year dataset with 7240 locations fixed and covering an area of 1,130,467 km2. A continuous-time movement model was employed to convert the data into a regularly spaced format to facilitate robust statistical analysis. Subsequently, we implemented an integrated step selection analysis to identify core use areas, investigate seasonal variations in residency patterns, and elucidate potential drivers of space use, including environmental variables. Our results show that most of the whales tagged strongly preferred a relatively small area southeast of Madeira Island (around 400 km2). Animals inside the residency area showed shorter step lengths but almost no difference in the turning angles. By shedding light on the residency dynamics of pilot whales, this study contributes valuable knowledge for conserving these socially and ecologically important marine mammals..
Authors: Martin Sköld (Swedish Museum of Natural Sciences)+.
Abstract: Managing and conserving species that traverse administrative borders is challenging. This work considers the problem of assigning individuals to their management region based on spatially and genetically referenced cues (e.g. scat) from a partial area search. I borrow models from the Spatial Capture-Recapture literature to derive explicit thresholds for when to assign an individual in the surveyed or neighbouring region. In a case study, the method is illustrated using data from the Swedish brown bear monitoring programme, where in particular I discuss consequences of ignoring individual variation in home-range size and detectability..
Authors: Jorge Mestre Tomás (Institute of Marine Sciences (ICM) - CSIC)+; Alba Fuster Alonso (Institute of Marine Sciences (ICM) - CSIC); Marta Coll (Institute of Marine Sciences (ICM) - CSIC).
Abstract: Species Distribution Models (SDMs) are widely used in ecology and conservation to predict current, past, and future geographic distribution and suitable habitats of species. SDMs serve as a useful tool for understanding ecological relationships and making informed decisions in conservation and resource management. Several studies have highlighted the worldwide impact of climate change on marine ecosystems, revealing how marine species are adapting and potentially changing their distributions. Therefore, predicting the spatial-temporal distribution of species is essential in ecological and conservation research. Bayesian Additive Regression Trees (BART), a nonparametric machine learning approach based on a sum-of-trees model, has emerged as a promising method for SDMs. By incorporating georeferenced records of global scale marine species and environmental data, BART can assess how species respond to environmental conditions and predict their possible past and future habitats based on environmental changes. In addition, these models could provide a valuable approach to inform and empower the inputs of global Marine Ecosystem Models (MEMs). To make this tool and this framework more accessible to users, we present GLOSSA (GLobal Ocean Species Spatiotemporal Analysis), a user-friendly Shiny application in R designed to apply BART to global scale marine data, allowing users to easily predict native ranges, suitable habitats, and response curves for environmental variables of marine species. With GLOSSA, researchers gain access to a powerful and easy-to-use tool for projecting the past, present and future geographical distribution of marine species..
Authors: Susan Jarvis (UK Centre for Ecology & Hydrology)+; Fiona M Seaton (UK Centre for Ecology & Hydrology); Joanna Staley (UK Centre for Ecology & Hydrology).
Abstract: Data collected by citizen scientists provides a large source of potentially valuable information on species distributions, abundance and responses to environmental pressures. Citizen science data is also hugely variable: it is collected in different ways, with different levels of accuracy and for differing motivations. Extracting true ecological signals from citizen science data can therefore be challenging. One suggested approach is to combine citizen science data with professionally collected data where bias and variability in observation processes are controlled or known. An integrated analysis can potentially make use of both the unbiased nature of professionally collected data and the high volume of citizen science data for improved inference.
Data integration for citizen science has become increasingly popular, particularly models for species distributions which employ shared spatial terms. However, integration could also be helpful where the focus is on drivers of species diversity and abundance. In the UK agri-environment schemes are key mechanisms to incentivise ecologically friendly management of the countryside, but direct evidence of their benefit for many species is difficult to find. Although professionally designed monitoring is in place, the cost of this means the number of locations surveyed is small and potentially non-representative of national responses.
We investigated whether integrating citizen science data with professional monitoring could enable more representative and robust estimates of agri-environment scheme effects on English butterflies. We describe a simple non-spatial integrated model designed to identify shared effects of agri-environment schemes while controlling for differences in observation processes. We discuss the potential benefits and drawbacks of our non-spatial integration and show how improved confidence in agri-environment effects on butterflies can be obtained from combining different sources of data. .
Authors: Bríd E. O’Connor (Marine Institute of Ireland)+.
Abstract: Biologically sensitive fish, such as slow-growing species with low fecundity, are more susceptible to long-term population declines as a result of anthropogenic pressures than fast-growing species with high fecundity. Many biologically sensitive species are not of commercial interest, resulting in limited data collection and availability, impacting our understanding of the spatiotemporal dynamics of their populations. As such, there is a pressing need for the development of new research methods to improve knowledge on the likely distribution of these species to successfully manage maritime activities for their protection.
Species distribution models (SDMs) are an important tool that can extrapolate predictions of habitat suitability geographically to regions with no known occurrence data, based on the environmental parameters of the areas in which the species are known to occur. SDMs have been widely used for terrestrial species but are as of now less commonly used in marine environments. Joint species distribution models (jSDMs) further build on individual SDMs by modelling species interactions to create habitat suitability predictions for multiple species at once, accounting for their biotic interactions which likely gives a more robust and realistic prediction of suitable habitats. We conducted individual SDMs and jSDMs, identifying substantial differences between their output.
Consequently, we used jSDMs to predict the distribution of biologically sensitive fish species in Irish waters, using fisheries observation and environmental data. We aim to identify areas in which sensitive species and threats, such as fisheries, consistently overlap in order to highlight where these species are most vulnerable to anthropogenic pressures. Outputs of this work will help to improve the understanding of the local variability of sensitive and data-poor fish species and provide support for the expansion and advancement of management approaches that benefit particularly vulnerable species in addition to overall marine biodiversity..
Authors: Yves Hingrat (Reneco International Wildlife Consultants, LLC.)+; Gabriele Sorci (Biogéosciences, UMR 6282 CNRS, Université de Bourgogne Franche-Comté); Michel Saint Jalme (Reneco International Wildlife Consultants, LLC.); Yves Hingrat (Reneco International Wildlife Consultants, LLC.).
Abstract: Climate change presents an urgent crisis, reshaping ecosystems and biodiversity. Birds are highly sensitive to climatic changes and undergo remarkable shifts in their range, reproduction, and biological timing. Understanding the intricate relationship between climate fluctuations and bird reproduction is challenging, hindered by two major obstacles: the difficulty of distinguishing direct climate effects on avian physiology and behavior from indirect habitat impacts (like food availability), and the lack of standardized quantitative approaches to determine suitable climate study windows to avoid a priori period selection based only on species-specific knowledge. Our study addresses these issues by analyzing the effects of temperature on the breeding parameters of the North African Houbara bustard (Chlamydotis undulata), a vulnerable species native to arid regions. Using a 20-year longitudinal dataset from captive breeding of over 1,700 prime-aged Houbara bustards (4 to 8 years old), we developed a novel approach combining a sliding window technique with generalized mixed regression models based on the Student’s t-distribution. This method identifies specific periods where temperature strongly influences reproductive parameters, effectively minimizing confounding habitat biases as birds were housed in individual cages and fed ad libitum. Our findings show that sperm count in males responds to short-term temperature fluctuations up to 41 days before sperm production, with optimal sperm quality at temperatures below 15°C. In contrast, female breeding timing is shaped by long-term climate effects, extending well beyond four months before egg laying, leading to earlier breeding in warmer conditions. Capturing each breeding parameter’s unique reaction to climate over specific timeframes, demonstrates the effectiveness of our methods in understanding animal responses to climate change. Future research will expand to include additional breeding and climate variables, such as precipitation and extreme weather events, to deepen our insights into these complex ecological dynamics..
Authors: Keshab Gogoi (Wildlife Institute of India)+.
Abstract: Deciphering the intricacies of human-carnivore interactions is paramount in the conservation efforts directed towards large carnivores inhabiting human-dominated landscapes. Balancing local communities’ welfare while safeguarding these species represents a significant global conservation dilemma. Therefore, examining systems where humans and large carnivores already coexist might offer valuable insights into how this delicate equilibrium could be maintained. The coexistence of Asiatic lions and humans within the Saurashtra landscape presents such a unique system. Here, robust statistical approaches were used to investigate the nuances of interactions between lions and humans by spatially combining the socio-economic attributes of individual attitudes obtained through questionnaire surveys with aspects of lion ecology obtained through radio telemetry, experience of living with lions, and various tangible and intangible factors with the magnitude of conflicts. Species Distribution modelling was used to understand the spatial processes of interactions between lions and humans (AUC >0.83). The spatial stratification of tolerance towards lions revealed that people have a higher positive attitude towards lions in high to moderate-conflict areas compared to areas with no conflicts. This was contrary to our expectations. To further investigate this dilemma, we used General Linear Mixed Effect Models to understand the factors governing positive attitudes among people. The odds of liking lions increased with increasing experience of living with lions, economic status, conflict magnitude, decreased among pastoralists and away from community tourism hotspots (MR2 0.54, CR2 0.78). We found that despite negative interactions, both are co-adapting with each other, people by socio-cultural tolerance, better animal-rearing practices, strict law-enforcement and lion centric livelihood opportunities, while lions by mostly scavenging and rarely predating prized livestock, rarely attack people, spatially and temporally segregating activity with humans. This exemplifies plausible model of syncing land-sharing with large carnivores through such participatory efforts, making long-term carnivore conservation sustainable in human landscapes elsewhere..
Authors: Gesa von Hirschheydt (Swiss Federal Research Institute WSL)+; Ariel Bergamini (Swiss Federal Research Institute WSL); Klaus Ecker (Swiss Federal Research Institute WSL); Andrin Gross (Swiss Federal Research Institute WSL); Markus Schlegel (Swiss Federal Research Institute WSL); Silvia Stofer (Swiss Federal Research Institute WSL).
Abstract: Comparing detection probabilities between studies or species is challenging, because they depend on many factors. First, species differ in their ecology (e.g., they can be mobile or sessile, seasonal or non-seasonal), their size, conspicuousness and identifiability. Second, sampling methods usually vary between organism groups with sampling units being of different sizes and more or less time being available for searching. And third, studies may also differ with respect to surveyor experience and ambition, and whether citizen scientists are involved. Despite these challenges, we here compare detection probabilities between three different organisms (plants, fungi and lichens) from separate surveys, all conducted at the national scale of Switzerland. Similarities include the sessile nature of all species, the fact that sampling units were clearly defined and systematically searched for all species, and the large total number of species detected (2825 plants, 1936 fungi, 512 lichens). Repeated-visit data are also available from a subset of the sampling units for all surveys, which allows the estimation of detection probabilities using occupancy models. Differences include the number and size of the sampling units (plants: 6769 sites à 10 m2, fungi: 389 sites à 10,000-40,000 m2, lichens: 500 sites à 500 m2), the seasonality of the organisms, and the involvement of volunteer surveyors in the fungal survey. We illustrate the effect of seasonality on detection probability for some example species and compare inter- and intra-study variability. To draw conclusions for the design and analysis of future studies of these organisms, we discuss the interplay between search time, size of the sampling unit, and number of visits to a site..
Authors: Christian Che-Castaldo (USGS)+; Mathew Schwaller (Stony Brook University).
Abstract: While remote imagery has long been used to monitor physical processes, it is only recently that satellites have advanced to the point where animal activity can be seen from space. This breakthrough affords researchers a new, lower-cost option for gathering species abundance and distribution data at large spatiotemporal scales. However, there are challenges associated with remotely sensed wildlife, including accounting for satellite observation errors and that biological processes often occur at spatial scales too small relative compared to the resolution of many satellites. Image super-resolution techniques, which enhance a sensor’s spatial resolution, provide an exciting opportunity to infer changes in environmental processes that occur at sub-pixel scales. In the case of the Landsat satellite archive, which is limited by its relatively coarse resolution, super-resolution can unlock the potential to study planet-wide changes in biodiversity going back nearly 50 years.
We develop a novel approach to incorporate Landsat imagery into models of Adélie penguins (Pygoscelis adeliae) distribution and abundance in Antarctica. Antarctic penguins serve as a key indicator species for Southern Ocean health due to their sensitivity to climate change. However, managers rely on penguin abundance data that is very sparse, raising concerns over how transferable these data are to broad scales. Building on algorithms that classify penguin guano in Landsat imagery, we construct a Bayesian hierarchical super-resolution model that leverages the repeat visits to penguin colonies during the breeding season to estimate the fraction of guano in each pixel while accounting for satellite detection failures. Because penguins packing consistently, these pixel guano area estimates can be summed annually to the colony level to provide a new data input in penguin population models. We show how the automated mapping of Adélie penguin colonies with Landsat allows for the near-complete monitoring of this species at the continental scale..
Authors: Sonia Illanas (Institute for Game and Wildlife Research (IREC-CSIC-UCLM))+; Javier Fernández-López (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); José Antonio Blanco-Aguiar (Institute for Game and Wildlife Research (IREC-CSIC-UCLM)); Carmen Ruiz-Rodríguez (Institute for Game and Wildlife Research (IREC-CSIC-UCLM)); David F Ferrer Ferrando (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); Pelayo Acevedo (Institute for Game and Wildlife Research (IREC-CSIC-UCLM)).
Abstract: Determining the spatial pattern of species abundance at large scales is a challenge that used to be solved by using hunting bags data as a proxy of game species abundance, but the nature of data makes modeling a need.
The use of correlative models for explaining or predicting the spatial patterns of abundance of species can be challenging, because predictors can vary regionally or locally their effect. Different solutions are proposed for considering the possible variations, among which there are adjusting independent models for each environmental homogeneous region (IND), including these regions as one more predictive in a single model (REG), or parametrizing the effects of predictors across space (SVC). These methodologies were applied for modeling hunting bags of different game species in mainland Spain. Models were calibrated under a Bayesian framework using R’ spAbundance package.
Spatial abundance patterns under all methodologies showed concordance with the known distribution pattern of species. Nonetheless, predictions obtained with IND and REG show, in general, a steeply change in predicted values between bioregions (resulting from model parameterization), while a more realistic pattern due to continuous values is observed when using SVC.
Although SVC approach is computationally more expensive, these models could be a good alternative for parameterizing distribution and abundance patterns of species at continental scale, in which their relationship with the environmental gradient can deeply change..
Authors: James A Campbell (Leibniz Institute for Freshwater Ecology and Inland Fisheries)+.
Abstract: Underwater acoustic telemetry is a core methodology for collecting fine-scale movement data for aquatic animals. When telemetry receivers are set up in closely spaced arrays with overlapping detection ranges, the detection times of a tagged animal can be used to estimate its movement paths. In practice, estimating these paths can be challenging. Traditional time-difference-of-arrival methods generally provide positioning accuracy too poor for inferring fine-scale behaviors, while more robust state-space positioning models can be computationally intensive and practically infeasible to run on large datasets.
Here, a novel telemetry positioning method is presented where time-of-arrival positioning is implemented as a state-space model within an extended Kalman filter. The resulting model, termed EK-TOA, provides closed-form solutions to track estimation. Simulated datasets of fish movement within a 2D telemetry array are used to validate the models performance and an example case study of high-volume positioning of a tagged fish is provided. The EK-TOA model can rapidly estimate accurate movement tracks without data-volume limitations while providing good measures of positioning accuracy.
As large, fine-scale acoustic telemetry datasets tracking underwater animals are becoming more prevalent, the EK-TOA model addresses the need for accurate positioning methods that can be applied to these large datasets. The resulting high-resolution movement paths facilitate the application of animal movement statistics to examine how environmental or experimental conditions affect animal movement behaviours..
Authors: CARLOS P.E. BEDSON (Natural England)+.
Abstract: Anthropogenic pressures are strongly implied causes of species range shifts or extinctions. Protected areas, historically chosen to maintain and conserve taxa have been inadequate to retain biodiversity. Amidst changing climates and constricted landscapes, we must increase natural connectivity to aid species’ responses to environmental change.
We considered the White Peak National Character Area of England: a limestone plateau measuring 529km^2 within which are fourteen fragmentary designated Sites of Special Scientific Interest (SSSI’s) totalling 51km^2: limestone valleys, rivers, ash woodland, calcareous grassland surrounded by low biodiversity dairy grassland. We assessed the connectivity requirements for a panel of nationally important locally associated species (three butterflies, one moth, five birds, five mammals, one reptile). We prepared England-wide species distribution models, referencing current and future climate conditions (SSP 245, year 2050). These informed omni-directional circuit theory analysis, identifying high value ecological connectivity areas.
Future climate projections showed for the White Peak eight species maintaining range; one arriving in the future; six others departing. Connectivity analysis for all species today, showed the SSSI network containing 27km^2 of high value connectivity land, the wider non-designated White Peak 95km^2. For the reduced panel of species occurring in future, projections forecast less high value connectivity land in the SSSI network: 19km^2 ; although in the wider non-designated White Peak, high value connectivity land of 69km^2 would still be present. High value connectivity areas were river drainages. Current climate connectivity requirements for water vole (Arvicola Ampibius) showed highest correlation with that of the combined species panel (R2 = 0.56); for future climate, white letter hairstreak butterfly (Satyrium W-Album) demonstrated the highest correlation (R2 = 0.64), reflecting the riparian woodland landscapes, and making both species useful conservation indicators.
This study demonstrates how species distribution models and connectivity analysis can inform nature network planning, connecting fragmented habitats, improving ecological dynamism and resilience to climate change. .
Authors: Daniel W Linden (NOAA Fisheries)+.
Abstract: Population estimation using capture-recapture modeling typically requires that individuals are identifiable by unique marks. North Atlantic right whales (Eubalaena glacialis) can be identified by natural callosity patterns on their heads that are established nearly a year after birth, facilitating extensive aerial surveys used for population monitoring. A well-maintained catalog of individual sightings has been used to annually estimate population size with a Jolly-Seber model using a Bayesian state-space framework. Given that young animals cannot enter the catalog before an established callosity pattern, the terminal year population estimate never includes new calves despite breeding area surveys that provide a near census of births. Here, we illustrate a simple modification to the Jolly-Seber likelihood whereby the number of expected entrants are a function of known births and a parameter representing initial calf survival. The calf-integrated model had more accurate terminal year estimates of population size that remained consistent during subsequent model fitting to additional years of sightings data. While the corrections were fairly small given that per capita calving rates rarely exceed 6%, the clear improvement in accuracy will be helpful to the conservation and management processes for this critically endangered species..
Authors: Marc Grimson (Cornell University)+; Courtney L Davis (Cornell Lab of Ornithology); Yingheng Wang (Cornell University); Brendan Rapazzo (Cornell); Shufeng Kong (Cornell University); Daniel Fink (Cornell Lab of Ornithology); Carla P Gomes (Cornell University).
Abstract: Advances in deep learning are revolutionizing our ability to utilize large observational and environmental datasets to model biodiversity at scale. The development of Transformers, a new architecture initially developed for large language models, allows models to adaptively learn important features from sequential and spatial datasets through a mechanism known as attention. Using this mechanism, spatial models, for example, can automatically learn which spatial patterns and relationships to focus on based directly on raster data, obviating the need to generate tabular summaries of the raster data. By doing so, transformers can unlock new opportunities to model species distributions by extracting novel information from spatial datasets unavailable in human-selected tabular summaries.
In this presentation, we leverage the power of transformers to integrate large-scale multi-modality (spatial raster and tabular) feature sets into a deep-learning joint species distribution modeling (JSDM) framework. Our framework employs a modular, scalable architecture via feature embeddings with a Variational Autoencoder-based decoder to capture the associations between species and features and associations among species. To evaluate the framework we model the occurrence of 200 bird species in the Northeastern U.S. using data from eBird, a participatory science program focused on birds. Because of the opportunistic nature in which eBird data are collected, we also include 6 additional tabular features (e.g. search duration and length) known to capture important sources of variation in detectability.
We evaluate the use of raster spatial feature sets, including surface reflectance data from Harmonized Landsat Sentinel-2 (i.e., high-resolution, seasonal information on land cover) and climate/weather data from Daymet, and find that incorporating these features significantly improves predictive performance compared to only using tabular data on annual land cover/land-use available from MODIS. The integration of multi-modality features promises to improve our understanding of species communities by incorporating other data modalities in JSDMs..
Authors: Ana Couto (BioSS)+; Thomas Cornulier (Biomathematics and Statistics Scotland); David L Miller (BioSS/UKCEH); Katherine Whyte (Biomathematics and Statistics Scotland (BioSS)); Adam Butler (Biomathematics and Statistics Scotland); Esther L Jones (Biomathematics & Statistics Scotland).
Abstract: Understanding resource selection and its associated fitness consequences is a key interest in ecology. Resource selection may be a function not only of intrinsic factors (e.g. sex and body condition) but also availability of extrinsic resources (e.g. habitat or prey). Popular modelling frameworks for studying resource selection (such as resource-selection functions and step-selection functions) rely on defining the availability of environmental space with respect to animal locations. Information pertaining to individual locations can be collected using different surveying techniques, including the use of telemetry devices. Whilst high resolution telemetry data have often been analysed using these frameworks, there are some statistical challenges in analysing such data. Telemetry data are often highly autocorrelated, as multiple observations can be recorded per day, thus violating assumption of independence between locations. Moreover, to deal with intrinsic issues in the data, temporal information is often aggregated to coarser scales in favour of spatially explicit frameworks. In this work, we will review how different statistical approaches for quantifying resource selection (such as inhomogeneous Poisson point-process and logistic regression models) can be applied to telemetry data, both in a frequentist (e.g. using generalized additive models) and Bayesian (e.g. using INLA) framework. Using simulated telemetry data, we aim to illustrate the strengths and weakness of each approach and how these can potentially be improved to account for the particularities of telemetry data..
Authors: Valentina Ruco (University of Turin)+; Ricardo Simon (Office Français de la Biodiversité); Mathieu Garel (Office Français de la Biodiversité, Service Anthropisation et Fonctionnement des Ecosystèmes Terrestres); Nives Pagon (Slovenia Forest Service); Rok Cerne (Slovenia Forest Service); Anne Loison (Laboratoire d’Ecologie Alpine, Université de Savoie Mont Blanc); Francesca Marucco (Department of Life Sciences and Systems Biology, University of Turin).
Abstract: In human dominated landscapes, animals face the dual threat of disturbance from human activities and the risk of being hunted and predated. This prompts them to carefully balance their time and space utilization between foraging resources and areas offering refuge from perceived predation risk. Numerous studies have confirmed that many species perceive humans as predators, leading to the same adjustments in spatial-temporal behavior. However, despite mounting evidence of the widespread impact of humans on animal behavior, our understanding of how human activities and the related lethal risk influence habitat selection is still limited and challenging, particularly in environments where the ‘landscape of fear’ is further shaped by the presence of natural predators for whom presence is by nature less predictable than for humans. We investigated roe deer spatial response in three mountain areas across the Alps (in France, Italy and Slovenia), differing in terms of presence of natural predators, hunting pressure, anthropogenic and environmental features. We analyzed habitat selection from GPS data of 45 adult roe deer (23 in France, 12 in Slovenia and 10 in Italy) during the hunting season (September to December), using resource selection functions and spatially fine-scale harvest data accounting for differences in hunting type (drive hunts vs. approach/wait hunt). Our results confirm the primacy of anthropogenic drivers on habitat selection by roe deer: animals exhibited strong avoidance both for areas with relatively high light pollution (an index of human urbanization) and a high risk of harvest by humans. Roe deer were also selected for proximity to human buildings, when hunting risk was relatively high, supporting the human shield hypothesis. Our results provide insights into the drivers that shape the habitat selection of a large ungulate in human-dominated landscapes, a key consideration for informing effective management and human-wildlife coexistence issues. .
Authors: Sara Martino (Department of Mathematical Sciences, NTNU); Jafet Belmont Osuna (School of Mathematics and Statistics, University of Glasgow); Erin Bryce (School of Mathematics and Statistics, University of Glasgow); Leonardo Capitani (Swiss Federal Institute for Forest, Snow and Landscape Research WSL); Virginia Morera-Pujol (University College Dublin)+; Megan Morton (School of Mathematics and Statistics, University of Glasgow); Janine B. Illian (School of Mathematics and Statistics, University of Glasgow).
Abstract: In ecology, spatial point processes (SPP) have become a popular statistical tool to model presence only data, both on their own, and as a building block for model complex models (i.e. integrated species distribution models). However, inference on SPP, particularly when incorporating spatially explicit random effects, poses significant complexity. A common strategy involves adopting the so-called log-Gaussian Cox process (LGCP) and relying on the INLA/SPDE implementation for efficient inference. Central to this approach is the representation of a spatially continuous Gaussian random field as the solution of a stochastic partial differential equation (SPDE) defined over a triangular mesh spanning the spatial domain. The mesh definition is an important part of the model implementation. Equally crucial, yet often overlooked, is the role of integration points employed in numerically solving the stochastic integral inherent in the LGCP likelihood. This study underscores the critical interplay between mesh resolution, integration points, and the spatial scales of both random fields and covariates. Inadequate model covariate resolution relative to that of the mesh can lead to highly unreliable numerical integration. This compromises parameter estimation, producing widely different covariate coefficients depending on the mesh and covariate resolutions.
Using both simulations and real-world ecological data, we illustrate these challenges and offer guidelines to empower users in navigating these complexities and mitigating potential pitfalls when using these methods in ecological applications such as species distribution models based on presence-only data..
- Session: P2 & P4
- Time: Tuesday (4:30-6:30pm) & Friday (10:00-11:00am)
- Room: Great Hall
Authors: Coralie C Williams (University of New South Wales)+.
Abstract: Simulation studies are essential tools to assess the performance of statistical methods. Functioning as controlled experiments, simulation studies generate data from known underlying processes. However, unclear or incomplete reporting of simulation studies can impact their reliability and potentially lead to the misuse of statistical methods. In this study, we conducted a survey of 100 articles in ecology and evolution journals that included a simulation study evaluating a statistical method. Our survey aimed to characterise these simulation studies and assess their reporting practices. To structure our assessment, we used the ADEMP framework proposed by Morris et al. (2019, Stat Med, 38, p2074), comprising of: Aims, Data-generating mechanisms, Methods, Estimand, and Performance measures. Expanding on the ADEMP framework, we developed items to assess reporting practices and quality of each simulation step. We found that a small portion of papers reported a Monte Carlo uncertainty (17%) and justification of the number of simulation replications (1%). Additionally, we contrasted our results with two comparable surveys of simulation studies in medicine and psychology. We found that simulation studies in ecology and evolution tend to report software (91%) more frequently than in medicine (62%) and provide comprehensive code (68%) more frequently than in psychology (36%). Our survey indicates a positive trend towards open science practices for reporting simulation studies in ecology and evolutionary biology. However, there is room for improvement for more transparent simulation studies to ensure credible evaluation of statistical methods. The reporting items we developed in our study, using the ADEMP framework as a basis, provide a preliminary template for broader guidelines on how to report a simulation study. Such reporting guidelines for simulation studies would be applicable not only in ecology and evolutionary biology but potentially to other fields of research..
Authors: Carole Niffenegger (Swiss Ornithological Institute)+; Fränzi Korner-Nievergelt (Swiss Ornithological Institute).
Abstract: Environmental conditions can have complex effects on reproductive phenology and reproductive success. So far, many studies report an earlier onset of reproduction in response to rising temperatures. However, most studies are based on a few model species with detailed long-term monitoring programs, mostly from temperate ecosystems. Similar studies from other habitats, for example from high elevation, are scarce which is probably often caused by a lack of long-term systematic monitoring programs. Here, we use citizen-science data to study breeding phenology of the White-winged snowfinch (Montifringilla nivalis), a high-elevation specialist which is considered climate-sensitive. We derived the start, end and duration of the breeding season for this species using hierarchical generalized additive models (GAMs) based on more than 12’000 observations from a citizen-science platform and explored the relationship between breeding phenology and environmental conditions. The results indicate that higher pre-breeding temperatures and reduced April precipitation were associated with earlier breeding onset, while higher temperatures during the breeding season shortened the breeding season. An overall increase in temperature is therefore expected to cause a shift towards earlier breeding and an overall contraction of the breeding season. In a second step, we attempt to link the phenology to reproductive success which allows us to improve predictions on the effects of environmental change on the population dynamics of this species..
Authors: Oluwadare O Ojo (Federal University of Technology Akure Nigeria )+.
Abstract: Birds have proved to be an excellent indicators of biodiversity because they are easily observed compared to other animals. In recent times, Bayesian method has gained popularity in the study of ecology. The major advantage of Bayesian method over classical method is the use of prior information of the situation being modelled. This research considered a Bayesian method of estimation to analyze the diversity and abundance of avian species in Nigeria. Data were collected during the wet and dry seasons for a period of twelve months. Two dominant families of the bird species observed were Estrilda and Malimbe. The results of Bayesian method of estimation was then compared to classical method of estimation. The results show that the Bayesian method of estimation outperformed classical method in the analysis of diversity and abundance of avian species in Nigeria. Also, Bayesian estimates revealed that abundance of bird species are higher in dry season than wet season. .
Authors: David L Miller (BioSS/UKCEH)+; Ken B Newman (Biomathematics & Statistics Scotland).
Abstract: Aphids are a pest insect that can cause serious issues for agriculture. While feeding on plants is destructive, they also serve as vectors for plant viruses. Prediction of aphid arrival date (from diapause or development from eggs) is therefore a key part of any mitigation strategy. Arrival date is influenced by weather: consecutive days of warm temperatures increase the chance of arrival. In this talk, we’ll explore the use of distributed lag models to characterize this relationship in order to improve predictions of arrival and provide nowcasts for farmers. These models fit within a generalized additive modelling framework and can be fitted in the popular R package mgcv. We will show examples of these models with aphid arrival data from the UK suction trap network..
Authors: Haidee Tang (NIAB)+.
Abstract: Developments in phenology models better relate mathematical models to biological phenomena. Previous models were restricted to using a fixed amount of overlap — either complete or none. This biologically confounds the model as we are uncertain whether endodormancy (plant induced inhibition) and ecodormancy (inhibition due to the environment) phases overlap. The PhenoFlex model introduces flexibility by allowing varying degrees of overlap between chilling and heat sub-models. The extent of overlap is dependent on the model parameters. Usually, the model parameters are trained on a single cultivar, so we explored the effect of parameterisation on multiple cultivars. In our study, we tested the effect of multi-cultivar parameterisation of the PhenoFlex model with thirty-one cultivars. We tested three parameterisation approaches: 1. Cultivar-specific: tailored parameters for each cultivar. 2. Group-specific (K-means): groups derived by cultivar-specific parameter estimates. 3. Common model: a model fitted with all thirty-one cultivars. We observed that the best model accuracy from the K-means grouped approach, suggesting that performance may depend on the grouping. We suspect that performance improves with strong genetic relationships between members of the same group and vice versa. The cultivar-specific approach yielded the best performance in the training data but had the worst overall performance on the test data, indicating the risk of overfitting. The common model centralised the predictions of the cultivar-specific approach, resulting in predictions that fall between the extremes observed in the cultivar-specific approach. Furthermore, the models of the cultivar-specific and grouped approaches were bias towards late flowering predictions in the test data. Lastly, we compared results of the PhenoFlex model with the StepChill model, a model which dis-allows any overlap, and we show that the PhenoFlex model results in better predictive accuracy..
Authors: Yuxin Zhang (RCEES, CAS)+.
Abstract: Herbivory, a classical negative biotic interaction between plants and herbivores, is influenced by a complex array of biotic and abiotic factors. Competing hypotheses—“top-down control” and “bottom-up control”—were developed in isolation to explain herbivory patterns in nature, but studies evaluating both are scarce. We used structural equation modeling, a casual network inference approach, to evaluate the role of top-down and bottom-up factors and their interactions in shaping the variation of herbivory on Oak (Quercus liaotungensis) in a deciduous temperate forest in Beijing, China. We found that the intensity of the ant-aphid mutualism, a key top-down driver and a prime example of a positive biotic interaction, exerts a significant direct negative effect on the herbivory of Oak trees. Moreover, the indirect effects of abiotic factors and plant defense on herbivory are mediated through the intensity of the ant-aphid mutualism. Our results highlight the importance of positive biotic interactions in shaping the pattern of negative interactions in the community..
Authors: Emy Guilbault (University of Helsinki)+; Jarno Vanhatalo (University of Helsinki); Mirkka Jones (Aalto University); Andrea Santangeli (Institute for Mediterranean Studies (IMEDEA)); Laura Antao (University of Helsinki); Anna-Liisa Laine (University of Helsinki); Tomas Roslin (University of Helsinki); Marjo Saastamoinen (University of Helsinki).
Abstract: Biodiversity is changing at an unprecedented rate, with land use and climate change being its major drivers. Yet, evaluating the magnitude and relative roles of their impact remains challenging, as their effects vary with species vulnerability and resilience, while the magnitude of climate and land use change varies over space and time. Here, we used 503 species from the major terrestrial taxa monitored over 20 years across Finland to evaluate the relative importance of climatic and land use changes using joint species distribution modelling. We used a predictive and conditional variance partitioning approach to characterize the relative importance of climate and habitat, and their context dependency, across the whole country. Overall, habitat change was the main driver of biodiversity, but its dominance and uncertainty in explaining species occurrences were highly context-dependent. Our findings emphasize that multiple drivers and their interplay need to be considered, such that conservation efforts should focus on habitat management without disregarding the relevance/effects of climate change, especially in northern regions where climatic is changing the most/fastest. To better tackle the biodiversity crisis, conservation actions need to be tailored based on species and spatial contexts. .
Authors: Louis Hunninck (Swiss Ornithological Institute)+; Fränzi Korner-Nievergelt (Swiss Ornithological Institute); Jan von Rönn (Swiss Ornithological Institute).
Abstract: Bayesian population models are often inhibitively slow for ecologist using large datasets or complex model structures. Following Yackulic et al. 2020, we applied marginalization to various multistate capture-mark-recapture-recovery (CMRR) models estimating survival for 3 common nocturnal and daytime raptor species: tawny owl (Strix aluco), barn owl (Tyto alba), and common kestrel (Falco tinnunculus). We used 15 years of ringing data collected mostly by volunteers of the Swiss Ornithological Institute, covering all of lowland Switzerland, and consisting of between 11,000 and 16,000 ringed individuals – yielding between 20,000 and 33,000 observations – per species. We used hierarchical spatial clustering to assign individuals to natal home ranges allowing us to estimate natal dispersal and/or account for geographic variation in population parameters. We show how marginalisation, with and without aggregating capture history matrices, vastly speeds up model runs, in JAGS but especially through Stan. We provide code for all our models to show that including marginalisation in multistate models, or switch from JAGS to Stan, does not present any additional coding difficulty. Adult survival of barn owl was highly variable among years and low survival rates correlated with extreme cold winters. Geographic variation in survival probabilities was low for all three species, though recovery and detection probabilities varied considerably between areas, probably due to local differences in ringing and recapture effort. We show that applying marginalization is straightforward and vastly reduces run time while yielding indistinguishable parameter estimates as from traditional models, opening up Bayesian methods for those users still discouraged by traditional run times..
Authors: Loïc Baptiste Pages (CNRS - CEFE)+; Olivier Gimenez (CNRS).
Abstract: Large carnivores are recovering in Europe, and their interactions with human activities increase consequently. In France, the wolf population is managed by legally killing animals for reducing livestock depredation. This management framework needs to be evidence-based and sustainable, so that quotas are justified and adjusted to avoid damaging the wolf conservation status.
Here we developed a Bayesian state-space model to support adaptive management of the wolf population in France. Our approach is used to predict the probability that the forecasted population size will be below or above the management objectives when subjected to different harvest quotas.
First, we present our model in two components. We used a Jolly-Seber capture-recapture model to estimate annual abundance. We include heterogeneity in detection through a mixture latent process to account for higher detectability of dominant wolves. Then, we quantified population dynamic by considering two ecological mechanisms, the exponential and the logistic models, which were fitted to the population size state variables from the capture-recapture model. The approach was implemented in a Bayesian framework using MCMC algorithms to fully propagate uncertainty between the two components.
Second, we evaluated our adaptive management approach in a simulation study considering contrasted management actions (no killing vs. increasing quotas) and contrasted mechanisms underlying population dynamics (increasing vs. stable population). We also considered a non-adaptive approach to illustrate the benefits of our proposal.
Third, we applied our approach to data on the French wolf population. We used encounter histories for almost 3000 wolves that have been identified individually through scat genotyping and monitored over 28 years.
Our approach highlights the effectiveness of adaptive management compared to non-adaptive population management that has been used for several decades on the French wolf population..
Authors: Louis Backstrom (University of St Andrews)+; Hannah Worthington (University of St Andrews); Alison Johnston (University of St Andrews).
Abstract: Correctly identifying which species have become extinct, and when, is an important component of biodiversity conservation. Extinctions are almost never directly observed, meaning that practitioners must rely on approaches that estimate the probability of extinction using information on a species’ biological traits, threats, and observation histories. Effective methods need to be conservative enough to avoid incorrectly inferring species as being likely extinct when they are still extant, without being so overly cautious that they fail to identify extinct species, as both of these issues have negative implications for the conservation of threatened species (i.e., methods must minimise both Type I and Type II errors). Over the past decade, several methods have been developed that focus on inferring extinction probability using observation histories of “lost species” for which there are few or no recent records. By and large, however, these approaches have not been spatially explicit (they do not consider how the process of extinction may vary across a species’ range), nor have they been designed for use with citizen science data, which are an increasingly important source of information in biodiversity conservation. This presentation will outline my current efforts to develop a method to estimate extinction probability that address both of these issues, summarising progress to date and highlighting ongoing challenges requiring further research..
Authors: Michael A Spence (Cefas)+.
Abstract: Often there are multiple models to describe a specific system and ultimately any decision can be sensitive to the model used. Choosing a single model from a suite of models is potentially throwing away information, meaning that the decision is not as well informed as it could be. Further, ignoring alternative models does not rigorously quantify the “true” uncertainty. An alternative to choosing a single model is to combine them using an ensemble model. Many approaches to ensemble modelling involved weighting models and follow some strong assumptions, that a) are often not met, and/or b) are not utilising all the information. In this talk I will discuss these methods and compare them with alternative schemes. I will demonstrate that even in cases when the models are similar to the true data generating process alternative schemes do better than weighting the models. In this talk I will discuss common ways of combining models, demonstrating them on a number examples and highlight that there is more to ensemble models than model weights.
Keywords: ensemble modelling, uncertainty quantification, risk
.
Authors: Emily L Coles (Aberystwyth University)+; Kim Kenobi (Aberystwyth University).
Abstract: Complex social structure exists among many mammals, including cetaceans, and plays an important role in population dynamics and patterns of behaviour. Social networks are widely used to investigate the social structure of animal populations. However, despite this, limited sampling of non-human animal observation data continues to present challenges. Within an animal social network, nodes typically represent individuals that are connected with edges that represent behavioural interactions or associations. Often, network studies use methods that rely on aggregations of the presence or absence of edges to construct images of network structure. However, many real-world data sets are subject to error. Observational methods of network data collection contain incomplete data, varying degrees of certainty, and potential observer bias. For example, boat-based surveys are limited to the number of boat trips and visibility may be influenced by weather, distance, and diving behaviour. The possibility of missing data due to small or incomplete samples can lead to uncertainty regarding the existence and strength of associations and interactions, which is then carried forward in statistical methods. Therefore, network measurements may only hint at network structure. In recent years, methods adopting a Bayesian modelling framework have been proposed to incorporate uncertainty into network analysis, such as those proposed by Young, Cantell, and Newman (2020). In this work these methods are applied to infer network structure from photo-identification data collected using boat-based surveys over thirty years for a well-studied population of bottlenose dolphins, Tursiops truncatus, along the east coast of Scotland. Animal interactions are dynamic processes, therefore temporal dynamics of social network structure are considered by generating time-aggregated networks for each year. We are developing existing methods and highlighting the applicability of Bayesian methods to analyse social networks of wild animal populations..
Authors: Sarah R Christofides (Cardiff University)+; Emile Blin (Maastricht University); Ben Stevenson (University of Auckland).
Abstract: Wood-decay fungi, and especially cord-forming basidiomycetes, are key regulators of biogeochemical cycles at the ecosystem and global scales. Yet the structure of their communities remains largely unstudied, with no estimates of population abundance and density to date. This is largely due to experimental difficulties associated with fungal surveys. These are strikingly similar to those associated with animal surveys, which have been successfully tackled with spatially explicit capture-recapture (SECR) modelling. The spatial behaviour of macro-fungi resembles that of animals in many respects. In the light of these similarities, we tested SECR modelling on wood-decay basidiomycetes to develop a method for the potentially generalised application of SECR to macro-fungi. This is the first study (to our knowledge) to apply SECR to a non-animal taxon. Woody debris were surveyed in a patch of forest for fungal cords, which were cultured and identified at the individual level by somatic incompatibility testing. The distribution of individual detections was used to estimate the orders of magnitude for mean detection probability, territory size, and population abundance and density of cord-forming fungi for the first time. We also identify experimental design recommendations for the promising application of SECR to fungi..
Authors: Rachel L Drake (University of St Andrews)+; Alison Johnston (University of St Andrews).
Abstract: Citizen Science data have become increasingly widespread in applied ecological research. The structure of these data can range from rigorous protocols to the submission of single observations to public databases. Inferring non-detections from complete checklist data is straightforward, however non-detections from opportunistic data are challenging to infer.
This study aims to both develop methods for improving absence inference in both checklist and opportunistic data types, as well as providing novel insight into possible bias present in model validation. We explored different ways to process opportunistic data to see how well we could retrieve similar results to occupancy models from structured data types. To accomplish this we use eBird and iNaturalist data to compare both checklist and opportunistic data types across two species designed to span a range of ecology, observer expertise and observer preference.
Our results reveal that occupancy models applied to both complete and opportunistic datasets yield divergent ecological inferences. In certain instances, these differences are as extreme as coefficients of landcover variables in opposite directions. Additionally, we compared model predictions from a variety of models to two benchmark datasets and two validation datasets, each derived from either checklist or opportunistic data. Predictions from these models exhibit significant variation across datasets and model types.
The findings emphasise the importance of preprocessing opportunistic data in occupancy modelling, along with careful planning of model validation to mitigate potential sources of bias.
Keywords: occupancy modelling, citizen science, community science, eBird, preprocessing, model validation..
Authors: Jorge Mestre Tomás (Institute of Marine Sciences (ICM) - CSIC)+; Alba Fuster Alonso (Institute of Marine Sciences (ICM) - CSIC); Marta Coll (Institute of Marine Sciences (ICM) - CSIC); David Conesa (Department of Statistics and Operational Research, Universitat de València); Jose Maria Bellido (Spanish Institute of Oceanography (IEO) - CSIC).
Abstract: Assessing the state of the world’s marine communities and monitoring those of fishery interest is essential in marine science, especially given the impact of intense fishing activities and global climate change on ocean ecosystems. These stressors, which can disproportionately affect different species, can spread their effects through the food web, prompting biodiversity changes and trophic levels shifts. The scientific literature often includes terms like “fishing down marine food webs”, “pelagification”, and “trophic amplification” of ocean food webs, emphasizing the interest and complexities of these ecological mechanisms. Our research aims to investigate if Large Marine Ecosystems (LMEs) have been moving towards homogenous community compositions over time. To do so, we analyze trends in catch estimates obtained from the Sea Around Us database of global catches between 1950 to 2019. Specifically, we investigate whether there has been homogenization in the proportion of catches across different functional groups. We apply Dirichlet regression within the INLA statistical framework, which allows us to explore temporal patterns in functional group catches over long-term periods and at global scales. The findings of our study will help us to understand whether changes in marine ecosystems are associated with increasingly uniform catches in LMEs, which could indicate a trend toward global homogeneity of functionality..
Authors: David F Ferrer Ferrando (Institute for Game and Wildlife Research IREC (CSIC-UCLM))+; Pedro Tarroso (CIBIO/InBIO, Centro de Investigação em Biodiversidade e Recursos Genéticos da Universidade do Porto); Jose L. Tellería (3Department of Biodiversity, Ecology and Evolution, Universidad Complutense, Madrid); Pelayo Acevedo Lavandera (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); Javier Fernández-López (Institute for Game and Wildlife Research IREC (CSIC-UCLM)).
Abstract: Understanding species abundance patterns across broad spatial scales is crucial for deciphering ecological processes. However, accurate abundance estimates often demands extensive sampling efforts, posing logistical challenges at large spatial scales. The use of suitability models (based on occurrences) as a proxy of abundance has garnered interest in this context. Previous studies suggest a triangular relationship between suitability and abundance but remains contentious due to varying results. This variation arises from limiting factors that affect abundance but are unaccounted in the suitability models. This study investigates how species spatial pattern (as an ecological factor) and spatial resolution (as a methodological factor) influence this relationship. Through simulations and real-world case studies of two passerine species in the western Mediterranean, we explore the effects of two different spatial patterns (aggregated vs. uniform) and three levels of resolution (macro, meso, micro) on the abundance-suitability relationship. Our findings reveal that higher spatial resolution improves the relationship between abundance and suitability, and that species exhibiting aggregated spatial patterns show stronger relationships. Additionally, we observe a possible interaction between spatial pattern and resolution, wherein species with aggregated patterns benefit more from the initial resolution improvements compared to those with uniform patterns. Our results underscore the importance of considering both ecological and methodological limiting factors in species suitability modeling when models are aimed to obtain accurate abundance patterns. We highlight the need for fine-resolution data (local-scale) to make reliable inferences about species abundance from distribution models and suggest potential strategies for integrating local information to improve abundance predictions. Overall, this study contributes valuable insights into the complex interplay between spatial pattern, resolution, and abundance-suitability relationships..
Authors: Wesley M. Hochachka (Cornell Lab of Ornithology)+; Tom Auer (Cornell Lab of Ornithology); Cynthia Crowley (Cornell Lab of Ornithology); Daniel Fink (Cornell University); Shawn Ligocki (Cornell Lab of Ornithology); Matt Strimas-Mackey (Cornell Lab of Ornithology).
Abstract: Imperfect detection is an essentially universal characteristic of field surveys of organisms. Highly structured surveys control for much of the variation in detection rates, but likely some variation in detection rates will be present in data from any survey. To model this variation requires predictors of varying detection rates, for example measures of observer effort like durations of (varying) observation periods. However, relevant information is not always recorded. In such cases, list length—the number of species recorded during an observation period—can be used as an index of observers’ effort, with the caveat that list length is only an appropriate index of observer effort when the pool of potentially observed species is constant (i.e. habitats and times of year are consistent across surveys). Here, we examine the practical limitations to the assumption of constant species pools by comparing the value of list length information with recorded information about observer effort using data from the participatory science project eBird, which contain multiple descriptors of observers’ effort. As expected, list length was largely a function of variation in observation effort. While this qualitative association was consistent over long distances, quantitative relationships between measures of observer effort and list length broke down, with biased predictions in cross validation even over distances of a few tens of kilometers in one data-rich region of the northeastern United States. Controlling for variation in observer effort using list length led to less accurate models of observations of individual bird species than using direct measures of observer effort. We conclude that list length should only be used as a descriptor of observation effort if there is no alternative, and even then list length cannot be assumed to provide a quantitatively stable index of variation in observer effort across even moderately large distances..
Authors: Murat Ersalman (University of Helsinki); Mervi Kunnasranta (Natural Resources Institute Finland); Markus Ahola (Swedish Museum of Natural History); Anja Carlsson (Swedish Museum of Natural History); Sara Persson (Swedish Museum of Natural History); Britt-Marie Backlin (Swedish Museum of Natural History); Inari Helle (Natural Resources Institute Finland); Linnea Cervin (Swedish Museum of Natural History); Jarno Vanhatalo (University of Helsinki)+.
Abstract: Integrated population models (IPMs) are a promising approach to assess and manage wildlife populations in dynamic and uncertain conditions. By combining multiple data sources into a single, unified model, they enable the parametrization of versatile, mechanistic population models that can predict population dynamics in novel circumstances. We developed a Bayesian IPM for the Baltic ringed seal Pusa hispida botnica population inhabiting the Bothnian Bay in the Baltic Sea. Despite the availability of long-term monitoring data, dynamic environmental conditions, varying reproductive rates, and the recently re-introduced seal hunting have hindered monitoring of the Baltic ringed seal for over a decade, leading to, for example, uninformed decisions regarding hunting quotas. We fit our Bayesian IPM to census and various demographic and reproductive data from 1988 to 2022 to provide a comprehensive assessment of the past development of the Baltic ringed seal population and an analysis of its future development under alternative hunting scenarios. Currently 20,000 to 40,000 ringed seals inhabit the Bothnian Bay and the population is increasing at a rate of 2.6% to 7.3% per year. Baltic ringed seal reproductive rates have increased since 1988. This led to a substantial increase in the population growth rate from 1988 to 2015. However, since 2015 the reintroduction of seal hunting has noticeably decreased the growth rate. Further decline in population growth can be expected if hunting quotas are raised by more than 30 licenses per year. Our results also support an hypothesis that a greater proportion of seals haul-out on ice under lower ice cover circumstances, leading to higher aerial survey counts in such years. In general, our study demonstrates the value of mechanistic IPMs for monitoring and managing natural populations under changing environments, and supporting science-based management decisions..
Authors: Taisiia Marchenkova (Land of the Leopard National Park)+.
Abstract: Understanding long-term demographic structure is essential for properly managing the population recovery process, especially with critically endangered species such as Far Eastern leopards. We assessed the population demographic parameters of the Far Eastern leopard using camera trap surveys from 2014 to 2023. We used open population spatial capture-recapture models with a maximum-likelihood approach to estimate density, abundance, number of resident animals sex ratio, recruitment and growth rate. The survival rate of adults was estimated by using the Bayesian Cormack-Jolly-Seber open population models. Over the decade, the population more than doubled, exhibiting an impressive annual growth rate of 12%. Throughout the study period, the sex ratio consistently favored females, initially with nearly twice the prevalence, gradually diminishing in later years. The percentage of resident animals for both females and males were similar. The survival rate for both genders was higher than 80%, but the recruitment rate of females was twice as low as that of males. This comprehensive analysis elucidates the intricate population demographic structure and its evolving trends, providing invaluable insights for the conservation of the Far Eastern leopard subspecies and other leopard populations globally. .
Authors: Varsha S Shastry (Mississippi State University)+; L. Mike Conner (The Jones Center at Ichauway, Georgia); Gail Morris (The Jones Center at Ichauway, Georgia); Andrew J. Royle (USGS Patuxent Wildlife Research Center); Dana Morin (Mississippi State University).
Abstract: Spatial capture recapture (SCR) models use trap locations in conjunction with encounter history data to account for imperfect detection and estimate population parameters. Typically, density is the parameter of interest, but the spatial information included also allows inference about habitat use and selection. Spatial covariates can be incorporated in inhomogeneous point process models with the point process model representing second order selection, and ecological distance models can be used to assess third order selection. SCR applications assessing habitat features in relation to frequency of detections at trap locations can also be interpreted as microhabitat use. Microhabitat use is studied in small mammal communities to better understand and test theories regarding niche theory and coexistence. Generally, capture-recapture data is reduced to trap level counts and estimated using generalized linear models. However, this approach does not take into consideration other confounding factors that can influence microhabitat use and realized niche including density and differences in home range movements that can vary with habitat, density, and competition. These confounding factors can be accounted for within SCR models by estimating density and spatial scaling parameters while inference is focused on baseline encounter probability. We applied an SCR approach to assess microhabitat use to a long-term small mammal capture-recapture data set collected at The Jones Center at Ichauway, Georgia, USA. We assessed drivers of microhabitat use for 3 small mammal species using multi-session SCR models and estimated differences in microhabitat selection among them and in response to differences in predation and disturbance. Top models indicated high variation in density across years and sites, and the spatial-scaling parameter also varied, demonstrating the value of accounting for both parameters in microhabitat use assessments..
Authors: Brage Førland (University of Bergen)+; Hans J. Skaug (University of Bergen); Megumi Takahashi (Institute of Cetacean Research); Luis Pastene (Institute of Cetacean Research).
Abstract: CKMR is a powerful method to estimate population size and other demographic parameters. In CKMR, sampled relatives has the same role as a recaptures in ordinary mark-recapture analysis. For example, a sampled individual can be considered a marking of its parents at the time of birth, and the capture of the parent as a recapture. Without age information, the time of marking is unknown, and standard CKMR methods are difficult to apply.
In this study, we use baleen whales as a motivating example, and develop a CKMR model for the situation where the ages of the sampled individuals are unknown. Using simplified assumptions about stable age distribution and life history, we derive equations for the sampling probabilities for parent-offspring pairs, half sibling and grandparent-grandchild pairs sampled at different points in time. The resulting model is parametrised in terms of population size, population growth rate and mortality, and estimates of these parameters can be found by maximising the pseudo likelihood obtained by summing over all pairs.
We apply the model to a sample of 161 biopsies from southern right whales. The biopsies are genotyped at 14 loci, which is insufficient for certain kinship classification. To account for this, we use a mixture model, with the true kinship as a latent parameter.
We find that the model is identifiable, and we are able to obtain estimates of population size, population growth rate, and mortality, but that the uncertainties are high for our example data set. We further discuss the impact of the assumptions on the results, and under what demographic assumptions and sampling scenarios the model is valid. .
Authors: Braden Scherting (Duke University)+; David Dunson (Duke University).
Abstract: Joint species distribution modeling (JSDM) is a powerful tool for linking theories of ecological community assembly with observed abundances. State-of-the-art probabilistic JSDMs rely on generalized linear latent variable models to summarize relationships between species’ abundances and environmental conditions as well as relationships among species. Modern semi-autonomous biomonitoring designs are currently generating more data on more species than ever before, which exhibit extreme sparsity, overdispersion, and complex interspecies dependences; such data resist the latent linear structure of JSDMs and simply defeat most methods. Motivated by large-scale Malaise and AudioMoth sampling efforts, we develop a new statistical model for high-dimensional integer-valued data that accommodates sparsity without additional model components (i.e., hurdle or zero-inflation processes), flexible marginal dependence between species, and auxiliary information about sites and species. As consequence of the pyramid-shaped model structure, binary “barcodes” are estimated for each site and species, providing a fine-scale clustering, and sites are further aggregated into robust habitat partitions. This method provides an important alternative to poor-fitting log-linear Poisson models or discarding non-detections in favor of presence-only modeling. Through simulations and challenging applications, we illustrate the utility of the proposed approach..
Authors: Chathuri Samarasekara (RMIT University); Ian Flint (the University of Melbourne); Yan Wang (RMIT University)+.
Abstract: Ecological point patterns, encompassing diverse objects like trees and animal presence, are often analyzed using spatial point processes in current studies. However, multivariate spatial point process applications in ecology mostly rely on descriptive methods, limiting their scope. To address this, the multivariate log-Gaussian Cox process is introduced, expanding beyond the traditional bivariate case. This model overcomes the limitations of the Poisson process by capturing the clustered patterns common in tree species. Additionally, the Saturated Pairwise Interaction Gibbs Point Process model is proposed to handle challenges arising from an increasing number of species. This work thoroughly investigates and compares the performance of the multivariate log Gaussian Cox process and the Saturated Pairwise Interaction Gibbs Process through simulation studies and real data examples. By highlighting the strengths and weaknesses of each model, the paper serves as a practical guide for applying these methods in ecological studies..
Authors: William B Lewis (University of Georgia)+; D. Clay Sisson (Tall Timbers Research Station); Justin Rectenwald (Tall Timbers Research Station); Chloé R Nater (Norwegian Institute for Nature Research); James Martin (University of Georgia).
Abstract: Effectively managing wildlife populations necessitates accurate estimates of vital rates and an understanding of how vital rates affect population growth. Juvenile survival can be a key driver of population growth; however, this parameter is often difficult to assess, and so data is lacking for many wildlife populations. Integrated population models (IPMs) explicitly model all demographic processes affecting population dynamics, and so could be used to inform estimates of juvenile survival in the absence of direct data. We extend the framework of Arnold (2018) in an IPM framework to show how data on breeding productivity and post-breeding population size and structure can be used to inform estimates of juvenile survival. Specifically, breeding productivity data can be used to estimate the number of young produced while post-breeding surveys and age ratios from trapping or harvest data can be used to estimate the number of surviving juveniles, with the difference inferring rates of juvenile survival. We apply this methodology to develop an IPM and estimate juvenile survival from long-term demographic data collected from a population of Northern Bobwhites (Colinus virginianus) in southern Georgia, USA. Juvenile survival exhibited limited model convergence when specified with a vague prior, though incorporating an informative prior increased convergence. Despite requiring an informative prior, integrating seasonal fecundity, post-breeding counts, and post-breeding age-ratio data successfully updated the prior distribution for juvenile survival. Our results suggest that combining productivity and post-breeding data with prior information could be broadly useful for improving estimates of juvenile survival in an IPM framework..
Authors: Chun-Huo Chiu (National Taiwan University)+.
Abstract: Sample coverage, the proportion of individuals belonging to observed species in a sample, is a crucial metric for assessing the completeness of a sample. Unlike equal sample size, equal sample coverage has emerged as a preferred standard for comparing diversity across multiple assemblages. This standardization facilitates a more accurate comparison of richness ratios between samples, reflecting the true relationship between assemblage diversities. Traditionally, the Good-Turing estimator has provided robust and accurate estimates when individuals are randomly sampled. However, quadrat-based abundance data are commonly utilized to assess species diversity. In quadrat-based abundance data, the sampling unit is a plot, net, trap, or transect randomly selected from the target area, and the number of individuals (or biomass; cover area) for each observed species within the sampled plot is recorded. In this context, the individual is no longer randomly and independently collected. The Good-Turing estimator of sample coverage may exhibit significant bias, particularly when individuals exhibit highly spatial aggregation patterns. In this study, according to the sampling with replacement and without replacement, I separately derive the estimator of sample coverage based on quadrat-based abundance data. Additionally, I derive the corresponding estimators’ variances using asymptotic methods. Finally, three censused forest data sets are utilized to evaluate the statistical performances of the newly proposed estimators..
Authors: Mia Goldman (University St Andrews )+.
Abstract: To make informed decisions regarding the management of grey seal populations in the UK, accurate pup production estimates are critical. Previous approaches face limitations such as inflexible observation processes and an inability to share information across colonies and years. This study proposes a novel approach using a Bayesian state-space model. Data are derived from serial aerial counts of colonies during the breeding season where only a portion of the total pup population is observed during a single survey. Initial analyses using simulated data demonstrate the model’s ability to recover parameters (i.e. 95% CIs cover the true parameter values) under diverse scenarios, including varying timing, number of surveys and population sizes. Future data integration from aerial, drone, and ground surveys will inform estimation of key observation parameters. Further developments include applying a hierarchical framework to the long-term data set across 60 regularly monitored colonies. This will allow the model to provide uncertainty estimates across regions that will improve the overall grey seal population model. This project also seeks to provide insights into how the birth curve (i.e. the distribution of births throughout the breeding season) is shifting across different colonies, over time and what potential environmental or anthropogenic impacts influence breeding phenology and pup production. This approach has the potential to significantly improve estimation compared to current methods, directly informing the overall grey seal population dynamics model and more effective management strategies..
Authors: Saskia Schirmer (University of Greifswald)+; Alexander Scheuerlein (University of Greifswald); Stefan Mayr (University of Greifswald); Gerald Kerth (University of Greifswald); Marcus Fritze (Competence Center for Bat Conservation Saxony-Anhalt).
Abstract: When individuals of a species cluster together in space, estimating their population size and trend may challenge existing population trend estimation methods in a different way than the already known problems. In a simulation study we generate population count data with an additive trend consisting of differing local and a common global part and show how this affects the trend estimate of established methods like TRIM and poptrend. We suggest that by incorporating hierarchical structures in GAMs or GLMs it becomes possible to overcome the difficulties created by clustering individuals. In a case study on annual count data of hibernating bats, we apply the method to real-world data and show the need of such a method when obtaining nationwide trends of bat species. The method is implemented on the website batlas.info to facilitate executing, visualization and further use of nationwide population trends of bat species in Germany. The project aims to bring together the data of volunteer bat conservationists and the knowledge of professional bat scientists and statisticians as well as politicians in order to improve the conservation of bat species. The project is funded by the German Federal Agency for Nature Conservation with funds of the Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection..
Authors: Janelle Badger (National Oceanic and Atmospheric Administration)+; Jennifer McCullough (National Oceanic and Atmospheric Administration); Jay Barlow (Cascadia Scientific Services); Robin Baird (Cascadia Research Collective); Taiki Sakai (National Oceanic and Atmospheric Administration); Erin Oleson (National Oceanic and Atmospheric Administration).
Abstract: Passive acoustic data collection on cetaceans has flourished in recent decades, proving particularly useful in remote regions where regular visual surveys are infeasible. However, there is a disconnect in translating these data sources to management needs, with limited methods available to analyze density and abundance data for various platforms and species. Here, we use echolocation pulses of Blainville’s and Cuvier’s beaked whales from drifting recorder buoys deployed around the Mariana Archipelago to estimate animal density using a Bayesian group-based point-transect analysis. We estimated the detection function using a maximum simulated likelihood approach fit to the observed distribution of signal arrival angles; the acoustic availability of whales during their dive cycle was estimated from the duration of acoustic encounters using a mark-recapture approach. This method relies upon prior information obtained from visual observations (group sizes) and telemetry tag deployments (dive cycle durations, echolocation depths); however, no such data are available from the Mariana Archipelago. We thus conducted a meta-analysis of available diving behavior data from 133 telemetry-tagged whales in seven locations to partition variance due to location, individuals, and tag types for each species in order to construct informative prior distributions for these parameters and propagate our uncertainty to density estimates. From 21 buoy deployments, Cuvier’s and Blainville’s beaked whales were detected 335 and 395 times, respectively, resulting in densities of 8.72 (95% CRI [5.21, 13.0]) and 28.5 (95% CRI [17.8, 41.0]) animals per 1000 km2, respectively. The study area was defined by a minimum convex polygon that covered all survey effort plus a 50 km buffer. The resulting estimated abundance of our focal species in the study area was 6,001 (95% credible interval (CRI): [3,904; 8,605]) Cuvier’s beaked whales and 15,667 (95% CRI: [10,144; 22,096]) Blainville’s beaked whales. Sensitivity analyses show that this method is not sensitive to non-random sampling due to entrainment in oceanic features, and that duty cycling does not introduce substantial bias in detection rates. This acoustic-based density estimation framework allows for the first abundance estimates for these remote, cryptic populations of large marine predators..
Authors: Amira Sharief (Zoological survey of India)+; Dr.vineet kumar (Zoological survey of India); hemant singh (Zoological survey of India); Dr.Bheem Dutt Joshi (Zoological survey of India); Dr.Mukesh Thakur (Zoological survey of India); Ritam Dutta (Zoological survey of India); Catherine Graham (WSL); Dr.lalit kumar sharma (Zoological survey of India).
Abstract: In the Himalayan region habitat loss, fragmentation and degradation through conversion of land use, agriculture expansion, linear developments, industrialization and overexploitation of natural resources are amongst the major threats faced by. In addition, the planned drivers of degradation, including the development of road, rail network, etc., may degrade existing natural biological corridors. The long-term persistence of species in fragmented landscapes depends on the successful movement of individuals between habitat patches. Over the years Protected areas (PAs) has increased as a way to safeguard nature but PAs cannot serve mankind alone. Therefore, conservationists now view PAs as a landscape context where conserving connectivity among PAs is a new and growing challenge. Protected areas form the cornerstone of conservation initiatives, serving as essential foundations for safeguarding biodiversity, promoting ecosystem well-being, and delivering a host of critical ecosystem services. Hence, we made efforts to map such vital biological corridors between protected areas which will facilitate geneflow for safeguarding the wildlife habitats during developmental activity planning. The pilot study conducted in Himachal Pradesh, India aimed to identify biodiversity corridors between protected areas using various survey methods such as sign survey and camera trapping. Ensemble modeling was employed for multispecies distribution modeling, and GIS tools such as Circuitscape, Core Mapper, and Linkage Mapper were used to map connectivity between the PAs. Results indicated high biodiversity regions between certain protected areas. Only a small portion of the landscape was found suitable for species habitat, with factors like maximum temperature of warmest quarter and elevation influencing distribution. Four high connectivity blocks were identified, with the highest connectivity between Inderkilla NP and Kais WLS. Core habitat patches were identified, and least-cost paths were mapped between them. The least-cost path analysis enabled us to map the optimal movement networks between the existing populations of biodiversity in the study region. We also highlighted cost-weighted distance which suggests the quality of each linkage and gives the average resistance encountered along the optimal path between the core areas. Pinch points indicating movement constriction were found, emphasizing the importance of certain linkages for network connectivity, particularly between Inderkilla NP and Kais WLS, and within Dhauladar Wildlife Sanctuary. Connectivity mapping can assist managers and policy makers to develop strategic plans for balancing wildlife conservation and other land uses in the landscape. This landscape has potential for sustaining biodiversity, with careful planning and tools that bring transparency to the process, such as these connectivity maps, we believe that pragmatic conservation mechanisms can be devised to sustain the functionality of this landscape for long-term biodiversity conservation.
.
Authors: Sierra Gillman (University of Washington)+; Scott Pearson (Washington Department of Fish and Wildlife); Beth Gardner (University of Washington).
Abstract: The Marbled Murrelet (Brachyramphus marmoratus) is a small seabird that ranges along the Pacific coast from Alaska to California, USA. Unlike other alcids that breed in dense colonies on the ground or in burrows, Marbled Murrelets nest in dispersed old-growth forests. This unique breeding strategy leaves them vulnerable to anthropogenic disturbances such as commercial logging. Currently, Marbled Murrelets are listed as threatened under the United States federal Endangered Species Act. Therefore, the conservation status of the Marbled Murrelet requires a robust analysis of environmental drivers of population trends. The population within the contiguous United States has been monitored through a structured designed based distance sampling approach since 2002. We developed an innovative spatio-temporal hierarchical distance sampling model to estimate long-term population trends of Marbled Murrelets across Washington, Oregon, and California. Our methodology leverages longitudinal survey data to enhance the accuracy of population trend estimates by accounting for imperfect detection, spatial and temporal structuring in abundance, and habitat-specific effects on ecological processes. The results of our study showed a more nuanced understanding of how environmental factors impact Marbled Murrelet populations across space and time, including human disturbances. Given the large scale of the data collection, our model’s capacity to account for subpopulation dynamics within the larger population context, allows for more fine-scale estimation of population trends, which is essential for informing targeted conservation strategies and policy-making to mitigate the species’ decline..
Authors: Saman Muthukumarana (University of Manitoba)+.
Session: S-5-4.
Where: G029.
When: 10:30-12:30.
Abstract: Acoustic telemetry systems technology is useful for studying fish movement patterns and habitat use. However, the data generated from omnidirectional acoustic receivers are prone to large observation errors since the tagged animal can be anywhere in the detection range of the receiver. In this study, we develop a Bayesian state-space modeling (BSSM) approach and different smoothing methods, including kernel smoothing and cross-validated local polynomial regression to reconstruct fish movement paths using data obtained from a telemetry receiver grid in Lake Winnipeg. As a part of the Lake Winnipeg Basin Fish Movement Project, Walleyes (Sander vitreus) were tagged and a grid of acoustic receivers was deployed in Lake Winnipeg and some of its major tributaries to detect the movements of tagged fish. From June 2017 to September 2018, the receivers recorded over 3.8 million detections.
To achieve our objective, we develop a BSSM and compare against three popular smoothing approaches (simple weighted average method, kernel smoothing with the Gaussian kernel, and cross-validated local polynomial regression approach) to reconstruct fish movement paths. Furthermore, we performed a simulation study to evaluate the performance of the different modelling approaches in estimating the movement paths. Using the BSSM approach, we obtain more realistic movement paths compared to the smoothing methods. We then use the estimated fish movement paths and fishing information from Lake Winnipeg, such as the amount of landings and quota assignments to investigate the potential interactions between fish movements and fishing activities in the lake. The newly developed fishery metrics include the probability of fish presence, probability of encounter and potential fishing pressure, which are useful in making effective fishery management decisions on Lake Winnipeg over the space and time. This is based on an ongoing collaborative project with the Department Fisheries and Oceans Canada..
Authors: Virginia Morera-Pujol (University College Dublin)+; Damien Barrett (Department of Agriculture, Food, and the Marine); Andrew W. Byrne (Department of Agriculture, Food, and the Marine); Guy McGrath (University College Dublin); David J. Quinn (Department of Agriculture, Food, and the Marine); Simone Ciuti (University College Dublin).
Session: S-4-3.
Where: G043.
When: 14:30-16:30.
Abstract: The European badger (Meles meles) is an important keystone species in Irish ecosystems, but it is also a reservoir and vector of bovine tuberculosis (bTb). Therefore, badgers find themselves in a paradoxical situation in Ireland sitting between two conflicting conservation and management policies: on one hand, they are protected by Irish and European laws, but on the other hand they are subject to a culling programme within the framework of the bTb eradication strategy. Additionally, since 2019 there is an attempt to move towards a vaccination-and-release programme, but the effects of this switch on badger populations, or on their role in the bTb episystem are still unknown. In this context, it is essential to know their distribution, abundance, and habitat preferences. We used point process models within the Integrated Nested Laplace Approximation (INLA) framework to model the distribution and habitat preferences of badger setts (burrows) and badgers themselves, using data collected through both the culling and vaccination programmes on location of both setts and badgers. We modelled non-linear relationships with environmental covariates such as elevation, slope, and aspect, and habitat presence based on the land use layer from the Copernicus European remote sensing server (CORINE), and used a spatial random field modelled as a Gaussian Markov Random Field through a Stochastic Partial Differential Equation (SPDE) model. We obtained distribution and abundance estimates for both setts and number of badgers, which are already a very useful tool for management of badger populations themselves. Combined with information about the badgers’ health and body condition, they can be linked to bTb relative risk spatial estimates to link badger occurrences to bTb breakouts in farms. In addition, the estimates of sett and badger abundance can help understand the effect that replacing the culling programme with a vaccination programme can have on badger populations..
Authors: Sara Stoudt (Bucknell University)+; Laura Melissa Guzman (University of Southern California); Benjamin R Goldstein (North Carolina State University); Jayme Lewthwaite (University of Southern California); Vaughn Shirey (University of Southern California).
Session: S-6-4.
Where: G037.
When: 10:30-12:30.
Abstract: What counts as a checklist for the purposes of occupancy models? Recent developments in occupancy models applied to presence-only data (typically coming from community sourced data or historic museum records) suggests that non-detections can be inferred from other species collected during the same visit. There has been ample discourse about how to delineate a visit in terms of the temporal extent of the search effort and the spatial range of each site. However, less has been discussed about the role of taxonomic scope. For example, if a data collector is recording all bees encountered, then an observation for one species implies a non-detection of all other bee species. On the other hand, if a data collector is doing a more targeted search than is expected by the modeler, those inferred absences might not actually represent non-detections. With presence-only data, modelers must choose, for a given target species, which non-target species’ detections should stand in to represent effort for the target species: should all records on the dataset be included, or just records of sufficiently similar species? Choosing a taxonomic scope when inferring effort in these data sources is especially challenging because we do not have explicit information about the users search preferences. How do assumptions about the taxonomic level that a data collector is actively searching for impact the information content of an assumed checklist, and what are the implications for modeling? In this work we show how different choices in the pre-processing of presence-only data, at the taxonomic level, can impact model estimates in a real-world case study (North American butterflies), explain drivers of these differences in estimates (for example, relative differences in user engagement between species), and provide a preliminary strategy for correcting for these differences. .
Authors: Sam Mason (UNSW)+.
Session: S-2-3.
Where: G049.
When: 15:30-16:30.
Abstract: Detecting species response to climate change is a critical concern in ecology requiring relevant climatic predictors over long temporal windows and large spatial extents. To date this has been technologically challenging resulting in the widespread use of 30 year averaged datasets which, while readily accessible, actually mask species responses by effectively treating the climate as temporally static.
Recent advances in data acquisition technology make it easy to obtain a wide range of environmental predictors at appropriate spatio-temporal scales and at consistent resolutions needed for effective research.
By using dynamic predictors and anomalies from their mean values in spatio-temporal species distribution models we show that we can detect species response to climate change and that this yields better predictive performance than current practice..
Authors: Becky Tang (Middlebury College)+.
Session: S-6-1.
Where: G043.
When: 10:30-12:30.
Abstract: A frequent challenge encountered in real-world applications is data having a high proportion of zeros. Focusing on ecological abundance data, much attention has been given to zero-inflated count data. Models for non-negative continuous abundance data with an excess of zeros are rarely discussed, even though such data are consistently collected in ecology. Work presented here considers the creation of statistical models that attempt to fill this gap. We consider a model that introduces a point mass at zero through both a left- censoring approach and through a hurdle approach. We incorporate both mechanisms to capture the analogue of zero-inflation for count data, where zeros due to suitability and zeros due to chance are both possible but arise from different sources. Additionally, primary attention has been given to univariate zero-inflated modeling (e.g., single species), whereas data often arise jointly (e.g., a community or collection of species). With multivariate abundance data, a key issue is to capture dependence among the species at a site, both in terms of positive abundance as well as absence. Therefore, we also propose a model for multivariate zero-inflated continuous data that are non-negative. Working in a Bayesian framework, we discuss the issue of separating the two sources of zeros and offer model comparison metrics for multivariate zero-inflated data. In an application, we model the total biomass for five tree species obtained from plots established in the Forest Inventory Analysis database in the Northeast region of the United States. We find that modeling the species jointly and accounting for dependence in both presence and absence can lead to superior predictions than when modeling species individually..
Authors: Erik Kusch (University of Oslo Natural History Museum)+.
Session: S-7-1.
Where: G049.
When: 14:30-16:30.
Abstract: While ecosystem projections have previously been made using ecological network approaches, mainly focusing on extinction cascade directionality as a driving force of ecological change, only a few studies have explicitly considered resilience mechanisms of ecological networks towards species extinctions.
These mechanisms are link-loss sensitivity (i.e., the inverse of the capacity to withstand loss of links when interaction partners go extinct) and realisation of rewiring potential (i.e., the capacity to re-allocate lost interaction potential/links to novel or already established interaction partners). We have developed a quantitative framework with which to do so and subsequently established simulated ecosystem projections following species-specific climate safety margin-driven extinction cascades that consider both resilience mechanisms as well as cascade directionality across an array of globally-distributed in-situ sampled mutualistic networks.
To identify the effects of network resilience mechanisms on loss of biodiversity and change in connectedness, our approach explicitly explores two-dimensional resilience landscapes defined by link loss sensitivity and realisation of rewiring potential. In addition to the traditional focus on bottom-up and top-down extinction cascades, we also explore the implications of bi-directional extinction cascades.
Our results indicate that neglecting to account for ecological network resilience mechanisms may lead to severe underpredictions of ecological change measured as loss of biodiversity and change in connectedness. We also find that increasing link-loss sensitivity results in far greater numbers of secondary extinctions, while increases in rewiring potential abate secondary extinctions. Furthermore, we find that the contemporary focus on differences between bottom-up and top-down outcomes may obscure the considerably larger risk of biodiversity loss posed through bidirectional extinction cascades..
Authors: Erik Kusch (University of Oslo Natural History Museum)+.
Session: S-5-1.
Where: G037.
When: 10:30-12:30.
Abstract: Rendering ecological networks is vital to assess ecosystem resilience to biodiversity loss and to predict community assemblies. The labour-intensive sampling requirements of ecological networks have spurred the creation of network inference methodology. Recent research has identified inconsistencies in networks inferred using different approaches thus necessitating quantification of inference performance to facilitate choice of network inference approach.
We have developed a data simulation method which generates data products fit for network inference. The simulation framework we present here can be parameterised using real-world data. (e.g., biological interactions observed in-situ and bioclimatic niche preferences), supports directed and undirected links in ecological networks as well as exporting of time-series or spatial products fit for currently available ecological network inference approaches. Next, we developed novel analysis procedures with which to explore inference and detection probabilities of association types of different identity and sign with respect to bioclimatic niche preferences and association strength between species. Consequently, we present a workflow for quantification of network inference reliability.
Applying our workflow to two well-established ecological interaction network inference methods, we identify a concerningly large range in accuracy of inferred networks confirming that choice of network inference approach is a non-trivial decision. These differences in inference accuracy are governed by a paradigm of input data types and environmental parameter estimation as previously suggested with network inference approaches accounting for environmental gradients and leveraging more nuanced biological inputs outperforming simpler approaches in inference accuracy..
Authors: David Warton (UNSW Sydney)+; Victor Tsang (UNSW Sydney).
Session: S-5-4.
Where: G029.
When: 10:30-12:30.
Abstract: An important problem in palaeoecology is estimating the extinction or invasion time of a species from the fossil record – whether because this is of interest in and of itself, or in order to understand the causes of extinctions and invasions, for which we need to know when they actually happened. There are two main sources of error to contend with - sampling error (because the last time you say a species need not be the last time it was there) and measurement error (dating specimens, usually of known magnitude). The paleobiology literature typically ignores one or other of these sources of error, leading to bias and underestimation of uncertainty to an extent that is often qualitatively important.
The problem is surprisingly difficult to address statistically, because while standard regularity conditions are technically satisfied, we are typically close to a boundary where they break down, and hence standard asymptotic approaches to inference perform poorly in practice. We propose using a novel method, which we call regression inversion, for exact inference, and we apply this technique to a compound uniform-truncated t (CUTT) model for fossil data. We show via simulation that this approach leads to unbiased estimators, and accurate interval inference, in contrast to its competitors. We show how to check the CUTT assumption visually, and provide software to apply all of the above in the reginv package..
Authors: David K. E. Chan (The University of New South Wales)+; Janice Seo (The University of Auckland); Ben Stevenson (The University of Auckland).
Session: S-3-3.
Where: G049.
When: 10:30-12:30.
Abstract: Animal density estimators from spatial capture-recapture (SCR) are commonly advertised as being remarkably robust to model misspecifcations, particularly to the choice of detection function. The literature since its inception supports this with sampling designs involving many detectors distributed across the study area. However, the typical number of detectors used in an acoustic survey for density estimation is noticeably smaller than other sampling methods, such as camera traps and DNA sampling surveys. Currently, there is no known guidance on whether the density estimators from SCR remain robust to the choice of detection function for a typical acoustic survey design. Our work looks at this particular scenario via simulation studies, focusing on whether model selection methods, such as AIC, assist with selecting a detection function that results in more robust density estimates in practice..
Authors: Ben J Maslen (University of New South Wales)+.
Session: S-3-1.
Where: G029.
When: 10:30-12:30.
Abstract: The rise of AI has seen an explosion in the use of deep learning methods in the automated analysis of underwater monitoring videos and camera traps, saving ecologists vast amounts of time and resources. Ecological imagery poses unique challenges however, with cryptic species struggling to be detected amongst poor visibility and diverse environments.
Through my talk, I propose leveraging freely available movement information, to attempt to improve the detections produced by the state-of-the-art object detection algorithm; YOLOv8. A wide range of methods that utilise movement information are trialled on over 35,000 annotated images sourced from over 3,000 terrestrial and marine locations, the most extensive investigation to date.
We found that leveraging movement patterns becomes more useful for smaller sized studies, however is not needed for well annotated studies (>1000 annotations per class). Out of the methods that utilise movement, we found that a simple ‘differencing’ of neighbouring frames generally performed the best, whilst attempting to track taxa to boost prediction scores performed poorly.
Other studies in this area tend to only focus on 1-2 datasets and a single method that utilises movement information, making it difficult for ecologists to generalise results. My talk provides key lessons for ecologists to determine the usefulness of incorporating methods that measure an animal’s movement patterns into their object detection pipeline. .
Authors: Olivier Gimenez (CNRS)+.
Session: S-8-3.
Where: G037.
When: 11:00-12:45.
Abstract: Streams and rivers are biodiversity hotspots that provide valuable ecosystem services to human populations. To monitor this biodiversity, we need to quantify the distribution of animal and plant species, and better understand the relationships between species and their environment. To do so, occupancy models are Species Distribution Models that are based on repeated visits of spatial sampling units. Importantly, these models allow the distinction between missing a species in the field despite its actual presence and its true absence. Recently, several extensions of occupancy models have been proposed to account for spatial autocorrelation in the occupancy probability through the Euclidean distance between the spatial sampling units. However, these models are not suited for streams and rivers because of their spatial structure in networks which requires a particular treatment. Here I propose spatial stream network occupancy models that allow spatial autocorrelation in occupancy probabilities on stream and river segments. Specifically, I use the stream distance defined as the shortest distance between two locations computed along the stream network. I first present the theoretical developments of the model. Then I conduct a simulation study to assess bias and precision in parameters estimates, and demonstrate flawed inference when spatial autocorrelation is modeled incorrectly or simply ignored. Finally, I illustrate the approach on a semi-aquatic mammal in French streams and rivers. Overall, spatial stream network occupancy models provide a formal approach to assess biodiversity in streams and rivers. .
Authors: Neil Gilbert (Michigan State University)+; Graziella V DiRenzo (U. S. Geological Survey); Elise Zipkin (Michigan State University).
Session: S-5-2.
Where: G049.
When: 10:30-12:30.
Abstract: Capture–recapture is one of the oldest analyses in statistical ecology. Historically, applications of capture–recapture focused on estimating abundance of a single species at individual study sites. However, coordinated monitoring programs are making these data increasingly available across multiple locations for multiple species. Therefore, we developed a multi-species, spatially stratified capture–recapture model to estimate species abundances from monitoring programs in which multiple species are sampled at multiple sites. The model uses random-effects structures to explain variation in abundance and detectability among species and sites. Moreover, the model constrains abundance estimates for a given species to be zero beyond the limits of its geographic range, since study sites may sample different regional species pools. We apply the model to small mammal trapping data collected by the National Ecological Observatory Network, a recently launched, government-funded ecological monitoring program in the United States. The goal of our application was to evaluate the relative influences of local host biodiversity versus host metacommunity diversity on disease risk. We used abundance estimates from the model to compute—with uncertainty—biodiversity metrics, such as species richness and Shannon diversity, for replicate plots within sites (local community) and aggregated across plots within sites (metacommunity). We subsequently used these biodiversity metrics as predictor variables in analyses to understand variation in Lyme disease infection prevalence in small mammals. Small mammal species richness and Shannon diversity showed stronger relationships with infection prevalence when measured at the metacommunity level versus the local community level. Our modeling framework expands the canon of capture–recapture to accommodate data from multispecies, broad-scale monitoring programs such as bird ringing programs and the National Ecological Observatory Network..
Authors: Keiichi Fukaya (Institute for Environmental Studies)+; Yuta Hasebe (Kanagawa Environmental Research Center).
Session: S-4-4.
Where: G037.
When: 14:30-16:30.
Abstract: Environmental DNA (eDNA) metabarcoding is a method of high-throughput sequencing of environmental samples to detect DNA sequences of specific taxa to assess species diversity. Although eDNA metabarcoding can detect species with high sensitivity, species detectability with this method can be variable. For reliable inference of species diversity and effective study design, methods and tools to understand the imperfect detection of species that can occur throughput the multi-step workflow of eDNA metabarcoding are essential.
A previous study proposed a class of multispecies site occupancy models for eDNA metabarcoding that fits sequence read count data, a data format unique to high-throughput sequencing. The talk will present the development of an R package occumb for statistical inference of this class of multispecies occupancy models. The occumb package provides functionalities to specify a model using R’s formula syntax and fit the model via the Markov Chain Monte Carlo method implemented in JAGS. It also allows users to assess the goodness of fit of the model and compare the effectiveness of species detection among different possible study settings in a straightforward manner. A case study of eDNA metabarcoding of riverine aquatic insects will illustrate the application of the package.
.
Authors: Hannah Correia (Johns Hopkins University)+; Paul Ferraro (Johns Hopkins University); Laura Dee (University of Colorado Boulder).
Session: S-5-4.
Where: G029.
When: 10:30-12:30.
Abstract: Ecologists seek to understand the intermediary processes through which changes in one component of an ecological system affects other components. However, quantifying the causal effects of these mediating processes in ecological systems is challenging, even in experimental settings. Researchers must define what they mean by a “mediated effect”, determine what assumptions are required to obtain unbiased causal estimates of the mediating component’s effect, and assess whether these assumptions are credible for a study. To overcome some of these challenges, ecologists can take advantage of significant advancements in causal mediation analysis that have been made in other fields over the past three decades. We use a hypothetical ecological study to describe how common designs in ecology for detecting or quantifying mediation effects may have biases. We then show how these biases can be addressed through alternative experimental or statistical designs and describe the various causal assumptions each design relies on to make causal claims about the magnitudes of mediator effects. We also discuss when these causal assumptions may be violated for ecological studies and review procedures to assess the sensitivity of a study’s results to potential violations of causal assumptions. The approaches we discuss will allow ecological researchers to clearly communicate the causal assumptions necessary for valid inferences and examine potential violations to these assumptions to enable rigorous and reproducible explanations of important intermediary processes in ecology..
Authors: Blanca Sarzo (Foundation for the Promotion of Health and Biomedical Research of Valencian Region, FISABIO); Ruth King (University of Edinburgh)+; Rachel McCrea (Lancaster University).
Session: S-5-2.
Where: G049.
When: 10:30-12:30.
Abstract: We consider individual heterogeneity Cormack-Jolly-Seber models for open population capture-recapture data. In particular we focus on models with a continuous (or infinite mixture) individual random effect component specified on the survival probabilities to reflect inherent (unobserved) individual heterogeneity. However, if individuals are initially observed at different ages we will observe a survivorship bias (i.e. a selection bias). This survivorship bias arises since individuals are observed conditional on having survived until their initial capture, and hence implicitly on having survived until their given age at initial capture. Thus it follows that individuals that are initially observed at an older age are likely to be stronger individuals, as weaker individuals are more likely to have died by this given age and so be unavailable for capture. We practically demonstrate the implications of failing to account for such survivorship bias via a simulation study, including biased estimates of the survival probabilities. Further, we describe how we can formulate a consistent model that corrects for the survivorship bias in the observed data, and demonstrate its improved performance for the simulated data. We conclude with an application to guillemot data, where within the study period birds are of known different ages at initial capture and demonstrate the importance of accounting for survivorship bias for this real study population..
Authors: Scott W Forrest (Queensland University of Technology)+; Daniel E Pagendam (CSIRO Data61); Chris Drovandi (Queensland University of Technology); Michael Bode (Queensland University of Technology); Jonathan Potts (University of Sheffield); Maryam Goldchin (CSIRO Health and Biosecurity); Andrew Hoskins (CSIRO Environment).
Session: S-8-1.
Where: G043.
When: 11:00-12:45.
Abstract: Predicting animal movement is difficult due to the fine-scale and complex decisions made by animals as they move. A common approach to infer the relationships between animals and their external environment is a step selection analysis (SSA). SSA approaches have recently been used to generate animal trajectories for prediction, although the SSA model fitting framework is limited in several ways, namely that regression is performed on point locations that ignore the structure and composition of habitat features, and it is prohibitively difficult to fit and interpret models that include interacting components such as multi-scale temporal dynamics. Fortunately, the goal of SSA model fitting is simply to determine the next step probability given the surrounding environment, and deep learning is aptly suited to this task. This study will train and compare between neural network architectures that receive scalar values and multiple habitat layers as inputs, and output a single layer representing the next-step probability. This is equivalent to SSA model fitting, but benefits offered by deep learning include the ability to learn features that are present in the habitat covariate layers, such as linear features (rivers, forest edges), and the composition or size of certain habitat areas. It can also represent the complex and seemingly abstract interactions between habitat covariates, time of day or year, and memory and social dynamic processes. Additionally, there is promising potential for integrating non-spatial data sources such as accelerometers and physiological sensors. We expect that the deep learning approach will lead to generating more accurate animal movement trajectories.Our research is motivated by the need to accurately predict invasive water buffalo and cattle locations in Northern Australia, both of which cause significant environmental damage and represent economic opportunities for traditional landowners. The data collection consists of 916 GPS devices..
Authors: Baptiste Alglave (Université Bretagne Sud)+; Benjamin Dufée (Université Bretagne Sud); Said Obakrim (University of Rennes 1); James Thorson (NOAA).
Session: S-4-3.
Where: G043.
When: 14:30-16:30.
Abstract: Spatio-temporal data are ubiquitous in ecology. These can be available over long time series and dimension reduction (a.k.a. “ordination”) methods are necessary to summarize, visualize, and communicate the information contained in the series of maps. Empirical Orthogonal Functions (EOF) is the basic method for dimension reduction of spatio-temporal data. It basically consists in realizing a Principal Component Analysis on spatio-temporal data. After many applications in meteorology and climate, these methods have been extended to noisy, zero-inflated, and multivariate data arising in ecological applications. Still, questions remains to identify which representation of spatio-temporal data is the best. EOF provides orthogonal patterns that capture the variance of the signal and these constraints might not be the best to represent the data. In this presentation, I will dig in the different alternative of EOF to represent spatio-temporal data. I present three alternative methods: (1) EOF where orthogonal constraints are relaxed, (2) EOF that are constrained by ancillary variables and (3) EOF where orthogonality constraints are modified so that plans are “spatially orthogonal”. To illustrate each extension of EOF, I apply the related methods on two distinct data sets: (1) plankton distribution from a biogeochemical model and (2) fish spatio-temporal distribution from a statistical model. Finally, I point out the limits of these methods and highlight future research avenues for this type of statistical approach..
Authors: Jan-Ole Koslik (Bielefeld University)+; Carlina C Feldmann (Bielefeld University); Sina Mews (Bielefeld University); Rouven Michels (Bielefeld University); Roland Langrock (Bielefeld University).
Session: S-6-2.
Where: G049.
When: 10:30-12:30.
Abstract: Over the last decade, hidden Markov models (HMMs) have become increasingly popular in statistical ecology, where they constitute natural tools for studying animal behavior based on complex sensor data. Corresponding analyses sometimes explicitly focus on — and in any case need to take into account — periodic variation, for example by quantifying the activity distribution over the daily cycle or seasonal variation such as migratory behavior.
For HMMs including periodic components, we discuss important mathematical properties that allow for comprehensive statistical inference related to periodic variation, thereby also providing guidance for model building and model checking. Specifically, we explore the periodically varying unconditional state distribution, along with the time-varying and overall state dwell-time distributions — all of which are of key interest when the inferential focus lies on the dynamics of the state process.
To illustrate the utility of these techniques, we apply the novel inference and model-checking tools to analyze the diel movement patterns of an elephant from the Ivory Coast. Crucially, we demonstrate that the dwell-time distributions of periodically inhomogeneous HMMs can deviate substantially from a geometric distribution, thus compensating for biologically unrealistic consequences of the Markov property..
Authors: Ioannis Rotous (University of Kent); Alex Diana (University of Essex); Alessio Farcomeni (Tor Vergata University of Rome); Eleni Matechou (University of Kent)+.
Session: S-6-1.
Where: G043.
When: 10:30-12:30.
Abstract: Hidden Markov models (HMMs) offer a robust and efficient framework for analyzing time series data, modelling both the underlying latent state progression over time and the observation process conditional on the latent state. However, a critical challenge lies in determining the appropriate number of underlying states, often unknown in practice. In this paper, we employ a Bayesian framework, treating the number of states as a random variable and employing reversible jump Markov chain Monte Carlo to sample from the posterior distributions of all parameters, including the number of states. Additionally, we introduce repulsive priors for the state parameters in HMMs, and hence avoid overfitting issues and promote parsimonious models with dissimilar state components. We demonstrate our proposed framework on two ecological case studies: GPS tracking data on muskox in Antarctica and acoustic data on Cape gannets in South Africa, Our results demonstrate how our framework effectively explores the model space, defined by models with different latent state dimensions, while leading to latent states that are distinguished better and hence are more interpretable, enabling understanding and interpretation of complex dynamic systems..
Authors: Rebecca Wilks (Edinburgh University)+; Ruth King (University of Edinburgh); Stuart King (Edinburgh University); David Williams (University of Leeds); Murray Collins (Space Intelligence); Niall McCann (National Park Rescue); Mike Chase (Elephants Without Borders).
Session: S-3-1.
Where: G029.
When: 10:30-12:30.
Abstract: Monitoring population sizes is critical for effective and accurate conservation management. However, obtaining accurate estimates of animal populations that range over vast areas is challenging. Traditionally, aerial surveys are used to observe populations and obtain population estimates. However, large-scale aerial surveys require significant man-power, time and cost, from planning, to surveying and post-processing of data. Research in recent years has turned towards exploring automated surveys, with Unmanned Aerial Vehicles (UAVs) performing the imaging, and bespoke AI algorithms used to automatically detect and count the size of animal populations.
Data from automated AI algorithms provide a range of statistical challenges including how to deal with the inevitable false negatives and false positives of the algorithm, corresponding to individuals not captured and non-individuals incorrectly recorded as individuals in the population of interest, respectively. Even the most performant AI will suffer from such limitations to some degree. Traditional abundance estimation models in statistical ecology are used to robustly account for false negatives (i.e. imperfect detection of individuals), particularly in relation to the use of capture-recapture-type approaches.
We develop a new modelling approach, applying the concepts of traditional capture-recapture methods to AI algorithms as observers to account for false negatives, before exploring further extensions to similarly account for false positives within the AI algorithm. We apply the novel approach to real single-capture data relating to elephants in southern Africa and demonstrate performance of the proposed methodology in terms of accuracy of estimating population counts..
Authors: Richard J Camp (U.S. Geological Survey)+; Len Thomas (University of St Andrews); David Miller (University of St Andrews); Steve Buckland (University of St Andrews); Steve Kendall (U.S. Fish and Wildlife Service).
Session: S-8-4.
Where: G029.
When: 11:00-12:45.
Abstract: Many wildlife monitoring programmes collect annual data on population abundance. The resulting abundance estimates fluctuate over time partly because of true population change and partly because of observation error. These two components of variation can be separated by fitting the estimates to a population dynamics model within a Bayesian state-space modelling framework. By constraining the population trajectory to be biologically realistic, more precise estimates can be obtained. Independent biological knowledge can be incorporated through choice of model structure and by specifying informative prior distributions on demographic parameters. We illustrate the approach using a 31-year point transect study of the Hawai‘i ‘ākepa (Loxops coccineus). We compiled demographic vital rates from the ‘ākepa literature. We initiated the population dynamic models as a SSM of the time series abundance estimates that included estimates of observation error and progressively added model complexity to include the demographic sub-processes population rate of change, recruitment and adult survival, thereby incorporating biological processes governing population changes. We fitted 5 models, each making different assumptions about how population change, recruitment and/or adult survival varied over time. Samples from the posterior parameter and state distributions were generated using Markov chain Monte Carlo procedures. Initial values were randomly selected from distributions based on demographic stochasticity from the literature. Overall, the ‘ākepa geometric mean growth rate was 1.02 indicating an increasing population over the 31-year time series, although there were periods of slow decline potentially associated with low recruitment and more rapid recovery associated with pulses of high recruitment. Abundance estimates derived from the population models were substantially more precise than the ``raw’’ point transect estimates: 95% CrI was on average 51.2% (SD=13.9%) narrower..
Authors: Amanda J Warlick (NOAA Fisheries AFSC)+; Brian Fadely (NOAA Fisheries AFSC); Peter Mahoney (NOAA Fisheries AFSC); Sharon Melin (NOAA Fisheries AFSC); Tom Gelatt (NOAA Fisheries AFSC); Kim Raum-Suryan (NOAA Fisheries AFSC); Sarah Converse (U.S. Geological Survey, WCFWRU, School of Aquatic and Fishery Sciences & School of Environmental and Forest Sciences, University of WA).
Session: S-1-4.
Where: G029.
When: 14:00-15:00.
Abstract: Effective monitoring is fundamental to estimating wildlife population parameters with a level of precision that is adequate to inform management decisions. However, trade-offs exist between survey effort, costs, and data quality. As such, evaluation of the expected performance of monitoring designs, prior to implementation, can facilitate identification of designs that require the least investment for a desired level of performance. In this study, we present a simulation framework for examining the precision of survival estimates and the probability of detecting a change in survival within the context of mark-resight monitoring programs. We examined model performance for monitoring survey designs that vary across size of the marked population, marking frequency, resight probability, and study duration. Our results showed that precision in age-specific survival estimates was most sensitive to marked cohort size, marking frequency, and study duration and less sensitive to resight probability. We evaluated these findings in the context of Steller sea lions (Eumetopias jubatus), where we found that historical mark-resight survey effort has been sufficient to reliably (>75% probability) achieve precision targets (coefficient of variation < 0.125) only for rookeries where abundance trends are stable or increasing. At rookeries where abundance trends are declining, however, the probability of achieving survival estimates with target levels of precision is low (<25%), being limited by smaller marked cohort sizes, lower marking frequency at remote rookeries, and fewer years of available data. Our findings highlight how the constraints of monitoring small, declining populations may limit the ability to detect changes in population dynamics on management-relevant time horizons. The framework presented here can be applied in a variety of contexts to assist natural resource managers in developing monitoring programs that efficiently meet monitoring objectives..
Authors: Fonya Irvine (Concordia University)+; Eric Pedersen (Concordia University).
Session: S-7-1.
Where: G049.
When: 14:30-16:30.
Abstract: The demographic rates of species are generally nonlinear functions of environmental drivers such as temperature, and understanding the shape of these relationships can help predict how species will respond to environmental change. While most studies focus on the responses of single species to these drivers, we expect closely related or functionally similar species to show similar responses to changes in environmental drivers. Currently, limited statistical tools are available for using information on trait or phylogenetic similarity to improve estimates of species-specific responses to environmental drivers.
We applied Hierarchical Generalized Additive Models (HGAMs) with Markov Random Field (MRF) smoothing terms to model simulated species responses to external drivers while accounting for shared functional traits or phylogenetic relatedness. HGAMs allowed us to model nonlinear relationships, while the MRFs smoothed differences between species based on functional traits or phylogenetic relatedness, enabling predictions for species with limited information. We applied the MRF-HGAM to evaluate whether species within a simulated community with similar functional traits or phylogenetic relatedness responded similarly to external pressures. .
Authors: Julien Gibaud (University of Monpellier)+.
Session: S-7-2.
Where: G043.
When: 14:30-16:30.
Abstract: In a context of component-based multivariate modeling we propose to model the residual dependence of the responses (here, the abundances of species). Each response of a response vector is assumed to depend, through a Generalized Linear Model, on a set of explanatory variables (here, environmental variables). Explanatory variables are partitioned into conceptually homogeneous variable groups, viewed as explanatory themes. Variables in themes are supposed many and redundant. Thus, generalized linear regression demands dimension reduction and regularization with respect to each theme. Regularization is performed searching each theme for an appropriate number of orthogonal components that both contribute to predict the responses and capture relevant structural information in themes. A small set of factors completes the model so as to model the covariance matrix of the linear predictors of the responses conditional on the components. This methodology is applied to an agricultural ecology dataset..
Authors: Chloé R Nater (Norwegian Institute for Nature Research)+; James Martin (University of Georgia); Erlend Nilsen (Norwegian Institute for Nature Research).
Session: S-8-4.
Where: G029.
When: 11:00-12:45.
Abstract: Quantifying temporal and spatial variation in animal population size and demography is a central theme in ecological research and important for directing management and policy. However, it requires field sampling at large spatial extents and over long periods of time, which is not only prohibitively costly but often politically untenable. Participatory monitoring programmes, particularly those that employ trained volunteers to collect data according to standardized protocols, can alleviate these constraints. Their usefulness is increased further by ongoing development of statistical models that are increasingly more powerful and able to make more efficient use of field data. Integrated population models (IPMs), for example, can use multiple streams of data from different field monitoring programmes and/or multiple aspects of single datasets to estimate population sizes and key vital rates. We developed a spatially explicit integrated distance sampling model (IDSM) and applied it to data from a large-scale participatory monitoring program to study spatio-temporal variation in population dynamics of willow ptarmigan (Lagopus lagopus) in Norway. In addition to providing robust spatially explicit abundance estimates, our model revealed that recruitment rates varied more across space than over time, while the opposite was the case for survival. Slower life history patterns (higher survival, lower recruitment) were more common at higher latitudes and altitudes and small rodent abundance had a positive effect on recruitment across sub-populations. Accurate estimation of covariate effects was only possible by integrating data from several monitoring areas for analysis. Last but not least, our study showcases how analysis workflows for IPMs can be set up in a reproducible and semi-automated way that increases their usefulness for informing management and regular reporting towards national and international biodiversity frameworks..
Authors: Thomas Neyens (Hasselt University & KU Leuven)+; Maxime Fajgenblat (KU Leuven).
Session: S-4-4.
Where: G037.
When: 14:30-16:30.
Abstract: Pitfall traps are buried containers set flush with the soil, in which invertebrates get trapped. Pitfall trapping provides a straightforward way to collect data on the occurrences and abundances of ground-dwelling invertebrate species, and it is therefore popular among citizen scientists interested in monitoring ground-dwelling invertebrate biodiversity. The traps are typically opportunistically placed for several days or weeks, after which the contents are collected and identified. Analysing pitfall trapping data can be challenging, especially when samples originate from multiple trapping periods or campaigns, as differences in sampling effort and seasonal patterns (i.e., the phenology of species) can strongly bias inference. We developed a Bayesian hierarchical model that circumvents these issues by incorporating a phenologically weighted measure of search effort. Specifically, we calculate this measure in a model-based fashion, by means of a latent phenological curve, as well as the starting and ending moment of each trapping event. We embedded this construct within a flexible joint species distribution model to comprehensively model ground-dwelling invertebrate communities through time and space, while simultaneously accounting for the data generating process. We applied our model to a large pitfall trapping dataset collected by citizen scientists, spanning almost 50 years and 10000 trapping events in the Belgian province of Limburg. By doing so, we uncovered spatio-temporal trends, phenological patterns and interspecific associations for over 500 spider species..
Authors: Ken B Newman (Biomathematics & Statistics Scotland)+; Lara Mitchell (US Fish and Wildlife Service); Leo Polansky (US Fish and Wildlife Service).
Session: S-7-3.
Where: G037.
When: 14:30-16:30.
Abstract: Scientific understanding and management of biological populations often requires estimates of population abundance and such estimates are calculated using samples drawn from the population. However, sampling can harm or harass populations, which is a particular concern for endangered species. We present a sequential adaptive sampling design focused on making population level inferences while limiting harm to the target population. The design incorporates stopping rules such that multiple samples are collected at a site until one or more individuals from the target population are captured, conditional on the number of samples falling within a predetermined range. We present distributional results for the sampling design and discuss how the resulting data may be used to fit models to estimate density. We use theoretical analyses and simulations to evaluate inference of population parameters and reduction in catch under the stopping rule sampling design compared to fixed sampling designs. Density point estimates based on stopping rules could theoretically have bias. In practice, they were comparable to estimates based on fixed sampling designs but had less precision. In an application to a case study involving sampling an endangered fish population, we show the stopping rules reduced catch by 58% compared to a fixed sampling design with maximum possible effort..
Authors: Guillaume Blanchet (Université de Sherbrooke)+.
Session: S-6-4.
Where: G037.
When: 10:30-12:30.
Abstract: Citizen science data are an amazing resource because they harness the effort of a huge number of citizens interested in biodiversity. However, although citizen science efforts often gather huge amount of data, they only convey the spatial and temporal information of specific individuals seen by citizen. These data are known as presence-only data. Citizen science data are also almost exclusively gathered where people live and travel. As such, little to no information is available in remote areas, including northern regions of Canada and Russia or the Sahara Desert. In this work, we propose to adapt the log-Gaussian Cox Process, a modelling framework designed for presence-only data, for it to explicitly account for sampling effort. Also, we propose a novel technique that combines convex polygons and cumulative density curves to assess whether there is enough data for a species to be modelled. This is particularly important to assess whether enough data is available to build a model. To illustrate our new approach, we relied on the research grade iNaturalist bird data gathered in North America and compare the modelled obtained with the ones from eBird for over 600 bird species. The models from eBird use much richer data but are ensemble models, which have limited interpretability. With our approach, we were able to reproduce the prediction performance of eBird using iNaturalist data (which is much less comprehensive) with the additional advantage that our approach is much more interpretable..
Authors: Tara Cunningham (University of Edinburgh)+; Stuart King (University of Edinburgh); Ruth King (University of Edinburgh); Norman Ratcliffe (British Antarctic Survey); Peter Fretwell (British Antarctic Survey).
Session: S-2-2.
Where: G043.
When: 15:30-16:30.
Abstract: As a key indicator species, penguins provide critical insight into the health of the Southern Ocean ecosystem. Accurate population counts of colonies are therefore integral to both conservation management of the species, as well as wider Antarctic research. Furthermore, identifying the demographic distribution within a colony – such as Breeder, Adult Non-Breeder, and Chick populations, are required for effective predictive population modelling. At all stages error must be quantified to attain population models with constrained error.
The use of Unmanned Aerial Vehicles (UAVs) for monitoring penguin colonies has become common practice due to the speed of data capture relative to traditional manual ground counts. Colonies are often large and dense (often >2 individuals per m^2), and so analysing the volume of captured images would require hundreds of hours of manual labelling. Developing automated or semi-automated methods for counting individuals in UAV imagery is therefore a key area of research.
We present a novel methodology for obtaining counts of adult breeding king penguins from UAV survey data through construction of a Digital Elevation Model (DEM) and unsupervised clustering. The method is significantly less computationally expensive and data intensive than deep learning models for object detection and has a detection accuracy comparable to expert manual counts. We additionally show that the overall accuracy is significantly improved by including an automatic delineation of the breeding colony area within the analysis via the regular spatial structure of the Breeders. Furthermore, this method could assist in standardising UAV survey operating procedures to ensure maximum information is gained from resource-intensive fieldwork..
Authors: Rishika Chopara (University of Auckland)+; Ben Stevenson (University of Auckland); Rachel Fewster (University of Auckland).
Session: S-3-3.
Where: G049.
When: 10:30-12:30.
Abstract: Goodness-of-fit (GOF) testing is vital to statistical analysis, as it allows us to validate the reliability of any statistical inference we make. In many modern capture-recapture models, such as spatial capture-recapture, effective methods of assessing GOF remain unresolved.
A model’s deviance is used to assess GOF by comparing it against a Chi-squared distribution. However, in some situations (e.g. when dealing with sparse counts) the deviance does not have a Chi-squared distribution, even approximately, yielding such tests unusable. By deriving tractable expressions for the mean and variance of the deviance, we use a Gamma distribution to accurately approximate the true underlying distribution of the deviance. Using this approximation, we enhance the usability and power of GOF tests while retaining the familiarity and convenience of the deviance statistic.
Using a range of capture-recapture models for illustration, we show how our method can be used to accurately approximate the distribution of the deviance when dealing with various levels of data sparsity. With this approach, we aim to provide an accessible and effective GOF testing framework for complex modelling scenarios such as spatial capture-recapture..
Authors: Maxime Pierron (CNRS-Université de Strasbourg)+; Carlos Ruiz-Miranda (Universidade Estadual do Norte Fluminense); Cédric Sueur (Université de Strasbourg); Valéria Romano (Institut Méditerranéen de Biodiversité et d’Ecologie marine et continentale).
Session: S-7-2.
Where: G043.
When: 14:30-16:30.
Abstract: Yellow fever is an endemic disease of South America and Africa, posing health issues to humans and non-human primates. In Brazil, one of the largest yellow fever outbreaks ever occurred between 2016 and 2019, causing the death of thousands of primates. To spatially predict future yellow fever emergence, there is a need to understand how species involved in its transmission cycle would be affected by environmental changes. Our project uses a Species Distribution Modelling approach to create predictive scenarios of the vulnerability of Brazilian primate species to yellow fever that consider global changes (such as climate variability and land use change). This approach identifies high-risk areas for yellow fever transmission by mapping the overlap of predicted distributions of the disease with both hosts (primates) and vectors (sylvatic mosquitoes). A presence-background modelling was used to estimate the current distribution of Brazilian primates and mosquitoes relevant to yellow fever transmission. We also model the current distribution of yellow fever itself, using historical outbreak data and the previously estimated distribution of potential hosts and vectors as covariates. We finally model the projected distributions for each study species for 2060 and 2100 under different Shared Socioeconomic Pathways (SSPs 245, 370, and 585). Our integrated approach offers insights into the present and future spatial distribution of yellow fever risk across the different biomes of Brazil. By delving into the factors influencing yellow fever emergence and primate vulnerability, our research aims to guide conservation efforts and public health initiatives, especially in regions where yellow fever is re-emerging and threatening primate and human populations. This research underscores the importance of predictive modelling to mitigate emerging issues in global health and biodiversity conservation..
Authors: Hanna BACAVE (INRAE)+; Pierre-Olivier CHEPTOU (CNRS); Nathalie PEYRARD (INRAE).
Session: S-5-4.
Where: G029.
When: 10:30-12:30.
Abstract: In ecology, the study of population dynamics is an important source of information for biodiversity conservation. Among the mathematical approaches used to model population dynamics, Hidden Markov Models (HMM) are well adapted in the case where the species of interest is difficult to observe. For a broader application of HMM in ecology, two limits need to be overcome. First, in most cases, HMM are used to deal with detection errors. But, another important situation is when only some life stages of the population can be observed while the others remain hidden. Second, conservation efforts actually require knowledge at the level of metapopulation more than a single population. Therefore there is a need to extend the HMM framework to the case of several couples of hidden and observed sub-populations in interaction. Depending on the species under study (plant, fungus, animal) the interaction between populations can be from and to observable or hidden stages and this structure must be explicitly modeled. In this work, we propose a conceptual guide to modeling and estimating parameters involved in metapopulation dynamics with partially observable populations, using the framework of Partially Observable Dynamic Bayesian Networks (PO-DBN). We show that only 4 interaction structures are needed to describe the main metapopulation models. Based on biological examples, we show how to build the associated PO-DBN for each of these structures. Finally, we consider parameter estimation using the EM algorithm for these models. We establish for which structures implementing EM is straightforward and for which there are computational limits. In these cases, we discuss methods from approximate inference that can be used to overcome them. This study provides the practical foundations for modeling and estimating the dynamics of a metapopulation with partially observable populations. It points out the computational challenges that remain to be tackled for a practical use..
Authors: Rebecca Muller (University of Cape Town)+.
Session: S-6-4.
Where: G037.
When: 10:30-12:30.
Abstract: Phenological changes are one of the most well recognised responses of animals to climate change. The ability to detect phenological changes is often reliant on long-term datasets, which are scarcely found in the global south. Given that the adaptive capacity of species is highly variable, it is important to improve our understanding of how species in southern-temperate systems may be responding to climate change through shifts in their annual cycles. Citizen science projects, like bird Nest Record Schemes, offer valuable long-term data, although digitisation and data quality pose challenges to their extensive use in research. We investigate the suitability of the South African Nest Record Scheme for estimating lay dates and preliminary exploration of phenological shifts in four well-represented species. Firstly, this study explored the composition of nest records for each species, specifically the proportion of single- to multi-visit cards. Secondly, we explored the accuracy of single-visit cards in estimating lay dates compared to highly accurate multi-visit cards. Lastly, we compared various data combinations to test for possible phenological shifts. For all species, a high proportion of records consisted of a single visit, however, our analysis suggested that both single- and multi-visit records mostly produced similar lay date estimates. This indicates that single-visit records could be used in combination with multi-visit records for estimations of First Egg Dates. We then modelled FEDs from different data combinations and found a high degree of consistency in the results from these different models across all species. Overall, a mixed model approach using all records, with card identity as a random term, was the most efficient and sensitive to assess effect of environmental predictors. This study highlights the usefulness of a large, long-term dataset in being able to detect phenological shifts in a region which is understudied in this regard..
Authors: Fränzi Korner-Nievergelt (Swiss Ornithological Institute)+; Sebastian Dirren (Swiss Ornithological Institute); Anne-Cathérine Gutzwiller (Swiss Ornithological Institute); Carole Niffenegger (Swiss Ornithological Institute); Elisenda Peris Morente (Catalanian Institute of Ornithology); Claire Pernollet (Swi); Christian Schano (Swiss Ornithological Institute); Irmi Zwahlen (Swiss Ornithological Institute).
Session: S-2-1.
Where: G037.
When: 15:30-16:30.
Abstract: Survival at high elevations is challenging both for birds and for the plastic rings used to individually mark them. Due to the cold, the plastic rings can crack and fall off. Individuals that lost their plastic rings cannot be re-sighted. As a consequence, apparent survival would be underestimated if mark loss is not accounted for in the survival model. The combination of re-sighting and re-capture data enables the disentangling of the survival of the ring from the survival of the birds by a multi-state model. We identified among-brand variance in the loss rate of the rings, and we describe seasonal variation in apparent survival of a high-elevation specialist species, the White-winged snowfinch Montifringilla nivalis. For assessing the climate sensitivity of such specialised species, it is important to be able to get unbiased estimates of their vital rates..
Authors: Théo Michelot (Dalhousie University)+.
Session: S-3-2.
Where: G043.
When: 10:30-12:30.
Abstract: Statistical approaches to estimate space use from tracking data roughly fall into two categories: small-scale models of animals’ movement decisions, and large-scale models of spatial distributions. The movement decisions give rise, in the long term, to large-scale distributions, but this mechanism is usually ignored, and it has been difficult to describe distributions as emerging properties of habitat-driven movements. I will show that irreversible continuous-time processes with explicit stationary distributions are promising multiscale models for the movement of animals. In particular, the underdamped Langevin process is a model from physics that can describe the movement of a particle that is subject to persistence in speed and direction, and to forces of attraction similar to habitat selection. This model includes many widely-used movement models as special or limiting cases, e.g., the Ornstein-Uhlenbeck process and the continuous-time correlated random walk. This approach can be used to understand how animals’ distributions are affected by the environment (e.g., habitat fragmentation), and makes it possible to combine data sources that have previously been analysed separately. I will describe some interesting properties of the underdamped Langevin model, and describe its usefulness for wildlife ecology using examples..
Authors: Anthony Rafael Charsley (National Institute of Water and Atmospheric (NIWA) Research)+.
Session: S-5-3.
Where: G043.
When: 10:30-12:30.
Abstract: Biodiversity data are “structured” when collected using standardised surveys or “unstructured” when collected opportunistically. While structured data are preferred, unstructured data are generally more numerous. Species distribution models that leverage the strengths of both data types can outperform models relying on a single data type. Here, we demonstrate a novel spatio-temporal modelling approach for integrating structured data and unstructured presence-only data. This approach is implemented with the vector autoregressive spatio-temporal (VAST) modelling platform, which is increasingly being employed in statistical ecology worldwide. Data integration is achieved by generating pseudo-absences for the presence-only data and estimating spatially varying catchability for all data sources relative to the most reliable structured dataset in a spatio-temporal model. We investigated a freshwater application in the Taranaki region, New Zealand, and a marine application in the South Pacific Ocean. We explored the impacts of the generation method and the number of pseudo-absences on model predictions. In the freshwater application, models integrating presence-only data and spatially structured pseudo-absences had increased true positive rates but decreased true negative rates compared to a model with structured data only. In the marine application, models employing spatially structured rather than randomly generated pseudo-absences had slightly reduced uncertainty in predictions. The number of pseudo-absences generated, relative to the sample size of the presence-only data, impacted model convergence. Five times the number was preferred in the freshwater application if the pseudo-absences generated matched the presence-only data spatial bias. Ten times the number was preferred in the marine application. Our approach has many other potential applications that can improve the quality of resource assessments and management efforts..
Authors: Fabiola Iannarilli (Max Planck Institute of Animal Behavior)+; Brian Gerber (USGS, Colorado Cooperative Fish and Wildlife Research Unit, Colorado State University); John Erb (Minnesota Department of Natural Resources); Martin Wikelski (Max Planck Institute of Animal Behavior); John Fieberg (University of Minnesota).
Session: S-4-4.
Where: G037.
When: 14:30-16:30.
Abstract: Anthropogenic disturbances increasingly impact wild animals, influencing risk and resource availability across spatio-temporal scales. To adapt to these novel conditions, animals may modify their behavior, potentially influencing both species- and community-level dynamics. Estimates of diel activity patterns can provide important information on how species adapt to both anthropogenic and natural disturbances, elucidate mechanisms of species co-existence through temporal partitioning, and assess broader effects of diel shifts at the community and ecosystem levels. Camera traps are commonly used to study animal diel activity by recording species presence and detection over time at multiple sites. Activity patterns often vary from site to site, yet this variability is averaged over when applying widely used Kernel Density Estimators (KDEs) of diel activity. Ignoring this intrinsic source of heterogeneity may lead to biased estimates of uncertainty and misleading conclusions regarding the drivers of diel activity. Thus, there is a need for more flexible statistical approaches for estimating activity patterns and testing hypotheses regarding their biotic and abiotic drivers. We illustrate how trigonometric terms and cyclic cubic splines combined with hierarchical models (i.e., Generalized Linear Mixed Models, GLMMs) can provide a valuable alternative to KDEs. Like KDEs, these models accommodate circular data, but can also account for site-to-site and other sources of variability, correlation among repeated measures, and variable sampling effort. They can also more readily quantify and test hypotheses related to the effects of covariates on activity patterns. Through an empirical case study focused on assessing the impact of human presence on the diel activity patterns of wild species in Europe, we demonstrate the flexibility and enhanced modeling capabilities of hierarchical GLMMs for analyzing diel activity patterns in relation to hypothesized drivers. .
Authors: Leonardo Capitani (WSL (Swiss Federal Institute for Forest, Snow and Landscape Research) & Eawag (Swiss Federal Institute of Aquatic Science))+; Valentin Moser (WSL (Swiss Federal Institute for Forest, Snow and Landscape Research) & Eawag (Swiss Federal Institute of Aquatic Science)); Francesco Pomati (Eawag (Swiss Federal Institute of Aquatic Science)); Anita Risch (WSL (Swiss Federal Institute for Forest, Snow and Landscape Research)).
Session: S-1-3.
Where: G037.
When: 14:00-15:00.
Abstract: Dissolved organic carbon (DOC) is important in many aquatic processes such as bacterial activity and primary production. The DOC concentration in the water can be related to bacterial proliferation. Therefore, the control of DOC has been recognized as an important parameter in drinking water monitoring and distribution systems. Beaver damming activity may influence the DOC content in the water by altering the lateral and longitudinal stream connectivity. Considered a protected species in Europe, beaver populations are increasing with consequences to the agriculture sector and human settlements. However, to what extent beavers influence the DOC content of the water and which mechanisms are the main drivers of such potential alterations remain to be clarified. In this study, we hypothesized that beaver damming activities may influence the DOC content in the water by increasing the abundance of aquatic primary producers (phytoplankton, macrophytes) and enhancing the terrestrial litter subsidies into the stream. We first drew directed acyclic graphs (DAG) that encode all prior ecological knowledge about the DOC content in the water. Then we expressed the DAG as a continuous Bayesian network where the prior knowledge is combined with the likelihood of observed data in the form of posterior evidence supporting or falsifying our hypothesis. We found that beaver damming activity caused a change in DOC content by adding 0.05 to 0.25 mg/L DOC downstream. Aquatic primary producers such as phytoplankton and macrophytes increased the DOC content by 2% to 8 %. Terrestrial litter subsidies in the beaver pond increased the DOC content in the water by 3-5%. These results allow us to quantify and predict how beaver damming activity influences stream biogeochemistry with notable consequences on ecosystem functioning, water quality and human wellbeing..
Authors: Dilsad Dagtekin (Swiss Federal Institute of Aquatic Science and Technology (Eawag))+; Dilsad Dagtekin (Department of Evolutionary Biology and Environmental Studies, University of Zurich); Dominik Behr (Department of Evolutionary Biology and Environmental Studies, University of Zurich); Claudia Fichtel (Behavioral Ecology and Sociobiology Unit, German Primate Center, Leibniz Institute for Primate Research); Peter M. Kappeler (Behavioral Ecology and Sociobiology Unit, German Primate Center, Leibniz Institute for Primate Research; Department of Sociobiology/Anthropology, University of Göttingen); Arpat Ozgul (Department of Evolutionary Biology and Environmental Studies, University of Zurich).
Session: S-6-1.
Where: G043.
When: 10:30-12:30.
Abstract: Tropical ecosystems are characterized by alternating dry and wet seasons. Wildlife populations already adapted to this seasonality via seasonally varying survival rates. However, seasonal patterns of temperature and rainfall are changing under climate change, especially in the tropics, forcing species to adapt their seasonal survival strategies. Therefore, it is important to understand how the survival of tropical species responds to seasonality and recent changes therein. To do so, we studied gray mouse lemurs (Microcebus murinus) of Kirindy Forest in Madagascar using three decades of capture–recapture data. We conducted multistate mark-recapture models using a hierarchical Bayesian framework and focused on age and sex-specific survival in response to rain, temperature, and population density. Mouse lemurs had a higher mean survival probability during the dry season than during the wet season. While survival of all groups decreased with increasing rainfall during the wet season, maximum temperature during the dry season affected survival depending on the density-dependence. Males were more sensitive to temperature changes than females. These results highlight the pronounced impact of temperature fluctuations on seasonal survival relative to variation in rainfall, while also emphasizing the influence of population density. This study also underscores the importance of accounting for seasonality in demographic analyses to pinpoint environmental drivers of population parameters and to refine predictions amid changing environmental conditions..
Authors: Helen Moor (Eawag - Swiss Federal Institute of Aquatic Science and Technology)+.
Session: S-8-3.
Where: G037.
When: 11:00-12:45.
Abstract: Amphibian population declines, emblematic of the freshwater biodiversity crisis, are driven by multiple stressors. Pond construction addresses habitat loss to restore the ecological infrastructure for pond-breeding amphibians. We evaluated a landscape-scale pond-construction program in a densely populated state of the Swiss lowlands to test whether this strategy benefited declining amphibian species in spite of other pressures, and to derive guidelines for where and how new ponds should be built to optimize long-term benefits.
We analysed 20 years of monitoring data for 12 pond-breeding amphibian species with dynamic occupancy models and quantified changes in metapopulation size (number of occupied ponds) while accounting for imperfect detection. We quantified effects of pond and landscape covariates, including different connectivity metrics, on both colonization and persistence probabilities in constructed ponds.
The estimated number of occupied ponds increased for 10 out of 12 species, while one species remained stable and one species further declined in the study period. Despite regional differences, in 77% of 43 regional metapopulations, the colonization of new ponds stabilized (14%) or increased (63%) metapopulation size. Species had individual preferences regarding pond characteristics. We summarized partial dependencies of colonization and persistence probabilities with the expected long-term occupancy probability (incidence) to derive concrete recommendations that would benefit multiple target species. Population connectivity mediated both colonization and persistence of conservation target species in new ponds, suggesting source-sink dynamics in newly established populations. The simple connectivity metric density of populations per km2 was a useful surrogate for dispersal kernel-weighted metrics commonly used in metapopulation theory and can be used by practitioners for site selection. .
Authors: Allison Binley (Cornell Lab of Ornithology)+; Scott Wilson (ECCC); Sara Barker (Cornell Lab of Ornithology); Amy Johnson (Smithsonian’s National Zoo & Conservation Biology Institute ); Joseph Lawrence (Cornell University); Daryl Nydham (Cornell University); Justin Proctor (Smithsonian’s National Zoo & Conservation Biology Institute); Kristan Reed (Cornell University); Bernadette Rigley ( Smithsonian’s National Zoo & Conservation Biology Institute); Joseph Waddell (Cornell University); Amanda Rodewald (Cornell Lab of Ornithology).
Session: S-1-3.
Where: G037.
When: 14:00-15:00.
Abstract: Grassland birds in North America are facing precipitous declines, largely due to habitat loss and threats posed by agricultural activities such as spring hay harvest. Our objective was to model the estimated active nesting dates for grassland bird species across the breeding range, to better understand the threats facing them and the critical dates during which they require protection. Furthermore, we aimed to identify predictors that provide enough advance notice to conservation practitioners to implement interventions in future years. We fit a logistic regression model with Stan using available data on nesting activity from various community science projects, breeding bird atlases and published literature, and climatic predictors such as monthly temperature and precipitation. The estimated probability that nests are active on a given date and in a given region were combined with the species’ relative abundance to quantify the nest destruction estimated if harvest were to occur on that date. We then explore two major applications of this model in current conservation science. First, we use these estimates to assess how the impacts of hay harvest on nesting grassland birds are changing over time and space. For example, in New York State, we found that the average recommended first alfalfa harvest date was almost two weeks earlier compared to a decade ago. Thus, even species with relatively consistent nesting dates over time were subject to changing harvest impact over time. Second, we incorporate these estimates into a structured decision-making framework that also accounts for agricultural objectives, to better understand how to maximize hay harvest quantity and quality and minimize nest destruction. Our analysis uncovers the dynamics of a major threat to declining grassland birds, while also providing tools to aid with conservation planning in agricultural landscapes moving forward..
Authors: Maëlis Kervellec (University of Montpellier)+; Thibaut Couturier (CEFE); Sarah Bauduin (OFB); Delphine Chenesseau (OFB); Pierre Defos du Rau (OFB); Nolwenn Drouet Hoguet (OFB); Christophe Duchamp (OFB); Julien Steinmetz (OFB); Jean-Michel Vandel (OFB); Olivier Gimenez (CNRS).
Session: S-8-3.
Where: G037.
When: 11:00-12:45.
Abstract: Occupancy models were originally developed to better understand species distribution while accounting for imperfect detection. Standard models account for the fact that species distribution is shaped by habitat quality but fail to accommodate the ability of individuals to reach suitable habitats. Spatial dynamic occupancy models have been proposed to extend the original framework by considering site colonisation as a function of the Euclidean distance to occupied sites. However, not all sites in the landscape are equally accessible due to the presence of barriers, corridors, etc. To account for connectivity between sites, the Euclidean distance has recently been replaced by a least-cost path distance which explicitly accounts for landscape resistance, but assumes that individuals will follow the optimal route. To relax this assumption, we first developed a new spatial occupancy model that incorporates commute distance derived from circuit theory to model colonisation across sites. This distance has the double advantage of modelling movement as a random walk and accounting that colonisation can be achieved through multiple paths. Our approach allows for the explicit estimation of landscape connectivity from detection/non-detection data and a direct measure of connectivity uncertainty. We implemented our model in the Bayesian framework using the R package NIMBLE which allows useful R connectivity functions to be called from within the model. Second, we carried out a simulation study to assess the performance of our model by considering four scenarios depicting an increasing level of landscape resistance. Third, to illustrate our new approach, we aimed to quantify the degree to which rivers facilitate Eurasian otter (Lutra lutra) dispersal and highways impede Eurasian lynx (Lynx lynx) recolonisation in France. Overall, spatial occupancy models provide a flexible framework to accommodate any distance metric designed to align with species dispersal ecology..
Authors: Otso Ovaskainen (University of Jyväskylä)+; Nerea Abrego (University of Jyväskylä).
Session: S-7-4.
Where: G029.
When: 14:30-16:30.
Abstract: Determining how the ecological niches are structured within local communities is fundamental for understanding the underlying evolutionary and ecological assembly processes. We introduce four new metrics that link joint species distribution models (JSDMs) to niche theory and are comparable among datasets. These consist of the shared and idiosyncratic responses of the species to measured and latent predictors, quantifying to what extent the known (measured) and unknown (latent) niche axes are similar (shared) or dissimilar (idiosyncratic) among the species. We illustrate with a virtual ecologist approach how the new metrics can capture niche overlap from the points of view of both fundamental niches (relevant at large spatial scales), and requirement and impact niches (relevant at small spatial scales). We use a case study on Finnish birds to illustrate how the shared response increases with increasing spatial scale, reflecting shared responses to a large-scale climatic gradient and idiosyncratic responses to small-scale habitat variation. The new methods are implemented in the JSDM framework of Hierarchical Modelling of Species Communities (HMSC) as post-processing capacity in the R-package Hmsc..
Authors: Lise Viollat (CEFE)+; Roger Pradel (CEFE); Alexandre Millon (IMBE); Cécile Ponchon (CEN PACA); Alain Ravaryol (La salsepareille); Aurélien Besnard (CEFE).
Session: S-2-2.
Where: G043.
When: 15:30-16:30.
Abstract: GPS tracking has enabled significant advances in the study of bird movements such as migration and habitat use. However, GPS devices can affect the behaviour of tagged individuals, especially when used for long periods of time, and may ultimately affect the reproduction and/or survival of individuals. While numerous studies have investigated the potential negative effects of GPS tags on birds, few have looked at these effects on both reproduction and survival simultaneously, with the latter sometimes suffering from methodological flaws. Here we investigated the effects of GPS tagging on the survival and breeding success of a medium-sized raptor, the Bonelli’s eagle (Aquilla fasciata). Forty-nine breeding adults were equipped with backpack-mounted GPS tags representing 1.2-3.5% of body mass for 36-2037 days. Using a newly developed capture-mark-recapture model that accounts for differences in detectability between the types of tags used (metal or coloured metal ring or GPS), we showed that annual survival was slightly lower for GPS-tagged individuals compared to banded-only individuals. Similarly, we found a slightly lower breeding success of GPS tagged individuals during the first breeding attempt after capture, regardless of the sex of the bird. This difference was not present in subsequent reproductions, indicating that if there is an effect of GPS tagging on breeding success, it may be relatively short-term. Furthermore, these differences in survival and reproductive success may also be due the fact that we successively tagged several young individuals on two territories that seem to concentrate mortalities induced by competition. Thus, we do not find any significant adverse effects of GPS tagging on Bonelli’s eagle population dynamics that would contraindicate the use of GPS in this or other medium to large birds, given the important information they provide for conservation purposes..
Authors: Marie-Pierre Etienne (Institut Agro)+; Marie Du Roy De Chaumary (Mathematical Research Institute of Rennes IRMAR); Salima El Koleil (Univ. Rennes, Ensai, CNRS, CREST); Matthieu Marbac (Univ. Rennes, Ensai, CNRS, CREST).
Session: S-5-4.
Where: G029.
When: 10:30-12:30.
Abstract: Hidden Markov models (HMMs) enjoy widespread popularity in ecology, particularly in movement ecology, where they serve to discern various patterns of movement interpreted as distinct internal states of individuals. The number of hidden states, termed the order of the HMM, represents a model parameter notoriously challenging to estimate. Classical approaches rely on utilizing a priori biological knowledge or likelihood-based information criteria such as AIC, BIC, and ICL. However, these methods tend to overestimate the HMM’s order, particularly when the emission distribution—i.e., the distribution of observed quantities given the hidden states—is misspecified. We propose estimating the HMM’s order within a non-parametric framework, wherein there’s no need to specify the parametric form of the emission distribution. The method relies on the fact that the order of the HMM can be identified from the distribution of a pair of consecutive observations and that this order is equal to the rank of some integral operator (i.e the number of its singular values that are non-zero). This method can accommodate various data types, including continuous, circular, or multivariate continuous data. Simulation studies indicate that this approach tends to favor a lower order compared to methods based on AIC, BIC, or ICL criteria. We illustrate the potential of this method by analyzing masked booby movement data..
Authors: Rebecca Akeresola (University of Edinburgh)+; Victor Elvira (University of Edinburgh); Ruth King (University of Edinburgh); Adam Butler (Biomathematics and Statistics Scotland); Esther L Jones (Biomathematics & Statistics Scotland); Gail Robertson (Biomathematics and Statistics Scotland).
Session: S-6-2.
Where: G049.
When: 10:30-12:30.
Abstract: Understanding animal movement and identifying the behaviours that influence animal movement is crucial to inform effective conservation management. Animal movement occurs in continuous time but is mostly sampled at discrete time points. The effectiveness of behavioural inference from animal movement trajectories may be limited by the temporal resolution at which the trajectories are sampled. Given that these behaviours are inferred from animals’ movement trajectories, it is crucial to examine how well the trajectories can be sampled using tracking devices such that little or no information is lost about the underlying latent process that is of interest to the ecological community.
We investigate this problem from the optic of the Nyquist-Shannon sampling theorem (NSST). NSST describes how to sample a continuous-time signal (or sub-sample a discrete-time signal) in a way that there is no information loss. In the context of animal movement, the theorem can be useful for selecting temporal resolutions in the animal movement trajectories such that there is no loss of information about the underlying latent behavioural process. To achieve this, we apply NSST to real and simulated animal movement trajectories to give guidance on how to sample these trajectories. We then assess how the sampled trajectory influences the ability to identify underlying behavioural latent states using HMMs. .
Authors: Alba Fuster Alonso (Institute of Marine Sciences (ICM) - CSIC)+; Jeroen Steenbeek (Ecopath International Initiative (EII)); Jorge Mestre Tomás (Institute of Marine Sciences (ICM) - CSIC); M. Grazia Pennino (Spanish Institute of Oceanography (IEO) - CSIC); Xavier Barber (Operations Research Center, Miguel Hernández University (UMH)); Jose M. Bellido (Spanish Institute of Oceanography (IEO) - CSIC); David Conesa (Department of Statistics and Operations Research (VaBar), University of Valencia (UV)); Antonio López-Quílez (Universitat de Valencia); Villy Christensen (Institute of the Oceans and Fisheries, University of British Columbia); Marta Coll (Institute of Marine Sciences (ICM) - CSIC).
Session: S-2-3.
Where: G049.
When: 15:30-16:30.
Abstract: Marine Ecosystem Models (MEMs) have been developed to analyse the past and future dynamics of the oceans. One of such efforts is EcoOcean, a complex, mechanistic and spatio temporal explicit MEM of the global oceans based on a trophodynamic core. EcoOcean can be informed with the species native ranges and suitable habitats. For key environmental variables, species’ functional responses and time-varying maps delivered by Earth System Models (ESMs) are needed. The different sources of uncertainty in these inputs may influence the validity and accuracy of EcoOcean results. For this reason, our study explores the use of global SDMs to reduce the uncertainty associated with these inputs. A promising new alternative to traditional SDMs classification tree methods is the Bayesian Additive Regression Trees (BART). BART is a non-parametric Bayesian regression approach based on a sum of-trees model. Our hypothesis is that BART can be a powerful approach to inform global-scale Marine Ecosystem Models (MEMs). In this study, we compare the projected results of EcoOcean when incorporating the outputs of BART as inputs and compare them with runs without informing the global model. Specifically, we perform a study on the combination of BART and EcoOcean targeting several species of marine mammals and top predators (marine turtles, tunas, sharks, etc.). All functional groups include species distributed very differently in the marine environment, some being highly commercial and other being non-commercial, and show a global distribution as a functional group..
Authors: Lucy Y Brown (University of Kent)+; Eleni Matechou (University of Kent); Bruno Santos (Stockholm University); Eleonora Mussino (Stockholm University).
Session: S-5-2.
Where: G049.
When: 10:30-12:30.
Abstract: Demographers have the essential task of estimating the number of individuals living in a country at a given time, but overcoverage makes this task difficult and can lead to serious bias in population estimates, negatively influencing policymaking and research. Overcoverage occurs when individuals are registered as living in the country but in fact live elsewhere, i.e. when we have imperfect emigration registration. Demographers in Sweden have been working towards estimating overcoverage, and in turn the true population size by using Swedish Population registers, which I have been given access to for this project. Hidden Markov Models (HMMs) provide an efficient and simple structure for computing likelihoods in Cormack-Jolly-Seber (CJS) type capture-recapture-recovery (CRR) models as they account for the observations as well as infer the latent states. I have created a CJS type CRR model which is able to account for Temporary Emigration of individuals, as well as using multinomial regression to incorporate an arbitrary number of, possibly interacting, observation lists. The model was built and tested using simulated data, but has since been applied to the real aforementioned data from Sweden, with promising present results which indicate that the model is working as expected. While this model was built with the aim to estimate overcoverage and the true population size from incomplete registers in Sweden, there is the possibility of application to other similar countries also..
Authors: Aimée Freiberg (University of Fribourg)+; Madleina Caduff (University of Fribourg); Daniel Wegmann (University of Fribourg).
Session: S-3-1.
Where: G029.
When: 10:30-12:30.
Abstract: Camera traps have been implemented in many studies regarding animal ecology, behavior, and conservation. While camera traps collect extensive observational data, their analysis is currently slowed because images remain to be annotated predominantly manually by experts or citizen scientists. Recently, a lot of effort has gone into developing and training neural networks to automate the labelling process. Their accuracy, however, is often much lower than manual annotation and as a consequence, only very few ecological studies are currently using automated labelling for their analyses. Since annotation accuracy is unlikely to reach the desired performance for many years, especially for critically endangered and hence rare species, we argue that to unlock the full potential of camera traps, downstream analyses must deal with imperfect classifications. To do so, we propose to explicitly model per-image AI classification errors in downstream, multi-species analyses, and show that they can be learned from both a test set of manually classified images, as well as through ecological information propagating in downstream models. We illustrate this approach through a multi-species occupancy model and show, using simulations and applications to real data that occupancy can be accurately inferred even for species that are poorly classified, if annotation errors are properly accounted for..
Authors: James A. Clarke (British Trust for Ornithology)+; Philipp Boersch-Supan (British Trust for Ornithology); Jeremy Smith (British Trust for Ornithology); Robert Robinson (British Trust for Ornithology).
Session: S-6-3.
Where: G029.
When: 10:30-12:30.
Abstract: Productivity is an important measure in determining population dynamics and its accurate estimation will help monitor populations in a changing environment. When applied to birds, productivity can be measured by counting the number of eggs (clutch size) and young (brood size) per nest. Typically, the Poisson distribution is used to model counts. However, this is often not appropriate when modelling clutch and brood sizes due to the Poisson’s strict distributional assumptions. For nest counts these are usually violated due to under-dispersion, when there is less variation in the counts than would be estimated by the Poisson, due to evolutionary constraints on these measures.
A possible method to account for under-dispersion is the exponentially weighted Poisson (EWP) (Ridout & Besbeas, 2004). This weighted Poisson utilises two additional parameters which control the compression of the distribution either side of the mean. This allows for greater control of the shape of the distribution and thereby provides insights into the magnitude of the lower and/or upper evolutionary constraints acting upon species traits. The Conway-Maxwell Poisson, a popular option for modelling under-dispersed counts, does not provide these insights because of its formulation with only one additional dispersion parameter.
We apply the EWP to citizen-science reproductive data of 62 UK bird species collected from 1960-2022 and present an R package that allows for convenient use of the EWP. We compared the results to outputs from both the standard Poisson and the Conway-Maxwell Poisson when used to model temporal trends in clutch and brood sizes and found that the EWP outperformed both based on AIC. The EWP was also able to provide insights such as that the upper constraint tended to be larger than the lower constraint, as there was greater evolutionary pressure on having too many chicks or eggs than too few. .
Authors: Simon Lacombe (CEFE)+; Sébastien Devillard (Laboratoire de biométrie et de biologie évolutive); Cécile Kauffmann (Société Francaise pour l’Etude et la Protection des Mammifères ); Mélanie Aznar (Groupe Mammalogique d’Auvergne); Xavier Birot-Colomb (LPO Auvergne-Rhône-Alpes); Ondine Dupuis (LPO Bourgogne-Franche-Comté); Christine Fournier-Chambrillon (Groupe de Recherche et d’Etude pour la Gestion de l’Environnmement); Pascal Fournier (Groupe de Recherche et d’Etude pour la Gestion de l’Environnmement); Camille Fraissard (LPO Occitanie); Nicolas Fuento (LPO PACA); Tiphaine Heugas (LPO Vendée); Alexandre Martin (LPO Pays de la Loire); Meggane Ramos (Groupe Mammalogique Breton); Antoine Roche (Groupe Mammalogique et Herpétologique du Limousin); Thomas Ruys (Groupe de Recherche et d’Investigation sur la Faune Sauvage); Franck Simonnet (Groupe Mammalogique Breton); Daniel Sirugue (Société d’histoire naturelle d’Autun); Bastien Thomas (Groupe Mammalogique Normand); Angélique Villéger (Sologne Nature Environnement); Olivier Gimenez (CNRS).
Session: S-7-2.
Where: G043.
When: 14:30-16:30.
Abstract: Monitoring changes in the distribution of a species on a large (e.g. national) scale requires a large amount of data. However, lack of centralized sampling strategy often results in heterogeneous datasets with variations in sampling effort across space and time. Integrating multiple datasets from different monitoring plans, including more local programs and citizen science initiatives, is a promising way to address this issue, but presents notable statistical challenges. As a case study, we study the Eurasian otter (Lutra lutra) in France, which, following an important decline in the past century, has undergone a slow recolonization process that has accelerated in the last 15 years. To date, the species failed to recover in some parts of the country, and a better knowledge of the species recent range expansion is required to understand the factors that still limit recolonization. However, data on this species’ dynamics remains limited for two main reasons. Firstly, otters’ elusive nature means assessing their presence relies heavily on scat identification, and therefore requires expert knowledge. Secondly, despite the recent introduction of a standardized monitoring program in France, the effort is not constant across all regions, and prospections often occurred after the species has established in an area. Here, we propose an integrated population model to map Eurasian otter’s recolonization in France between 2009 and 2023. Our model combines several datasets across the country with different temporal extents, while accounting for imperfect detection and spatio-temporal autocorrelation in the species’ presence. In particular, our model includes both detection/non-detection data from several monitoring programs and presence-only opportunistic data. By providing distribution maps with an annual temporal resolution across 15 years of recolonization, our model will provide a tool to further explore the factors still limiting otter’s range, and to ultimately identify areas with a high potential for installation of this species in the near future..
Authors: Toryn Schafer (Texas A&M University)+.
Session: S-8-1.
Where: G043.
When: 11:00-12:45.
Abstract: The evolution of ecological technology has ushered in an era of abundant and diverse wildlife movement data, necessitating robust modeling approaches. This presentation introduces Bayesian methodologies tailored to tackle the scientific challenges posed by these datasets. I highlight multiscale data collected from remote wildlife devices including high-dimensional accelerometers and video collars and GPS based locations. Emphasis is placed on addressing nonlinear interactions among organisms and their environment, accounting for non-stationarity, managing multiple data sources, and handling instances of missing-not-at-random. The proposed Bayesian models offer a systematic framework to extract insights into animal behavior, decision-making processes, and ecological dynamics. By integrating prior scientific knowledge and quantifying uncertainty, these models contribute to a more rigorous understanding of the complexities inherent in wildlife movement data analysis..
Authors: Eloise Bray (University of Sheffield)+.
Session: S-1-2.
Where: G049.
When: 14:00-15:00.
Abstract: Discrete-time step-and-turn type models for animal movement remain one of the most widely adopted methods in movement ecology, due to their simple formulation and implementation. But what if we could use a similar model, formulated in continuous time, that could overcome the common drawbacks of discrete time? Velocity-jump models are a relatively under-used class of movement models, where we assume an animal moves with a velocity that stays constant until some ‘event’ causes it to change, or ‘jump’. This produces a path very similar to a step-and-turn model, but on a continuous timescale.
A paper by Blackwell (submitted for publication) describes a way of performing exact Bayesian inference for these velocity-jump models, but unfortunately it is relatively slow and therefore infeasible for larger datasets. In my talk I will be describing a method I am developing for approximate Bayesian inference for these models by framing the movement process as a Hidden Markov Model. This means we can utilise the Forward Algorithm, therefore speeding up the process.
We can demonstrate this technique with simulated or real animal movement data. In particular, the eventual goal is to realistically model the out and back movement paths of seabirds during the breeding season. By showcasing its effectiveness in this way, we aim to highlight the practical significance of this approach and its potential for broader adoption in movement ecology research. .
Authors: Jenny A Hodgson (University of Liverpool)+; Claudia Gutierrez-Arellano (University of Liverpool).
Session: S-7-4.
Where: G029.
When: 14:30-16:30.
Abstract: Reversing biodiversity losses depends greatly on our understanding of the impacts of land use change. Several influential indicators attempt to track global biodiversity trends, but some biases in the data that underlie these indicators are challenging to overcome. Species exist on a gradient of tolerance to disturbance, and the more susceptible species might be prone to under-sampling. Here, we use a statistical approach to reveal the impact of this under-sampling and then to correct for it. We used land cover and intensity-related local abundance data compiled in the PREDICTS database, because this database feeds into global biodiversity indicators including the Biodiversity Intactness Index (BII; based on abundance and compositional similarity). As an independent metric of species ‘recordability’ relative to their taxonomic group we used the number of GBIF (Global Biodiversity Information Facility) occurrence records. Our model fitted to >4000 species of birds, plants and spiders reveals an interaction whereby species with fewer occurrence records are consistently more negatively affected by anthropogenic land uses (compared to primary vegetation). Furthermore, the species captured in PREDICTS studies are disproportionately those with more occurrence records (given their taxonomic group). The model allows extrapolation of the mean log change in abundance, to species outside the PREDICTS dataset, for which the number of GBIF records is available. The simple decline indicator based on our model —either with or without extrapolated species — predicts a greater proportional loss arising from land use change than the BII does. The use of readily available occurrence data as a metric of species sensitivity - even though it is affected by multiple biological and human factors that cannot be disentangled - could offer important insights into the state of global biodiversity, by better representing the gradient of responses across species that are easier and more difficult to record. .
Authors: Jonathan Potts (University of Sheffield)+.
Session: S-3-2.
Where: G043.
When: 10:30-12:30.
Abstract: This talk will describe various techniques for deriving broad-scale utilisation distributions from a class of movement kernels that can be parametrised by step selection analysis. We give examples of applying these techniques to empirical data, highlighting how they can be used to assess the goodness of fit between models and data, thus informing future modelling and data gathering experiments. We also show how this modelling approach can lead the researcher to a class of partial differential equations with rich pattern formation properties, linking statistical/empirical ecology to mathematical biology and nonlinear dynamics..
Authors: Toshihide Kitakado (Tokyo University of Marine Science and Technology)+.
Session: S-5-1.
Where: G037.
When: 10:30-12:30.
Abstract: Scientific assessment frameworks for wildlife and fisheries management are based on two main paradigms: “estimation of population status” and “evaluation of management procedure”. Normally, a population dynamics model is fitted to time-series data of population indices, removals and age/size compositions. The model is then used to assess historical population status relative to reference points, and to predict the outcomes of alternative management options. As a normal course of procedure, model diagnostics are conducted to test goodness of fit and inspect residual patterns. Given the ultimate goal of development of a management advice based on the fitted model, prediction error of the model should also be evaluated. Here, we propose a hindcasting approach as a method to evaluate models through their prediction skills. Models are retrospectively re-run by removing recent years’ data and the population trajectories are forecasted up to the most recent year to compare with the observed time-series of a population index (which is used in the original model fitting). The presentation introduces examples of application of the hindcasting approach to pelagic fish and marine mammal populations with some quantitative indicators and discuss some caveats of the approach..
Authors: Benjamin R Goldstein (North Carolina State University)+; Alex Jensen (North Carolina Museum of Natural Sciences); Roland Kays (North Carolina Museum of Natural Sciences); Elizabeth Kierepka (North Carolina Museum of Natural Sciences); Michael Cove (North Carolina Museum of Natural Sciences); William McShea (Smithsonian’s National Zoo and Conservation Biology Institute); Brigit Rooney (Smithsonian’s National Zoo and Conservation Biology Institute); Krishna Pacifici (North Carolina State University).
Session: S-5-3.
Where: G043.
When: 10:30-12:30.
Abstract: Ecologists are discovering that species’ environmental associations are nonstationary across their ranges–the effect of environmental conditions on the species’ distribution may differ from one region to another. To understand why such nonstationarity occurs, we need a method for differentiating between hypotheses for this variation, which include local adaptation; regional differences in covariate effects; or intra-species genetic differences between populations. Here, we synthesize several recent advances in wildlife monitoring and species distribution modeling to develop highly resolved, large-scale estimates of species occupancy. We use data integration to jointly model detections arising from two datasets, a camera trap dataset (comprising Snapshot USA and supplemental sites for a total of 400,000 trap-nights over 4 years) and iNaturalist (1.5 million participatory science observations), to estimate species occupancy at a 5km scale across the continental U.S. For each of a set of land use and climate covariates, we estimate spatially varying coefficients (SVCs). We consider four representations of SVCs, representing competing hypotheses for nonstationarity. Finally, we use on-line variable selection during MCMC sampling in NIMBLE to differentiate between SVC representations. By identifying relative model support for these competing hypotheses, we provide the first test distinguishing between hypothesized drivers of nonstationarity in mammals’ environmental responses..
Authors: Alec BM van Helsdingen (University of Auckland)+.
Session: S-3-3.
Where: G049.
When: 10:30-12:30.
Abstract: Spatial capture-recapture (SCR) is a well-established method used to estimate animal population size from animal sighting or trapping data. Standard SCR methods assume animal movements are temporally independent and consequently cannot incorporate site fidelity (attachment to a particular region) nor the temporal correlation of an animal’s location. Recent work has sought to solve these issues by explicitly modelling animal movement.
In this talk we propose an alternative solution for camera trapping surveys based on a multivariate self-exciting Hawkes process. Here the rates of detection of a given animal at a given camera are a function of not only the location and its proximity to the animal’s activity center, but also where and when the animal has been previously detected.
Our model is based on a mixture of Gaussian distributions, one of which is the home range as in SCR, and the other is centered at the site of the last detection. The weight on the second Gaussian decays exponentially in time after each detection. This formulation, we believe, better reflects animal behaviour because shortly after detection, we expect to see an individual close to where it was last seen. Only after a sufficiently long time could it be anywhere within its home range. Thus, our model allows us to account for both site fidelity and the inherent temporal correlation in detections that have not previously been accounted for in SCR-type models.
In this talk, I will 1) give an overview of Self-Exciting Spatial Capture-Recapture (SESCR) models, and 2) demonstrate the additional inference that can be drawn from such models and 3) apply the framework using a few case studies to compare traditional SCR and SESCR. .
Authors: LineekelaOmwene Tuhafeni Nauyoma (University Of Namibia)+; Camille H Warbington (University of Alberta); Fernanda C Azevedo (Universidade Federal de Catalão); Frederico G Lemos (Universidade Federal de Catalão,); Fernando Sequeira (Universidade Do Porto,); Ezequiel C Fabiano (University Of Namibia).
Session: S-7-3.
Where: G037.
When: 14:30-16:30.
Abstract: Density and abundance estimates are critical to effective wildlife management and are essential for monitoring population trends and setting effective quotas for harvesting. Management of roan (Hippotragus equinus) and sable (H. niger) antelopes in Mudumu National Park (MNP), Namibia, is challenging because they are elusive, naturally unmarked, and believed to occur at low densities. The species are threatened by habitat fragmentation, human population growth, and illegal hunting, and reliable density and abundance estimates have not been quantified, hampering management and conservation plans. Our objective was to estimate roan and sable densities and abundances using the time in front of the camera model (TIFC) and the Poisson-binomial N-mixture model (PB), respectively. We also evaluated the effects of environmental and ecological variables on roan and sable abundance. We used data from two camera trap surveys conducted between March and September 2021 in the MNP. Results showed that the TIFC model provided low-density estimates of 1.62 (95% CI 1.61–1.64) roans/km2 and 2.46 (95% CI 2.42–2.50) sables/km2, consistent with trends reported in Africa where these species occur at low densities. In addition, the total abundance of roans and sables in the MNP from the PB model were 57 and 242, respectively. Higher roan abundance occurred in sites with higher grass cover. This study provides the first accurate camera trap-derived density and abundance estimates for roan and sable in the MNP, which will be critical for developing comprehensive conservation programs and strategies that are likely to reduce the risk of extinction for both species..
Authors: Devin Johnson (National Oceanic and Atmospheric Administration)+; Mevin Hooten (University of Texas at Austin); Brian Brost (National Oceanic and Atmospheric Administration ).
Session: S-5-3.
Where: G043.
When: 10:30-12:30.
Abstract: We present a multistage method for making inference at all levels of a Bayesian hierarchical model (BHM) using natural data partitions to increase efficiency by allowing computations to take place in parallel form using software that is most appropriate for each data partition. The full hierarchical model is then approximated by the product of independent normal distributions for the data component of the model. In the second stage, the Bayesian maximum a posteriori (MAP) estimator is found by maximizing the approximated posterior density with respect to the parameters. If the parameters of the model can be represented as normally distributed random effects, then the second-stage optimization is equivalent to fitting a multivariate normal linear mixed model. We consider a third stage that updates the estimates of distinct parameters for each data partition based on the results of the second stage. The method is demonstrated with two ecological data sets and models, a generalized linear mixed effects model (GLMM) and an integrated population model (IPM). The multistage results were compared to estimates from models fit in single stages to the entire data set. In both cases, multistage results were very similar to a full MCMC analysis..
Authors: Hideyasu Shimadzu (Kitasato University)+.
Session: S-7-4.
Where: G029.
When: 14:30-16:30.
Abstract: The UN’s declaration of 2021–2030 as the Decade on Ecosystem Restoration underlines the imperative to address the unprecedented degradation of ecosystems and biodiversity worldwide. In light of this urgency, there is a pressing demand for a thorough comprehension of the extent to which biodiversity changes across both time and space to facilitate effective restoration efforts. Despite recent advancements and investigations into various biodiversity indices, discrepancies in these outcomes remain a challenge. A pivotal aspect lacking in current understanding is a more unified perspective indicating the essence of what these biodiversity indices aim to quantify.
This presentation commences by revisiting classical ecological concepts of biodiversity, namely alpha-, beta and gamma-diversities. Subsequently, we propose a structured interpretation rooted in abundance distributions, where shifts therein correspond to types of biodiversity changes. This suggests that prevalent biodiversity indices are fundamentally estimators of deviation in abundance distributions from one state to another. We introduce the Kullback-Leibler divergence and adopt an information geometric approach to consolidate a range of commonly studied biodiversity indices within a unified theoretical framework. We also illustrate how this proposed framework facilitates biodiversity analyses to elucidate the dominant factors influencing observed variations and shifts in biological diversity patterns..
Authors: Francisca Powell-Romero (The University of Queensland)+; Konstans Wells (Swansea University); Nicholas Clark (The University of Queensland).
Session: S-8-2.
Where: G049.
When: 11:00-12:45.
Abstract: Understanding biotic interactions is a crucial goal in community ecology and multi-species modelling. With the rise of readily available big data and the broad interest in understanding species distributions in times of global change, ecologists have often sought to draw inferences on biotic interactions using cross-sectional data of species co-occurrences. Large strides have been made towards improving multivariate computational methods with the aim of quantifying biotic interactions and improving predictions of species occurrence. At the same time, a growing body of literature has also emerged, cautioning against inferences of these interactions with binary co-occurrence data. Yet, while considerable attention has given to methodological approaches and the interpretation of these methods, the importance of sampling design to reveal these biotic interactions has received little consideration in comparison.
We explored the influential role of priority effects, that is the order of habitat colonization, in shaping our ability to detect biotic interactions from occurrence data alone. Despite the longstanding recognition of its profound effects on the structure of communities and possible asymmetric interaction outcomes, these effects have been overlooked in the vast literature on joint species distribution and biotic interaction modelling. Using a simple set of simulations, we demonstrate that common approaches to detect interactions from binary data simply cannot make reliable inferences. This is the case even if existing models perform well in predicting the occurrence of species. We then show how sampling designs that consider priority effects can accurately infer biotic interactions. We urge for caution when drawing inferences from cross-sectional binary co-occurrence data, and we conclude with guidance on the types of sampling designs that may provide ecologists with the necessary data to tackle this longstanding challenge..
Authors: Matthieu Paquet (SETE CNRS)+; Frédéric Barraquand (Institute of Mathematics of Bordeaux, University of Bordeaux, CNRS, Bordeaux INP).
Session: S-1-1.
Where: G043.
When: 14:00-15:00.
Abstract: Most communities of competing species have an age or stage structure, with possible competition between and within species at various life-stages. Predicting competitive outcomes and inferring coexistence mechanisms in such stage- or age-structured communities typically requires fitting dynamical models to data, from which invasion criteria and eventually coexistence indicators can be derived. Methods that allow to fully propagate parameter uncertainty and demographic stochasticity while estimating interaction strength and invasion or coexistence criteria are particularly indicated. These should ideally make the best out of multiple data sources, each of which can be relatively scarce by statistical standards. Here, we embed a mathematical model of stage-structured competition between two species, producing analytical invasion criteria, into a two-species Integrated Population Model framework. The community-level IPM fulfills the above requirements, and allows to combine for both species multiple data sources (counts, capture-recapture, fecundity data) into a single statistical framework. Our Bayesian formulation of the IPM fully propagates parameter uncertainty. Model fitting demonstrates that we can predict correctly coexistence through reciprocal invasion when present, though cases with extinction are little harder. Prior specification can have a relatively large influence on the results, with some priors leading to nonidentifiability in species interaction parameters, so we recommend to always plan for some form of prior sensitivity analysis in such Bayesian dynamic community model fitting exercises..
Authors: Anthony Seveque (Senckenberg Research Institude)+; Robert Lonsinger (USGS OK Cooperative Fish & Wildlife Research Unit ); Lisette Waits (University of Idaho); Kristin Brzeski (Michigan Technological University); Caitlin Ott-Conn (Michigan Department of Natural Resources); Sarah Mayhews (Michigan Department of Natural Resources); Cody Norton (Michigan Department of Natural Resources); Tyler Petroeljet (Michigan Department of Natural Resources); Anaïs Tallon (Mississippi State University); Dana Morin (Mississippi State University).
Session: S-5-2.
Where: G049.
When: 10:30-12:30.
Abstract: Close-kin mark–recapture (CKMR) methods use information on genetic relatedness among individuals to estimate demographic parameters such as adult abundance and survival probability. By treating an individual’s genotype as a “recapture” of each of its parent’s genotype, CKMR uses the frequency of kin-pair matches detected in population sample to inform estimates of population parameters. A notable advantage of this novel method is that it does not require recapturing individuals, and thus enables the opportunistic use of lethal samples such as those from harvest or road kills. CKMR compares the “true” kinship between two individuals (confirmed via genotyping) to their prior probability of relatedness given a set of covariates. Factors related to reproductive biology and sampling protocol influence which individual covariates and demographic parameters are of importance to the kinship probabilities and models being fitted. In this study, we introduce CKMR models applied to an American black bear (Ursus americanus) population in Michigan, using lethal samples from harvest to estimate population size. Our models incorporate various reproductive traits such as breeding age variability, age-dependent fecundity, and intermittent breeding. We compare the abundance and survival estimates with models using parent-offspring pairs and half-sibling pairs probabilities. Furthermore, we explore the effects of incomplete mixing and dispersal limitation on the accuracy of CKMR estimates. We propose a spatially-explicit formulation of CKMR models that integrates sex-specific probability density functions of natal dispersal distances in terrestrial species. In conclusion, our findings demonstrate that CKMR has a great potential to be used as a population monitoring tool, but its effective implementation requires rigorous tailoring to the characteristics of the target population and consideration of potential sampling bias..
Authors: Edward Lavender (Eawag)+; Andreas Scheidegger (Eawag); Carlo Albert (Eawag); Stanisław Biber (University of Bristol); Janine Illian (University of Glasgow); James Thorburn (Edinburgh Napier University); Helen Moor (Eawag - Swiss Federal Institute of Aquatic Science and Technology).
Session: S-4-1.
Where: G029.
When: 14:30-16:30.
Abstract: 1. Particle filters and particle smoothers are powerful sequential Monte Carlo algorithms used to fit non-linear, non-Gaussian state-space models. These algorithms are well placed to fit process-orientated models to animal tracking data, especially in autonomous receiver networks, but to date they have received limited attention in the ecological literature. 2. We introduce a Bayesian forward-filtering backward-smoothing algorithm that reconstructs individual movements and patterns of space use from animal tracking data, with a focus on passive acoustic telemetry systems. Within a sound probabilistic framework, the methodology uniquely integrates the movement process (both within and between periods of detection) and the observation processes of disparate datasets, while correctly representing uncertainty. We provide flexible R and Julia packages that implement the methodology. In a comprehensive simulation-based analysis, we compare the performance of our algorithm to the prevailing, heuristic methods used in passive acoustic telemetry systems. 3. We find the particle smoothing methodology outperforms heuristic methods across the board. Particle-based maps consistently represent simulated movements more accurately, even in dense receiver networks, and are better suited to analyses of residency, habitat preferences and home ranges. We are currently applying the methods to data from the Critically Endangered flapper skate (Dipturus intermedius) to support management of this species in Scotland. 4. This work sets a new state-of-the-art for movement modelling in autonomous receiver networks. The particle filter is a robust, flexible and intuitive modelling framework with potential applications in many ecological settings..
Authors: Diana Cole (University of Kent)+; Daniel Bearup (University of Leicester).
Session: S-6-1.
Where: G043.
When: 10:30-12:30.
Abstract: Assessing identifiability in Bayesian analyses often involves checking the overlap between the prior and posterior distributions (Garrett and Zeger, 2000, Gimenez et al., 2009). Typically, an overlap of more than 35% is used; however, it is possible for a non-identifiable model to have an overlap less than 35%.
For an identifiable parameter, the marginal posterior distribution will tend to a normal distribution with a variance that tends to 0, as the sample size tends to infinity (Walker, 1969). This implies that the overlap for an identifiable parameter will decrease towards zero as the sample size increases. However, this is not the case for a non-identifiable parameters. Instead, the marginal posterior distribution tends to a fixed distribution that is dependent on how the parameters are confounded, resulting in the overlap tends to a fixed value, which could be less than 35%. This is equivalent to the theory behind testing for estimability using data cloning (Lele et. al, 2010).
Using the prior and posterior overlap as a test for identifiability is unreliable. Instead, we illustrate, using occupancy models and integrated population models, how the method can be used more reliably, with large simulated data sets or data cloning.
Garrett, E.S. and Zeger, S.L. (2000) Latent class model diagnosis. Biometrics, 56, 1055-1066.
Gimenez, O., Morgan, B. J. T., and Brooks, S. P. (2009) Weak identifiability in models for markrecapture-recovery data. In Modeling demographic processes in marked populations. Thomson, D., Cooch, E.G. and Conroy, M.J. (Eds.) Springer, 3, 1057-1070.
Lele, S. R., Nadeem, K. and Schmuland, B. (2010) Estimability and likelihood inference for generalized linear mixed models using data cloning. Journal of the American Statistical Association, 105, 1617-1625.
Walker, A. M. (1969) On the Asymptotic Behaviour of Posterior Distributions, Journal of the Royal Statistical Society. Series B, 31, 80-88..
Authors: Elliot Dovers (UNSW Sydney)+.
Session: S-8-2.
Where: G049.
When: 11:00-12:45.
Abstract: The biplot has become an important tool in analysing multivariate data in ecology. For occurrence data from multiple species, collected at survey sites, the visualisation technique offers an efficient way of communicating species composition at sites and co-occurrence patterns across species. Biplots can be produced by multivariate statistical models with a factor analytical structure, and have been applied to a variety of data types such as: counts of individuals, presence/absences, and continuous measurements (such as biomass). However, they have not previously been applied to multivariate presence-only data, where model-fitting is difficult, and the notion of site is not well-defined. We present a novel, model-based approach to generate biplots for presence-only data on multiple species by fitting a multivariate log-Gaussian Cox process. We approximate species-specific latent Gaussian random fields using spatially explicit basis functions, which play the role of sites in our analysis, and use a factor-analytic approach to fit the model simultaneously to many species. This approach allows us to reexpress the problem as a generalised latent variable model and so can be implemented using popular generalised linear mixed modelling software. We demonstrate fitting the model and visualising the co-occurrence patterns through a biplot on a range of multivariate presence-only datasets and present simulations indicating how key decisions on model design can be made in a data-driven way..
Authors: Karel J Kaurila (University of Helsinki)+; Sanna Kuningas (Natural Resources Institute Finland); Antti Lappalainen (Natural Resources Institute Finland); Jarno Vanhatalo (University of Helsinki).
Session: S-2-3.
Where: G049.
When: 15:30-16:30.
Abstract: Species distribution models (SDMs) are key tools in ecology, conservation, and natural resources management. They are commonly informed by scientific survey data but, since surveys are expensive, there is an urgent need to augment them with cheaper, complementary information sources. One such source is expert knowledge in the form of assessments on species occurrence probabilities. Expert assessments are, however, inherently subjective and prone to biases, both of which are challenging deficiencies to account for with existing expert elicitation approaches. Moreover, there are still few examples demonstrating improvement in species distribution predictions when augmenting survey data with expert elicitation. We have developed a novel approach for using expert knowledge within SDMs, where we estimate experts’ reliability, correct for local biases in their assessments, and weight assessments based on how well they align with survey data. In our approach experts summarize their beliefs on species occurrence probabilities over regions in the study area. These summaries are then combined with survey data in a hierarchical Bayesian model to predict species abundance in the study area. We tested our approach with a simulation study and applied it in a real world case study on spring spawning pikeperch larvae in a coastal area in the Gulf of Finland.
In both simulated and real world cases, augmenting survey data with expert information improved species distribution predictions compared to only using survey data. In the case study, experts’ reliability varied considerably. Out of 10 experts only five provided useful information for inference and prediction. Even the generally reliable experts had spatially structured local biases in their assessments. Our results show that expert elicitation can be an efficient tool for improving species distribution model predictions and, thus, can help ecological studies, natural resources management and conservation area planning..
Authors: Kwaku Peprah Adjei (Norwegian Institute for Nature Research)+; Robert B O’Hara (NTNU); Nick Isaac (UK Centre for Ecology & Hydrology ); Francesca Mancini (UK Centre for Ecology & Hydrology); Claire Carvell (UK Centre for Ecology & Hydrology).
Session: S-4-2.
Where: G049.
When: 14:30-16:30.
Abstract: Developing methods for making inferences and predictions about diversity and species distributions is necessary, especially with biodiversity monitoring programs generating disparate data types. One of these methods is the integrated (community) distribution models (IDMs), which combine different data types from different sampling protocols to capture the features of both datasets.
Most issues that IDMs have been used for are case studies of particular applications or explorations of some statistical challenges data integration presents. In most of these studies, IDMs have been shown to improve predictive performance and have higher precision of estimated parameters as compared to single-dataset models.
Here, we highlight an unrealised benefit of IDMs by presenting a model that combines presence-absence and count data obtained at different taxonomic levels to estimate the alpha diversity of insect pollinators in the UK.
Our proposed model outperforms the single-dataset model’s prediction accuracy for some of the insect groups and provides a better estimate of community alpha diversity. With various biodiversity monitoring technologies that are generating data at different resolutions, our proposed model further extends the range of applications for IDMs in ecology as the most available data is utilised to provide better inference about biodiversity..
Authors: Audun Rugstad (NTNU)+; Robert B O’Hara (NTNU); Bert van der Veen (Norwegian University of Science and Technology).
Session: S-7-1.
Where: G049.
When: 14:30-16:30.
Abstract: In emerging fields such as restoration ecology, there is a very real demand for statistical models and approaches that allow for inference about the amount of time an ecosystem uses to meaningfully “recover” from a disturbed or impacted state to a purportedly more natural reference state, given different treatments and environmental conditions. However, the few current multispecies methods that attempt to address this challenge, are usually dependent on non-statistical dimension reduction methods such as NMDS in combination with linear modeling (Rydgren et al., 2019), making the robustness and meaningfulness of the resulting estimates potentially hard to assess both statistically and ecologically.
Here, we will instead propose and outline a novel approach to modeling multispecies time series based on Generalized Linear Latent Variable Models (GLLVMs) (Niku et al., 2023). The field of GLLVM modeling has seen major advances in the last decade – making it a fast and flexible model-based alternative to older ecological ordination methods, one which also includes a wide range of possible model specifications designed to better account for ecological data. By incorporating models non-stationary time series analysis into the GLLVM framework, we believe it possible to come up with a more statistically coherent way of modeling ecosystem change over time, with better ways to estimate statistical uncertainty, as well as providing ways to assess the relative fit of different time series models that might be realistic in terms of ecosystem change, such as directional random walks, Ornstein-Uhlenbeck processes and cyclic trends, to a time-structured species-site dataset. The goal is that this will result in novel and robust tools for inference and prediction about the trajectories of ecosystems that can be applied to real-world ecological monitoring and management, with ecological restoration as the referential case study.
References Niku, J., Brooks, W., Herliansyah, R., Hui, F. K. C., Korhonen, P., Taskinen, S., Veen, B. van der, & Warton, D. I. (2023). gllvm: Generalized Linear Latent Variable Models. Rydgren, K., Halvorsen, R., Töpper, J. P., Auestad, I., Hamre, L. N., Jongejans, E., & Sulavik, J. (2019). Advancing restoration ecology: A new approach to predict time to recovery. Journal of Applied Ecology, 56(1), 225–234. https://doi.org/10.1111/1365-2664.13254 .
Authors: Anouk Glad (INP-ENSAT, UMR 1201 Dynafor, Auzeville-Tolosan)+; Sylvain Moulherat (OïkoLab, TerrOïko, 2 place Dom Devic, Sorèze,); Emilie Andrieu (INRAE, UMR 1201 Dynafor, Castanet-Tolosan); David Sheeren (INP-ENSAT, UMR 1201 Dynafor, Auzeville-Tolosan); Annie Ouin (INP-ENSAT, UMR 1201 Dynafor, Auzeville-Tolosan).
Session: S-1-2.
Where: G049.
When: 14:00-15:00.
Abstract: Wild bees are known to participate in the pollination of a wide variety of flowering plants worldwide. To predict the pollination potential in a landscape, a comprehensive knowledge of their foraging movement behavior is essential. Nevertheless, in the majority of the models aiming to estimate pollination at a landscape scale, the probability of discovering a resource mainly depends on the distance to the nest without considering landscape heterogeneity.
Animal movement has been modeled by a large variety of algorithms. Among them, the stochastic movement simulator (SMS) presents better performances in estimating relative connectivity (Palmer et al., 2011). This method is based on a series of sequential movement decisions incorporating path memory, a directional parameter, a perceptual range and the movement cost, allowing to take into account the spatial arrangement of the landscape.
This study aims to evaluate the contribution of landscape heterogeneity (composition and configuration) on wild bee’s pollination services estimates using a “central place forager” (CPF) model and SMS. First, the role of the landscape heterogeneity on the pollination estimates was investigated by performing simulations on a set of virtual landscapes to cover a gradient of fragmentation, patch isolation, and composition. The results obtained by the CPF-SMS model were compared to those obtained by a distance-weighted kernel model (InVEST Lonsdorf et al., 2009). In the second part, the model sensitivity to path memory, directional and perceptual range parameters was explored using a real agricultural landscape located in the long-term Socio-Ecological Research site PYGAR (Occitanie, France) and two species differing by their size and movement ability (Lasioglossum marginatum and Andrena flavipes). The preliminary results show large differences in pollination estimates for intermediate fragmented landscapes. CPF-SMS model is sensitive to the memory path by increasing the total number of passage whereas the perceptual parameter augmentation slightly decreases the number of passage..
Authors: Valentin Lauret (CEFE - University of Montpellier)+; Nicolas Courbin (CEFE - CNRS); Aurélien Besnard (CEFE - EPHE ).
Session: S-4-2.
Where: G049.
When: 14:30-16:30.
Abstract: Telemetry data and population counts contribute to understand how animals select the landscape features despite highlighting complementary aspects of habitat selection, from detailed insights on few individuals to raw inferences for the population, respectively. Previous works showed that both data sources can be fitted within the common statistical framework of Inhomogeneous Poisson Point processes (IPP). Building on that, we developed an integrated model that provides a joint estimation of habitat selection for telemetry data fitted with Resource Selection Function (RSF) and count data fitted with a Poisson Generalized Linear Model (GLM), respecting the statistical conditions for converging with an IPP. We tested our integrated model using movement simulated data and Sandwich Tern (Thalasseus sandvicensis) data collected in the French Mediterranean Sea. Simulations showed that the integrated model correctly estimated habitat selection coefficients. Importantly, when the amount of data is limited, the integrated model benefited from both data sources and had a better accuracy and precision than RSF and Poisson GLM in isolation. Overall, our study formalized the integration of telemetry and count data to estimate habitat selection, contributing to a promising research avenue since telemetry and count monitoring are abundant in many ecological contexts..
Authors: Teresa Morán López (University of Oviedo)+; Roquer-Beni Laura (CREAF, E08193 Bellaterra (Cerdanyola del Vallés) Spain); Jordi Bosch (CREAF, E08193 Bellaterra (Cerdanyola del Vallés) Spain); Peter Hambäck (Department of Ecology, Environment and Plant Sciences, Stockholm University); Alexandra-Maria Klein (Chair of Nature Conservation and Landscape Ecology, University of Freiburg); Marcos Miñarro (Servico Regional de Investigación y Desarrollo Agroalimentario. Villaviciosa); Ulrika Samnegard (Department of Ecology, Environment and Plant Sciences, Stockholm University); Otso Ovaskainen (University of Jyväskylä); Daniel García (University of Oviedo).
Session: S-4-4.
Where: G037.
When: 14:30-16:30.
Abstract: Agriculture sustainability represents one of the greatest humankind’s challenges today. In this context, ecological intensification of agriculture represents a win-win strategy to reconcile biodiversity conservation and food production. By fostering biodiversity services to agriculture (e.g., pollination), we can reduce environmental footprint without compromising crop yields. Nonetheless, designing effective ecological intensification practices is often a challenging task. Management effects depend on the ecology of individual species, but such information is often unavailable at spatial scales relevant in agroecosystems (landscape or regional scales). Consequently, services are frequently modeled as a function of community-level metrics (e.g., pollinator functional richness), which may not reflect the re-arrangement of individual species or their functionality. If service providers differ in their response to management actions (due to their environmental niche) or in their functional effects (due to their abundance or capabilities to deliver services), community metrics may not be sensitive enough. We propose to model service provision to agroecosystems using the response-effect trait framework by means of HMSC models (Hierarchical Modeling of Species Communities). We estimate species distribution based on their “response” traits and their per capita contribution to services as a function of their “effect” traits. Then, the magnitude of services is calculated as the product of species-specific abundances and their per capita contribution. By retaining information at the species level (from landscape distribution to functional contribution), we attain accurate estimates of ecological intensification effects on service provision. In addition, since all ecological processes are modeled as a function of species traits, HMSC-Ser yields reliable estimates in out-of-sample data. Something particularly relevant when mapping ecosystem services at regional scales, which may depend on a variety of species from the regional pool. We will present the HMSC-Ser model and test its performance with in silico experiments, and illustrate its suitability with pollination data in apple orchards..
Authors: Karunarathna K. A. N. K. (UQ Spatial Epidemiology Laboratory, School of Veterinary Science, Faculty of Science, The University of Queensland, Queensland, 4343, Australia)+.
Session: S-4-3.
Where: G043.
When: 14:30-16:30.
Abstract: Ecologists are increasingly gathering data across spatial and temporal domains with the aim of generating both reliable inference and accurate forecasts. But tackling spatiotemporal dependence is challenging, and most models that ecologists employ assume either the independence of spatial and temporal processes or make use of multidimensional splines. Neither of these approaches can produce reliable out of sample predictions because they do not well represent dependency structures present in real systems. The field of spatiotemporal statistics offers some solutions, for example by demonstrating how spatial basis functions with time-varying coefficients can approximate spatiotemporal covariance structures and produce accurate forecasts. However, modelling the time series of basis functions demands significant computational resources, making them unappealing to practicing ecologists. In this study, we propose a new approach by employing low-rank dynamic factor models to capture the temporal evolution of spatial basis coefficients. Further, we demonstrate the use of low-rank Gaussian Process factors to ensure smoothness across both spatial and temporal dimensions. Using a series of simulations, we show that our approach produces reliable inferences and out of sample predictions over large spatiotemporal domains. By estimating model parameters using Bayesian inference with Hamiltonian Monte Carlo (HMC) simulation in Stan, our approach can be readily incorporated into flexible Generalized Additive Models opening new doors for the analysis of complex spatiotemporal ecological datasets. .
Authors: Eiren K Jacobson (Centre for Research into Ecological and Environmental Modelling, University of St Andrews)+; Mark Bravington (Estimark Research); David Miller (NA); Irina Trukhanova (North Pacific Wildlife Consulting, LLC, Seattle, Washington and US Fish and Wildlife Service, Marine Mammals Management, Anchorage, Alaska); Rebecca Taylor (US Geological Survey, Alaska Science Center, Anchorage, Alaska); William Beatty (US Geological Survey, Alaska Science Center, Anchorage, Alaska and US Geological Survey, Upper Midwest Environmental Sciences Center, La Crosse, Wisconsin).
Session: S-1-4.
Where: G029.
When: 14:00-15:00.
Abstract: The Pacific walrus (Odobenus rosmarus divergens) is an ice-associated marine mammal found in the Bering and Chukchi Seas, where they have been hunted for subsistence for time immemorial. In the late 20th century, the population declined, likely because it had reached carrying capacity and was subject to high harvests. Currently, Pacific walrus is species of conservation concern due to the potential impacts of climate change, particularly related to loss of sea ice. To reduce uncertainty in estimates of population size and trend, the FWS and USGS undertook an individual genetic mark-recapture (IMR) sampling campaign from 2013-2017 and collected tissue samples from over 8,000 individuals. Another campaign of a similar scale is ongoing (2023-2027). While sample collection was designed for IMR, methodological advances in close-kin mark-recapture (CKMR) and associated molecular methods (particularly epigenetic ageing) mean these samples could also be suitable for CKMR. The advantages of CKMR over IMR include increased effective sample size (since each individual tags not only itself, but also its parents, siblings, and offspring) and additional insights into demographic quantities of interest. Here, we combine individual and close-kin mark-recapture in a single modelling framework (ICKMR) and investigate whether different sampling strategies can increase precision in estimates of abundance and trend. Our modelling approach includes special considerations for walrus life-history, including a multi-year inter-birth interval. We implement our model in R and TMB and use an individual-based simulation to test performance of the ICKMR model when assumptions are violated. We find that the expected precision of the ICKMR estimates of abundance are higher than those expected from IMR alone. This result suggests that ICKMR is a promising approach for assessing population size and trend of species which have been difficult to survey using more traditional methods..
Authors: Sanet Hugo (University of Venda)+.
Session: S-6-3.
Where: G029.
When: 10:30-12:30.
Abstract: Indicator Value Analysis remains a popular method for identifying indicator species; however, it can be influenced by imperfect detection. Urban et al. (2012) demonstrated that unbiased indicator values (IndVals) can be obtained from the posterior distribution of abundances estimated by N-mixture models. To date, this proposed solution has been disregarded. I build on this idea by using simulated and real data to test occupancy, N-mixture and Royle-Nichols models that were modified in JAGS to include the relevant occurrence-based or abundance-based IndVal formulas. Real data were counts of selected ant species from a long-term (> 5 years) pitfall-trapping transect covering 44 sites across both aspects of the Soutpansberg mountain range in the UNESCO Vhembe Biosphere Reserve, South Africa. Simulated data were based on this transect design, and bias related to pitfall-level variables was introduced to the simulated observations. Site groups for IndVal analyses were defined by five vegetation types across the transect. IndVals supplied by hierarchical models were close to ‘true’ IndVals of the simulated data but differed substantially from ‘naïve’ values supplied by classical IndVal analysis of the simulated observations. For real data, hierarchical models that account for significant detection bias supplied IndVals that differed substantially from classical IndVal analysis. For example, Ocymyrmex fortior occurs predominantly in the woodlands, and observations were negatively related to maximum monthly temperature and canopy cover. An occupancy model and a classical IndVal analysis supplied woodland IndVals of, respectively, 76% and 64%. A 70% threshold is typically used to choose indicator species. Accounting for detection bias can therefore assist in choosing appropriate indicator species. Further studies on multi-species and dynamic hierarchical models may discover additional improvements to IndVal analysis. Urban, N.A., Swihart R.K., Malloy, M.C., Dunning, J.B. (2012) Improving selection of indicator species when detection is imperfect. Ecological Indicators, 15, 188–197..
Authors: Oscar Rodriguez de Rivera Ortega (University of Exeter)+.
Session: S-4-3.
Where: G043.
When: 14:30-16:30.
Abstract: In recent years, various regions worldwide, including South America, southern and western Europe, and even unexpected areas above the Arctic Circle such as Sweden, have experienced a notable increase in extreme wildfires. Particularly, the Mediterranean region has witnessed a surge in larger and more frequent fires, exemplified by significant incidents such as the approximately 200 thousand hectares burned in Portugal in mid-October 2017, and 65 thousand hectares in two fires in the same area in Spain in June 2022. This trend is attributed to socioeconomic shifts leading to rural depopulation and alterations in traditional land use, coupled with prolonged drought periods and heightened flammability of dominant species.
Despite substantial efforts to elucidate the impact of climate change on natural hazards, such as forest fires, and to develop models for characterizing and quantifying changes in climatic patterns, understanding the relative contribution of human factors in shaping wildfire occurrences remains an ongoing challenge. In this presentation, I will discuss two distinct spatio-temporal methodologies applied to ascertain causality and forecast the distribution and intensity of wildfires. Specifically, I will showcase the application of disease mapping models and marked point process models, which leverage environmental and socio-economic variables from differing perspectives. The former approach involves aggregating information within administrative areas to predict fire occurrences, while the latter examines the spatial distribution and intensity of individual fires, integrating environmental and socioeconomic data at various scales.
This presentation will conclude by highlighting the critical role of scale selection in spatial models. We will demonstrate that this choice is not only consequential for model accuracy but also for effective communication of the model’s outputs. .
Authors: Fiona M Seaton (UK Centre for Ecology & Hydrology)+; Susan Jarvis (UK Centre for Ecology & Hydrology); Pete Henrys (UK Centre for Ecology & Hydrology).
Session: S-4-2.
Where: G049.
When: 14:30-16:30.
Abstract: Species distribution modelling is a highly used tool for understanding and predicting biodiversity change, and recent work has emphasised the importance of understanding how species distributions change over both time and space. Spatio-temporal models require large amounts of data spread over time and space, and as such are clear candidates to benefit from model-based integration of different data sources. However, spatio-temporal models are highly computationally intensive and integrating different data sources together can make this approach even more unfeasible for ecologists. Here we demonstrate how the R-INLA methodology can be used for model-based data integration for spatio-temporally explicit modelling of species distribution change. We demonstrate that this method can be applied to both point and areal data with two contrasting case studies, one using the SPDE approach for modelling spatio-temporal change in the Gatekeeper butterfly (Pyronia tithonus) across Great Britain and the second using a spatio-temporal areal model to describe change in caddisfly (Trichoptera) populations across the River Thames catchment. We show that in the caddisfly case study integrating together different data sources led to greater understanding of the change in abundance across the River Thames both seasonally and over five years of data. However, in the butterfly case study moving to a spatio-temporal context exacerbated differences between the data sources and resulted in no greater ecological insight into change in the Gatekeeper population. Our work provides a computationally feasible framework for spatio-temporally explicit integration of data within SDMs and demonstrates both the potential benefits and the challenges in applying this methodology to real ecological data..
Authors: Benjamin Rosenbaum (iDiv)+.
Session: S-5-1.
Where: G037.
When: 10:30-12:30.
Abstract: Multi-species functional responses (MSFR) describe the feeding rates of consumers that interact with multiple resource species. They are the backbone of dynamic foodweb models which predict biomass changes of communities over time. However, empirical support for their specific usage and parameterization is lacking, largely due to the rarity of multi-species feeding experiments, which pose logistical challenges and require non-trivial statistical analysis.
One key issue is the depletion of prey over the course of the experiment, which must be statistically accounted for correctly. We have developed a novel approach that uses a population dynamics model to generate predictions (by numerically simulating differential equations) and fits them to data, allowing for the accurate estimation of model parameters such as attack rates, handling times, and prey preferences. This method has already been proven valuable for single species functional responses, and our study shows that it can be successfully extended to MSFRs.
To test this approach, we performed an extensive simulation study that crossed several classes of MSFR models, experimental designs (combinations and amount of available prey), sample sizes and noise levels. Specifically, we simulated realistic feeding trial data with known parameters using a stochastic algorithm, fitted population dynamic MSFR models, and evaluated the accuracy of estimated model parameters. Results showed an overall good performance and the investigated MSFRs were identifiable from the data. We found that sample size was more important than experimental design. Notably, analyzing the combination of all 2-species mixtures was sufficient to estimate all model parameters accurately in systems with 3 prey species, eliminating the need for trials with more than 2 prey species and simplifying the logistics of such experiments.
We propose a framework for estimating MSFRs from data by combining dynamical prediction models with Bayesian inference. This study offers guidelines for empiricists who want to perform MSFR experiments, and opens up new possibilities for testing hypotheses on topics such as prey preference and switching, optimal and adaptive foraging, and foodweb stability..
Authors: Célian Monchy (CEFE-CNRS)+; Marie-Pierre Etienne (Institut de recherche mathématique de Rennes); Olivier Gimenez (CNRS).
Session: S-6-3.
Where: G029.
When: 10:30-12:30.
Abstract: Passive sensors like camera-traps and autonomous recording units for acoustic have become important non-invasive monitoring techniques to collect environmental data with the objective to better understand species distribution. Those new data have fueled the development of statistical models to suit with specific sampling designs and get reliable ecological inferences. Among others, site occupancy models enable the estimation of occurrence patterns for one or more target species, considering that their presence may be observed or not. In that context, it is useful to distinguish a) the detection process, which is unobserved and corresponds to the situation where the species of interest leaves signs of its presence (by triggering a camera-trap for example), and b) the identification process, which allows to recognize the species of interest from its signs of presence. Detection and identification are both imperfect processes in which false-negative and false-positive errors may occur, especially when collected data are massive and required automated treatment. Misclassification at both steps can lead to significant biases in ecological estimates, and several extensions have been proposed to correct for these potential errors. The naïve model suffers from identifiability issues, hence the existence of alternative models that combine several data sources to allow the identification of error parameters. These models however are more data-hungry and require to include a source of data without false-positives. As an alternative, we propose to reuse available data from the identification process in a Bayesian framework through an informative prior, in order to overcome identifiability issue and reduce estimation bias. We compare through a simulation study these different approaches considering various monitoring designs. Overall, what is at stake is enhancing statistical methods together with sampling non-invasive technologies, in a way to provide ecological outcomes suitable for conservation decision-making..
Authors: Ying-Chi Chan (Swiss Ornithological Institute)+; Matthias Tschumi (Swiss Ornithological Institute); Fränzi Korner-Nievergelt (Swiss Ornithological Institute); Martin Gruebler (Swiss Ornithological Institute).
Session: S-1-2.
Where: G049.
When: 14:00-15:00.
Abstract: The lunar cycle has been shown to influence animal behaviours and key life history events, yet how it affects dispersal timing remained underexplored. During natal dispersal of nocturnal animals, prospecting of an unfamiliar landscape could be facilitated by moon illumination. We used radio-tracking data of natal dispersal movements of 126 juvenile little owls (Athene noctua), a nocturnal bird of prey, to explore the effect of the lunar cycle on natal dispersal timing. To deal with data gaps caused by variations in field effort, we fitted a Bayesian hierarchical multi-state model which account for imperfect detection and mortality, and movement is modelled as transition between states. We found that the probability of starting natal dispersal is highest around full moon. Females, which tend to disperse further, also departed earlier than males, at the phase before full moon (‘waxing gibbous’). Little owls tend to disperse during the full moon, likely because this timing enhances their ability to search for suitable habitats and mates..
Authors: Wei Zhang (University of Glasgow)+; Oliver Stoner (University of Glasgow).
Session: S-6-1.
Where: G043.
When: 10:30-12:30.
Abstract: In Bayesian statistics, Markov chain Monte Carlo (MCMC) methods are widely used to sample from target posterior distributions. MCMC methods are appealing due to their general applicability, however these may be time-consuming for complicated or large-scale hierarchical models. In this talk, I will introduce a new approximate MCMC sampling strategy that includes implementing a Gaussian approximation to the conditional posterior distribution for part of a model within MCMC sampling. I will demonstrate that this strategy can greatly improve the computational efficiency for many complicated models, compared to common MCMC samplers like the multivariate random-walk and slice samplers. I will present results on some examples with spatial components and/or non-linear relationships modelled through splines, including a spatial capture-recapture example..
Authors: Michael A Spence (Cefas)+.
Session: S-2-4.
Where: G029.
When: 15:30-16:30.
Abstract: Mechanistic ecosystem models, based on describing ecological processes rather than statistical correlations, can used to qualitatively predict to the long-term affects. Due to the robustness of mechanism, they can describe the effects of phenomena that have not occurred such as long-term effects of climate change. However, due to the complexity of the ecosystem, they are not very good at quantitative or short-term predictions, and therefore cannot be used for tactical advice. On the other hand, statistical models, such as single-species assessment models, that describe the system using statistical correlations are much better at describing the current state of the system and forecasts on a shorter-time scale, making them ideal for tactical advice. However, they often ignore more interactions with the ecosystem that can happen over longer time periods, such as multispecies and climate effects and so the long-term predictions are not as robust.
Combining these different classes of models in a coherent way is a difficult problem. In this talk I will introduce a framework that combines predictions from different types of models to make predictions across different time scales, potentially providing a basis for coherent tactical and strategic advice. I demonstrate the method using five stocks in the North Sea along with four ecosystem models to make predictions seamlessly across different timescales. Keywords: Ecosystem-based fisheries management, ensemble model, strategic advice, tactical advice, single-species models, ecosystem models .
Authors: Milly Jones (University of Kent)+.
Session: S-8-3.
Where: G037.
When: 11:00-12:45.
Abstract: Occupancy studies rely on temporal or spatial replication to estimate probabilities of species occupancy at surveyed sites, whilst accounting for observation error. If a site does not permit independent spatial replicates, line transects divided into segments are used as the replicates instead. In this case, an observer travels along the transect at each site and records a detection/non-detection at each of the segments. This approach is often employed for monitoring large carnivores, such as the case study on tigers in Indonesia that motivated this work. However, observations from consecutive segments are often not independent due to the tendency for individuals to use (or partly use) established trails to travel. Species presence on adjacent segments is often therefore correlated, so that detections are clustered. Models that incorporate Markovian dependence on segment occupancy, so that the probability a segment is occupied depends on the occupancy status of the previous segment, are now widely used. However, in practice, these models often produce unrealistic parameter estimates, as is the case for the motivating case study. In this talk, we investigate the identifiability of these models under a range of different parameter scenarios. Through simulation, we compare different data collection methods to propose alternatives that prove identifiable when these standard models are not. We consider both discrete (detection/non-detection) and continuous (distance to next detection) observations along with survey design considerations to account for effort levels in data collection in practice..
Authors: Abdulmajeed Alharbi (Sheffield University)+.
Session: S-5-4.
Where: G029.
When: 10:30-12:30.
Abstract: Ultra-high-resolution biologging has the potential to revolutionise our understanding of animal behaviour. However, current methods for analysing animal movement often fall short when data points are closely spaced in time, limiting the effectiveness of existing statistical approaches. Specifically, the conventional focus on displacement between recorded locations and changes in movement direction proves inadequate at such fine scales.
In response to this challenge, it is essential to develop techniques capable of identifying deviations from straight-line trajectories within the movement data. Detecting the time and location at which an animal turns from recorded trajectories poses a non-trivial problem, crucial for subsequent analyses. We have developed a likelihood-based method for inference that is tailored to identify deviations from straight-line trajectories within high-resolution movement data, which is fast computationally, accurate, and well-grounded statistically .
Authors: Marine Ballutaud (La Rochelle Université)+; Mathieu Doray (Ifremer Nantes); Maxime Olmos (Ifremer Brest); Matthieu Authier (La Rochelle Université).
Session: S-7-2.
Where: G043.
When: 14:30-16:30.
Abstract: Understanding the interactions between marine species and human activities is critical to implement conservation measures. In the Bay of Biscay, common dolphin (Delphinus delphis) by-catches during winter have increased in recent years but the underlying causes remain largely unknown. An hypothesis is a change in spatial distribution with common dolphins occurring closer to the coast to feed on small pelagic fish stocks such as sardines and anchovies, leading to increased interactions with fisheries. We developed a simulation framework to investigate the spatio-temporal overlap between predator (dolphins) and prey (small pelagic fish) with the ultimate aim of shedding light on the mechanisms behinds dolphins by-catch. Our framework includes two modelling approaches. First, we built a simulation model of predator-prey distribution where different parameters values can be associated to different spatial interactions between simulated populations of predator and prey. In this simulation model we represent the density distribution of dolphin as the sum of the spatial variations in prey at local and global scales, the spatial variation in dolphin not determined by predation on small pelagic fish, and a spatial random effect. Then, we used this model as an estimating model on observed data of small pelagic fish and dolphins in the Bay of Biscay. Our simulation-estimation approaches allowed us to investigate the sensitivity of various predator-prey overlap metrics and conclude about the ecological mechanisms underlying dolphins by-catch. Our results show that the distribution of small pelagic fish has shifted from offshore to inshore over the last five years and that the distribution of dolphins has followed the distribution of their prey, increasing the probability of interaction with small pelagic fish fisheries..
Authors: William P Kay (Cardiff University)+; Crispin Jordan (University of Edinburgh); Nicola Romano (University of Edinburgh); Kasia Banas (University of Edinburgh); Vanessa Armstrong (Newcastle University); Jenny Terry (University of Sussex).
Session: S-6-3.
Where: G029.
When: 10:30-12:30.
Abstract: Statistical analysis lies at the heart of inference for the Life Sciences, including Ecology. Yet, a reproducibility crisis across multiple disciplines questions whether, and how, the teaching of statistics should change. We approached UK universities to survey what, how much, and how statistics and experimental design are taught across four Life Sciences disciplines: Biology, Biomedical Sciences, Psychology, and Medicine. Our survey sought to determine which topics are under-taught, and whether disciplines differ in ways that allow them to learn from each other. In this talk we will share our results, highlighting the contrasts and similarities between how study design and analysis is taught in Biology (encompassing Ecology) compared to other disciplines. For example, Psychology provided more “stand-alone” courses, while other disciplines integrated topics into other teaching, and the survey revealed diversity in whether statistics was taught from a mathematical or conceptual perspective, highlighting opportunities to identify approaches that mitigate ‘Statistics Anxiety’. All disciplines typically taught core statistical tests (e.g., linear models) but differed with respect to teaching more advanced techniques (e.g., mixed models). Strikingly, most disciplines tended to fail to teach topics linked to the reproducibility crisis, including pseudo-replication and power analysis. Further, all disciplines often did not teach “Questionable Research Practices” (e.g., HARKing, p-hacking). Similarly, all disciplines often omitted topics related to Open Science. Our results highlight topics that require more teaching across all disciplines, and opportunities for disciplines to learn from each other to improve training in study design and analysis. We suggest that courses switch emphasis from teaching “experimental design and analysis” to teaching “skills required to produce reliable research” and support undergraduates as future researchers. This talk is designed to stimulate thought among delegates at ISEC 2024 and prime our roundtable discussion on: “What are the key statistical methods and concepts that all [undergraduate] ecologists should learn?”..
Authors: Yan Ru Choo (University of St Andrews)+; Alison Johnston (University of St Andrews); Chris Sutherland (University of St Andrews).
Session: S-3-3.
Where: G049.
When: 10:30-12:30.
Abstract: Spatial capture-recapture (SCR) models provide estimates of animal density from spatially referenced encounter data and has become the most widely-adopted approach for estimating density. Despite rapid growth in the development and application of SCR, approaches for assessing model fit have received relatively little attention. In this talk, we present a Monte Carlo resampling approach for assessing goodness-of-fit for SCR models fit using maximum likelihood. We use the point estimates and variance-covariance matrix of the fitted SCR model to resample SCR parameter values. For each set of resampled parameters, we derive probability distributions of individual activity centre locations from which we draw realisations of their activity centres. This process allows us to propagate the uncertainty of the estimates and activity centre locations via Monte Carlo simulation. Using resamples of parameter estimates and activity centers, we calculate the expected encounter probability for each individual in each trap, and calculate discrepancy measures for simulated and observed capture histories using Freeman-Tukey tests. We simulate capture histories and fit them to two sets of SCR models: one assuming no heterogeneity, and another where the model follows the data-generating process in the simulation. Comparing observed and expected encounters at specific traps was found to be the most powerful test when detection varied spatially. Whereas comparing observed and expected encounters of individual animals across traps was more powerful when the simulated detection variation was not spatially structured. Through simulations we demonstrate that these tests are able to detect lack-of-fit when unmodelled heterogeneity in the detection sub-model is present. We also show that our approach is consistent with analogous Bayesian GoF measures thereby providing a consistent means of assessing model fit regardless of inference paradigm..
Authors: Philip S Mostert (Norwegian University of Science and Technology)+; Kwaku Peprah Adjei (Norwegian Institute for Nature Research); Ron R. Togunov (Norwegian University of Science and Technology); Sam Perrin (Norwegian University of Science and Technology); Robert B O’Hara (NTNU); Anders Finstad (Norwegian University of Science and Technology); Joseph Chipperfield (Norwegian Institute for Nature Research).
Session: S-1-1.
Where: G043.
When: 14:00-15:00.
Abstract: Integration of data is needed to address many of the current threats to biodiversity. There has been an exponential increase in quantity and type of biodiversity data in recent years, including presence-absence, counts, and presence-only citizen science data. Species Distribution Models (SDMs) are frequently used in ecology to predict current and future ranges of species, and are a common tool used when making conservation prioritisation decisions. Current SDM practice typically underutilizes the large amount of publicly available biodiversity data and does not often follow a set of standard best practices. Integrating data types with open-source tools and reproducible workflows saves time, increases collaboration opportunities, and increases the power of data inference in SDMs.
Here, we present a fully automated pipeline to construct large-scale integrated species distribution models, by obtaining species occurrence data and environmental covariates from publicly available repositories. Integration of the datasets is done in through a state-space point process formulation: which combines a process model, describing the actual distribution of the species, with data-specific observation models, describing the data collection process. We provide an R package, code, and guidance on how to accommodate users’ diverse needs and ecological questions with different data types available on the Global Biodiversity Information Facility (GBIF), the largest biodiversity data aggregator in the world. Lastly, we reflect on our experiences, outlining both strengths and weaknesses of our approach, and suggest avenues for future refinement. .
Authors: Mehnaz Jahid (University of Victoria)+.
Session: S-6-3.
Where: G029.
When: 10:30-12:30.
Abstract: Collection of genetic capture recapture data can be paired with another field data collection method that collects presence-absence data, such as camera traps. Since the individual identification might not be possible for all the data captured in the first source, the second source can offer valuable information about the species. Both these data can be used in integrated modelling to estimate wildlife density and usually this method provides estimates more precise than single data source estimates. However, fulfilling the assumption that the data sources need to be independent often restrict this application since these data sources are certainly not independent. We formulated a model which can accommodate information from both capture recapture and presence-absence data to estimate wildlife abundance when these data sources are dependent. This model can accommodate more than one season by considering parameters for survival, fecundity, and movement. Capture recapture data are modelled with a spatially explicit capture recapture approach while the presence-absence data are modelled conditional on the latent capture recapture information. We applied this model to estimate the population density of grizzly bears of the central Rocky Mountains of Alberta, Canada using capture recapture data from hair traps and presence-absence data from camera traps. We also applied this data to two other available integrated models to compare the bias and precision of the estimates..
Authors: Laura Cowen (University of Victoria)+; Kehinde Olobatuyi (University of Victoria); Matthew Parker (Simon Fraser University).
Session: S-7-3.
Where: G037.
When: 14:30-16:30.
Abstract: Extended batch-mark models (EBMMs) are applied to small animals when unique identifiers cannot be used, forbidding the use of capture-recapture methods. We have made EBMM developments in the form of 1) extBatchMarking: a new R package that implements the hidden Markov model of Cowen et al (2017) and 2) a novel Bayesian EBBM. In the Bayesian model, we address issues of computational tractability and remove the need for state binning when dealing with large hidden populations. Avoiding state binning improves computational precision and computational efficiency, allowing analysis of more complex model structures with parameter covariates. We incorporate both time-varying and batch-varying covariates. A comparative analysis between the extBatchMarking R package and our newly developed EBMM shows that our proposed approach outperforms previous methods in terms of compute time, and provides similar nested model ordering for model selection purposes. We applied our new methods to study a Pacific sand lance (Ammodytes personatus) population in British Columbia..
Authors: Arjun Banik (University of Victoria)+; Laura Cowen (University of Victoria); Saman Muthukumarana (University of Manitoba).
Session: S-7-3.
Where: G037.
When: 14:30-16:30.
Abstract: Recent advancements in the Cormack-Jolly-Seber (CJS) model, used for analyzing mark-recapture data, have primarily focused on accommodating variations in capture and survival rates among individuals. While several methods have emerged to link capture and survival probabilities to auxiliary variables—ranging from discrete to time-constant to population-wide, and even individual continuous covariates that evolve over time—the challenge remains unresolved for continuous covariates that vary over both time, individual, groups, and states. Our study focuses on examining the impact of fork length and reproduction (breeding state) on the survival of the anadromous Northern form of Dolly Varden, Salvelinus malma malma. We introduce a novel Bayesian multi-state mark-recapture model adept at handling such covariates. This model incorporates temporal and sex-based changes via a diffusion process, and it establishes connections between covariates and capture and state transition rates using logistic functions. We validate the model’s performance through simulations and subsequently apply it to real-world Dolly Varden data collected from five river systems in the Northwest Territories, Canada. However, we acknowledge that parameter identifiability may occasionally be compromised due to the complex life history of Dolly Varden and the high scarcity of data..
Authors: Simon Bonner (University of Western Ontario)+; Alex Draghici (University of Western Ontario).
Session: S-2-1.
Where: G037.
When: 15:30-16:30.
Abstract: A key assumption of the Cormack-Jolly-Seber model is that individuals have independent fates with respect to both their capture and survival. This assumption may be violated in different ways, and standard errors will be underestimated if the dependence is ignored. A common solution is to inflate the standard errors by a correction factor that is estimated from the data. However, Draghici (2020) showed that existing correction factors fail to address the underestimated standard errors when there is dependence within mated pairs. Moreover, it may be of interest to make inference about the degree of dependence itself as well as to account for the dependence when estimating other parameters. We address these problems by developing a specific approach to study dependence in both the capture and survival of mated pairs. Furthermore, we propose an alternative correction factor that can be used to correct standard errors and confidence intervals to account when this form of dependence is present. Following initial attempts to implement a complete model of the mating process, we have developed a simple, multi-stage approach to inference that relies on conditional inference. Our method is easy to implement, fast to compute, and avoids specific assumptions about how mates are selected. We present results of a simulation study to demonstrate the validity of our approach and to assess the power to test hypotheses regarding the correlation in both survival and capture. We also present results from the analysis of data obtained from a 28-year study of Harlequin ducks in Alberta, Canada, a long-lived waterfowl known to form monogamous pair bonds..
Authors: Bastien Mourguiart (IFREMER)+.
Session: S-2-3.
Where: G049.
When: 15:30-16:30.
Abstract: Species distribution models (SDMs) are extensively used to estimate species–environment relationships (SERs) and predict species distribution across space and time. For this purpose, it is key to choose relevant spatial grains for predictor and response variables at the onset of the modelling process. However, environmental variables are often derived from large-scale climate models at a grain that can be coarser than the one of the response variable. Such area-to-point spatial misalignment can bias estimates of SER and jeopardise the robustness of predictions. We used a virtual species approach, running simulations across different levels of area-to-point spatial misalignment to seek statistical solutions to this problem. We specifically compared accuracy of SER estimates and predictive performances, assessed across different degrees of spatial heterogeneity in environmental conditions, of three SDMs: a GLM, a spatial GLM and a Berkson error model (BEM) that accounts for fine-grain environmental heterogeneity within coarse-grain cells. Only the BEM accurately estimates SER from relatively coarse-grain environmental data (up to 50 times coarser than the response grain), while the two GLMs provide flattened SER. However, all three models perform poorly when predicting from coarse-grain data, particularly in environments that are more heterogeneous than the training conditions. Conversely, decreasing environmental heterogeneity relative to the training dataset reduces the predictive biases. Because predictions are made from covariate-grain data, the BEM displays lower predictive performance than the two GLMs. Thus, standard model selection methods would fail to select the model that best estimates SERs (here, the BEM), which could lead to false interpretations about the environmental drivers of species distributions. Overall, we conclude that the BEM, because it can robustly estimate SER at the response grain, holds great promise to overcome area-to-point misalignment..
Authors: Simon Bonner (University of Western Ontario)+; Wei Zhang (University of Glasgow); Jiaqi Mu (University of Western Ontario).
Session: S-5-2.
Where: G049.
When: 10:30-12:30.
Abstract: Continuous predictors of survival present a challenge in the analysis of data from studies of marked individuals. If the predictor varies over time and between individuals, like most measures of individual fitness, then its value can only be observed when individuals are captured. Existing methods to study the effects of such variables have followed one of two approaches. The first is to model the joint distribution of the predictor and the observed capture histories, and the second is to draw inference from the likelihood conditional on events that depend only on observed predictor values, called the trinomial model. Previous comparison of these approaches found that joint modelling provided more precise inference about the effect of the covariate while the trinomial model was less prone to issues of model mis-specification. However, we believe that an important issue was missed. We show through mathematical analysis and numerical simulation that the trinomial model is not identifiable when the predictor has no effect on the survival probability. Perhaps more importantly, this causes inferences from the trinomial model to be imprecise when the effect of the covariate on the survival probability is small. We provide further demonstration of the importance of this issue in real applications through an analysis of the effect of body mass on the survival of meadow voles..
Authors: Frances Buderman (Pennsylvania State University)+; Ephraim M Hanks (Penn State University); Viviana Ruiz Gutierrez (Cornell Lab of Ornithology); Michael Schull (Pennsylvania State University); Robert Murphy (Eagle Environmental, Inc.); David Miller (Pennsylvania State University).
Session: S-4-2.
Where: G049.
When: 14:30-16:30.
Abstract: While the quantity, quality, and variety of movement data have increased, methods that allow for jointly estimating population- and species-level movement parameters are still needed. We present a formal integrated movement models (IMMs) approach to combine individual-level movement and population-level distribution data. Our IMM includes a model for individual movement, a model for among-individual heterogeneity, and a model to quantify changes in species distribution; these models are then formally linked through shared parameters. We outline a general IMM framework and develop and apply a specific stochastic differential equation model to a case study of telemetry and species distribution data for golden eagles in western North American during spring migration. We estimated eagle movements during spring migration based on data collected during 2011 - 2019. Individual heterogeneity in migration behavior was modeled for two sub-populations, individuals that made significant northward spring migrations and those that remained in the southern Rocky Mountain region through the summer. As in most tracking studies, the sample population of individual telemetered birds was not representative of the population, and underrepresented the proportion of long-distance migrants in. The IMM provided a more biological accurate subpopulation structure by jointly estimating the structure, using the species distribution data. In addition, the integrated approach a) improved accuracy of other estimated movement parameters, b) allowed us to estimate the proportion of migratory versus non-migratory birds in a given location and time, and c) estimated future spatio-temporal distributions of birds in a given a wintering location, from which seasonal connectivity and migratory routes can be derived. Our approach can be generalized to a broad range of available movement models and data types, allowing us to significantly improve our knowledge of migration ecology across taxonomic groups, and address population- and continental-level information needs for conservation and management..
Authors: Brett T McClintock (NOAA-NMFS AFSC Marine Mammal Laboratory)+.
Session: S-3-4.
Where: G037.
When: 10:30-12:30.
Abstract: Since its release in 2018, R package momentuHMM has become a popular tool for analyzing biotelemetry (e.g., location, accelerometer, dive activity) data using discrete-time hidden Markov models (HMMs). HMMs are particularly useful for inference when animals exhibit distinct movement behavior states (e.g., resting, foraging, migrating), but applications tend to characterize animal movement as a time series of steps and turns that is blind to the spatial mechanisms that give rise to certain behaviors. Discrete-time HMMs also require regularly sampled data, which can be difficult to accomplish in field studies. Here we describe recent extensions and developments that have expanded the breadth of movement models that can be implemented in momentuHMM version 2.0, while also providing faster and more stable optimization during model fitting. New features include:
• Continuous-time HMMs for irregularly sampled data
• Spatially-explicit HMMs based on position (instead of steps
and turns)
• Potential functions describing drift along environmental
gradients
• Langevin diffusions for inferring behavior-specific resource
selection and utilization distributions
• Ornstein-Uhlenbeck processes with unknown center(s) of
attraction
• Recharge dynamics models for aggregated physiological
processes
• Hierarchical HMMs for multi-scale behavioral processes
• Model fitting using R package TMBThese features are largely integrated into existing functions, thereby making their implementation straightforward to users already familiar with the package. momentuHMM 2.0 also provides new utility functions to aid in data preparation and model specification, including the calculation of environmental gradients from raster covariate data. Functions are also provided for simulating and assessing the fit of any given model. We present several example applications demonstrating these new features, with many more to be found in the package vignette. We hope momentuHMM 2.0 will help make these more advanced methods accessible to practitioners, thereby motivating more realistic hypothesis-driven animal movement analyses..
Authors: Lena L Payne (University of Kent)+.
Session: S-4-3.
Where: G043.
When: 14:30-16:30.
Abstract: I will show how different environmental factors, and different spatial constraints can affect the spatial structure of herds of prey, and how the structure of these herds affect the potential longevity of the species.
I will also show how different metrics (pair correlation function, adjacency, average shortest distance, average distance in a herd) can vary in utility as a measure of the spatial structure depending on the initial conditions..
Authors: Yacob Haddou (University of Glasgow)+; Rebecca Mancy (University of Glasgow); Sofie Spatharis (University of Glasgow); Davide Dominoni (University of Glasgow); Jason Matthiopoulos (University of Glasgow).
Session: S-7-2.
Where: G043.
When: 14:30-16:30.
Abstract: The impact of environmental changes on species often unfolds gradually, with legacies of past landscapes influencing present-day population abundances. Despite this, Species Distribution Models (SDMs) typically rely on contemporary sampled data to estimate species-environment relationships, overlooking populations potentially being in a transient state between past and present environmental conditions. This neglect could result in biased inference, leading to inaccurate management decisions, yet these consequences have remained largely unexamined. In this study, we bridge temporal simulations of population dynamics with SDMs to assess how the strength of the estimated relationships between organisms and environment varies with past landscape changes. We investigate how inference and subsequent predictions of species abundances across space are affected by changes in environmental homogeneity and the correlation of change sizes. Additionally, we explore these dynamics in breeding bird species across the USA. Our findings reveal that populations influenced by legacy effects from past landscapes exhibit shrinkage or inflation in their estimated relationship with the current environment. Shrinkage, characterized by a weaker relationship magnitude than the truth, is associated with past land cover changes resulting in environmental homogenization, while inflation is caused by environments becoming more heterogeneous. We also found that lower correlation in the magnitude of changes leads to shrinkage. We highlight how inflated inference can result in spatial predictions of abundance where species appear rarer and more heterogeneously distributed than reality. Conversely, shrinkage leads to species seeming more common and widely distributed. Legacy effects from past land cover changes continue to influence present populations. We emphasize the importance of acknowledging these processes to avoid misinformed conservation policies and management strategies targeted at incorrect spatial scales..
Authors: Simon English (University of British Columbia)+; Peter Arcese (University of British Columbia); Amanda Rodewald (Cornell Lab of Ornithology); Scott Wilson (Environment and Climate Change Canada).
Session: S-4-3.
Where: G043.
When: 14:30-16:30.
Abstract: The human socioeconomic factors that underlie habitat loss for migratory birds are widely recognized as important barriers to the success of conservation initiatives. Factors including human population densities, accessibility by roads, local incomes, and illicit activities like narco-trafficking all contribute to habitat loss through land cover changes in forested ecosystems of Mesoamerica and the Neotropics. Despite their role in driving deforestation, these drivers of habitat loss have never been quantified in a unified, spatially-explicit framework. We therefore compiled time-series land cover change maps from 2016 to 2021 at a multinational scale spanning Mexico and Colombia, biophysical maps of agricultural suitability, human population density maps, and road density maps. Next, we geocoded records of human socioeconomics at a municipality scale for both countries, including incidence of poverty, median income, and the annual incidence of violent conflicts perpetrated by cartels to understand how and where human needs may conflict with conservation objectives. Because collecting specific data on illicit economic activity was not feasible, especially for rural municipalities, we used public records of the incidence of violent conflicts with cartels as a proxy for narco-trafficking activity. We intersected these data layers with biodiversity hotspot maps of Neotropical migratory birds and implemented a spatial autoregressive Markov model to disentangle the influence of these socioeconomic drivers of habitat loss on the probability of land cover change. The applicability of our model for conservation of birds is twofold: firstly, we are able to predict risk of habitat loss for migratory birds based on a suite of local socioeconomic variables in a unified and spatially explicit framework; secondly, we identify which human socioeconomic factors must be addressed most urgently in tandem with conservation action to the benefit of biodiversity and people..
Authors: Brandon PM Edwards (Carleton University)+; Marcel Gahbauer (Environment and Climate Change Canada); Alexis Grinde (University of Minnesota, Duluth); David Hope (Environment and Climate Change Canada); Elly Knight (Boreal Avian Modelling Project); Nicole Michel (National Audubon Society); Barry Robinson (Environment and Climate Change Canada); Péter Sólymos (University of Alberta); Joseph Bennett (Carleton University); Adam Smith (Environment and Climate Change Canada).
Session: S-8-4.
Where: G029.
When: 11:00-12:45.
Abstract: Accurate assessments of population status and trends rely on accounting for probability of detection, making it an important metric in conservation. Yet probability of detection, or detectability, is often unaccounted for or unavailable on a species level. Although detection probability methods and estimates exist for many North American birds, probability estimates are most accurate for common, easy-to-detect species. There is therefore a need for methodology that provides detection estimates for species that are rare or undersampled, which generally do not have sufficient data for accurate estimates. Here, we take advantage of similarities in detection probabilities among phylogenetically-related species and species with similar traits to estimate detection probabilities in seven species of rare North American birds, using a hierarchical Bayesian multi-species removal and distance models. For two species, Bicknell’s Thrush and Kirtland’s Warbler, we demonstrate our multi-species model’s ability to predict detection probabilities given little to no observation data, and compare these predictions to previous estimates found in the literature. We also demonstrate the multi-species models’ ability to provide more precise estimates of detectability for species where there is sufficient data. We recommend that this multi-species model be combined with survey-level covariates to augment previous estimates of detection probabilities in the NA-POPS database of detectability offsets for North American birds..
Authors: Louise McMillan (Victoria University of Wellington)+; Daniel Fernández (Universitat Politècnica de Catalunya–BarcelonaTech); Shirley Pledger (Victoria University of Wellington); Richard Arnold (Victoria University of Wellington); Ivy Liu (Victoria University of Wellington); Murray Efford (Otago University).
Session: S-3-4.
Where: G037.
When: 10:30-12:30.
Abstract: clustglm and clustord: R packages for clustering with covariates for binary, count, and ordinal data
We present two R packages for model-based clustering with covariates. Both packages can perform clustering and biclustering (clustering sites and species simultaneously, for example). Both use likelihood-based methods for clustering, which enables users to compare models using AIC and BIC as measures of relative goodness of fit.
The models implemented in both packages use linear predictor terms, and so look more like regression models than clustering models. This allows for the incorporation of regression-style covariates alongside clustering effects, and enables the use of models suitable for non-continuous data. Both clustglm and clustord can include the effects of numerical or categorical covariates alongside cluster effect, or can fit pattern-detection models that include individual-level effects alongside cluster effects. For example, when applied to presence/absence data, you can cluster sites and species while also taking into account any single-species effects, and any additional covariates.
clustglm implements techniques from Pledger and Arnold (2014) for handling binary and count data. It leverages the R function glm and can accommodate balanced and non-balanced designs. It provides the clustering equivalent of biplots, and also profile plots.
clustord handles ordinal categorical data, using techniques outlined in Matechou et al. (2016), Fernández et al. (2016) and Fernández et al. (2019). It can fit the proportional odds model or the ordered stereotype model, a more flexible model whose fitted parameters can reveal when two ordinal categories are effectively equivalent to each other.
We will illustrate the use of clustglm and clustord with a selection of ecological datasets..
Authors: Orin J Robinson (Cornell Lab of Ornithology)+; Tom Auer (Cornell Lab of Ornithology); Wesley M. Hochachka (Cornell Lab of Ornithology); Alison Johnston (University of St Andrews); Daniel Fink (Cornell University).
Session: S-2-2.
Where: G043.
When: 15:30-16:30.
Abstract: The use of population trend information has been vital to prioritizing conservation and management efforts, as well as for other applications. The North American Breeding Bird Survey (BBS) has provided trend information for North America’s avifauna at regional and national scales since 1966. The BBS relies on a highly structured protocol to minimize interannual variation during data collection. In contrast, the eBird-based trend estimates use a recently developed analytical technique to control for interannual variation among these less structured surveys. Here we evaluate the differences between these two sets of trend estimates, baises in each, and consider the complementarity of both approaches. We compare the trend estimates from each survey at the range-wide and Bird Conservation Region (BCR) level and investigate likely causes of any differences in the estimates. We show that while there is general agreement between estimates from the two surveys, there is a large amount of variation across species/region combinations, although with the majority of trend estimates from both surveys being non-significant. While there are differences in the trend estimates across species/region combinations, there is no consistent bias in the direction or magnitude of these differences. Nevertheless, we also show that some of the differences can be attributed taxonomy, habitat type and survey coverage within a species’ range..
Authors: Mohammad MF Farhadinia (DICE, University of Kent)+; Luciano Atzeni (Oxforrd); Jose Hernandez (Severtzov); Anna Yachmennikova (Severtzov); Viatcheslav Rozhnov (Severtzov); Maria Chistopolova (Severtzov); Minaev Alexander (Severtzov); Natalie Dronova (Severtsov Institute of Ecology and Evolution); Alim Pkhitikov (Institute of Ecology of Mountain Territories of the Russian Academy of Sciences); Paul Johnson (WildCRU, University of Oxford); David Macdonald (WildCRU, University of Oxford).
Session: S-4-1.
Where: G029.
When: 14:30-16:30.
Abstract: Understanding the behavioural characteristics and movement patterns of re-wilded animals is paramount, particularly in assessing the success of re-wilding initiatives. This study focused on comparing the behaviour of re-wilded Persian Leopards in Russia to their wild counterparts in Iran, offering insights into the adaptations and behaviours necessary for their successful establishment in new habitats. We employed Behavioural Change Point Analysis (BCPA) with selection procedures for model parameters and temporal lags to define switch points. Each break was analysed for Net Squared Displacement (NSD), with linear and non-linear models applied to delineate five distinct movement modes. For each mode, we characterised intrinsic average movement parameters (turning angle, speed, persistence velocity, turning velocity) and spatio-temporal characteristics (duration, NSD and distance walked). Our findings revealed significant disparities in movement parameters across modes and inter-mode differences within countries. Speed and persistence velocity of Russians leopards were consistently slower than Iranian animals. There was significant evidence for slower speed and persistence velocity during encamped states in Russian animals, unlike Iranian leopards, which were homogeneous for speed but variable across round-trip and wandering modes for persistence velocity. For spatio-temporal characteristics, we found significant intra-country variability. In Russia, all modes presented significantly higher NSD compared to encamped, duration was significantly higher only for ranging, and distance walked for ranging and round-trip modes. In Iran, we found higher NSD for ranging, significantly quicker round-trip and wandering modes, and non-variable distance walked. Russian animals travelled less in each state compared to Iranian leopards. Linear Discriminant Analyses conducted on intrinsic parameters and spatio-temporal attributes both significantly separated the two countries, suggesting different behavioural patterns. These results highlight the intricacies of animal movement ecology and emphasize the need for targeted conservation strategies that consider the behavioural ecology of re-wilded animals to ensure their successful integration and sustainability in new habitats..
Authors: Jeffrey Doser (Michigan State University)+; Sarah Saunders (National Audubon Society); Shannon Reault (National Audubon Society); Brooke Bateman (National Audubon Society); Joanna Grand (National Audubon Society); Elise Zipkin (Michigan State University).
Session: S-5-3.
Where: G043.
When: 10:30-12:30.
Abstract: The 21st century has seen a proliferation of data from large-scale monitoring programs, citizen science programs, and autonomous monitoring approaches to understand occurrence trends, distribution shifts, biodiversity patterns, and their associated drivers. As such data sources are often piecemeal and have different limitations (e.g., preferential or biased sampling), model-based data integration is an attractive solution to make the most of all available data. However, integrating disparate data sets is complicated by differences in spatial resolution, spatio-temporal extent, and observational biases under different sampling protocols. Further, existing data integration approaches are highly computationally intensive when integrating multiple big data sources across large spatio-temporal regions. Here we present a multi-stage hierarchical Bayesian modelling approach for integrating multiple data sets with a focus on estimating spatial variation in occurrence trends across a species’ range. Our multi-stage approach fits a distinct model to each data source in individual stages and uses model estimates from previous stages as predictor variables in subsequent stages to improve estimates of the ecological process of interest. We showcase our modelling framework to quantify spatial variation in the relative contributions of climate and land use change in driving spatially varying occurrence trends in bird communities across the continental USA from 2000-2019. By fitting models in stages, our proposed framework can integrate a variety of data types (e.g., count, biomass, presence-absence), integrate single species and/or multi-species data sources, and can be implemented using user-friendly, formula-based R packages (e.g., spOccupancy, spAbundance)..
Authors: Clara Panchaud (University of Edinburgh)+; Ruth King (University of Edinburgh); David Borchers (University of St Andrews); Hannah Worthington (University of St Andrews); Ian Durbach (University of St Andrews); Paul Van Dam-Bates (Fisheries and Oceans Canada, Pacific Biological Station).
Session: S-6-1.
Where: G043.
When: 10:30-12:30.
Abstract: Estimating wildlife species abundance and distribution underpins the conservation and management of animal populations and natural reserves. Reliable and precise estimates are important to assess the conservation status and the impact of conservation actions, as well as to detect population trends. Capture-recapture studies involving repeated surveys are used to collect data on uniquely identifiable individuals, and increasingly rely on remote sensors, for example, through a spatial array of camera traps. Population and spatial density estimates can be obtained from the data through application of spatially explicit capture-recapture (SCR) models. SCR models consider spatial correlation by assuming that animals are more likely to be observed by sensors close to their activity centre. However, these traditional SCR models rely on the assumption that the probability an individual is observed at a given trap depends solely on the (unobserved) spatial location of their activity centre. This assumption implies that an individual’s known location from its previous trap sighting does not influence the probability of being seen at future times at the given trap locations. This implication is ecologically unrealistic given that animals move through space and time smoothly. We present a new continuous-time modeling framework to account for spatial correlation of observations due to both an individual’s (latent) activity centre and (known) observed locations from previous captures. We show that allowing for the smooth movement of animals over space and time can lead to biased estimates in standard SCR models that only consider the activity centre to explain the spatial correlation of observations; while the new proposed model performs substantially better. We also apply the different models to a real study of pine martens, and demonstrate a significant improvement in model fit to the data when accounting for the known observed locations of previous captures..
Authors: Kylee D Dunham (Cornell University Lab of Ornithology)+; Stephanie Harris (Bangor University); Orin J Robinson (Cornell Lab of Ornithology); Yves Hingrat (Reneco International Wildlife Consultants); Viviana Ruiz-Gutierrez (Cornell Lab of Ornithology).
Session: S-4-2.
Where: G049.
When: 14:30-16:30.
Abstract: Population reinforcement can be an effective strategy for stabilizing small and declining populations. However, evaluating the success of these efforts requires comprehensive information on the demography of wild-born and captive-bred individuals. Here, we evaluate the large-scale reinforcement of houbara bustard (Chlamydotis undulata undulata) in eastern Morocco. Using data from a long-term intensive monitoring effort between 2011 and 2022, we developed a coupled integrated population model (IPM) and population viability analysis to evaluate the efficacy of houbara bustard reinforcement and predict population responses to conservation strategies. First, we modeled point count data using a hierarchical distance sampling model to generate annual estimates of total population size. Within the IPM, we modeled abundance using a state-space model to account for process and observation errors. We analyzed individual tracking data from over 3000 birds using a multi-state model to estimate age, sex, and origin (captive-bred, wild-born) specific annual survival rates. Productivity was modeled as the product of breeding probability, number of breeding attempts, clutch size, hatch rate, nest survival, and brood survival using individual tracking and nest monitoring data. Using the IPM, we estimated origin, age, and sex-specific demographic rates and abundance, providing detailed information on the contribution of captive-bred individuals to wild population dynamics for the first time. We then predicted wild population responses to conservation scenarios including changes to the number of releases and demographic rates. Our results suggest that the wild bustard population is declining, and the total population consists largely of captive-bred individuals. While reinforcement has supported hunting opportunities, these efforts have failed to establish a viable population of houbara bustards due to variable survival and low productivity under current management and environmental conditions. Conservation projections indicate substantial increases in demographic rates and continued releases are required to maintain a viable wild population of houbara bustards..
Authors: Rochelle Kennedy (SRUC)+; Nick Littlewood (SRUC); Elisa Fuentes-Montemayor (University of Stirling); Kirsty Park (University of Stirling); Sarah Marley (SRUC).
Session: S-7-4.
Where: G029.
When: 14:30-16:30.
Abstract: Biodiversity in Europe is in rapid decline, largely due to intensive farming. Population declines of moths are particularly alarming given their important pollination role. Mixed farming systems, where livestock are integrated into the crop rotation, have the potential to increase biodiversity through various pathways. The presence of pasture on the farm and woodland shelter can increase the diversity of habitats available for moths. In addition, grazing livestock provide natural fertiliser and weed control, potentially reducing the need for synthetic fertilisers and herbicides that are known to be harmful to insects. However, grazing livestock can also have detrimental impacts on plant-feeding moths through direct competition for food. Hence the disadvantages of mixed farming may detract from the benefits. This study takes a whole-system approach using causal pathway methods to understand the direct and indirect effects of mixed farming on moths. Thirteen mixed farms and thirteen ‘arable’ farms (where livestock were absent) were selected across North-East Scotland. Each farm was surveyed for moths using light traps between June and August 2022. Moths were identified to species level and were categorised into functional groups depending on their larval food habits. Piecewise structural equation models showed that mixed farming marginally positively affected the agricultural landscape by increasing woodland edge density, this in turn had significant positive effects for ‘micro’ moth (smaller moths) abundance and richness, which was not driven by wood-feeding moth species. However, there was also a significant negative direct effect of mixed farming on micro moths, though this was weaker than indirect effects via increased woodland. This pathway analysis allowed the disentanglement of positive from negative effects in a complex ecological system and highlights areas where more research is needed, for example to investigate whether the direct negative effects of mixed farming are due to pasture management..
Authors: Mario Figueira (University of Valencia)+; David Conesa (University of Valencia); Antonio López-Quílez (Universitat de Valencia); Iosu Paradinas (AZTI).
Session: S-4-2.
Where: G049.
When: 14:30-16:30.
Abstract: In diverse scientific fields like ecology, econometrics, and epidemiology, abundant information sources and extensive databases are available, notably in fisheries science, where data may stem from varying sampling processes like scientific surveys and commercial vessel data. Analysing such data requires diverse strategies, with integrated models gaining traction, though large and complex datasets pose computational challenges.
Our study addresses big data issues while maintaining model complexity. We compare the integrated model strategy with two approaches: a sequential consensus method and a parallel consensus approach. We also investigate scenarios where data differences arise from different sampling designs, contrasting these with simplified sampling assumptions to simplify analysis complexity.
The sequential strategy entails defining prior distributions based on posterior distributions derived from modelling a subset of the initial dataset. This approach sequentially combines information, incorporating feedback of non-random effects and a consensus of random latent field components. The parallel strategy, in turn, uses consensus techniques to process the entire dataset in parallel.
We conduct simulation studies using different data types and sampling designs to compare these approaches. In addition, real data analyses further validate their effectiveness. Using the Integrated Nested Laplace Approximation (INLA) for inference in both simulation and real data analyses, we find highly comparable or nearly identical results between the integrated model and consensus approaches. Furthermore, these methods facilitate the segmentation of the dataset, allowing the reconstruction of the posterior distributions.
Finally, we show the superiority of integrated and consensus approaches over simplified sampling schemes in handling differences arising from different sampling structures. Both simulated and real data examples underline the effectiveness of the consensus methods, highlighting their usefulness in managing large and heterogeneous datasets..
Authors: Seth Harju (Heron Ecological)+; Scott Cambrin (Clark County Desert Conservation Program); Jodi Berg (Alta Science and Engineering).
Session: S-6-2.
Where: G049.
When: 10:30-12:30.
Abstract: Roads have often been identified as barriers to the movement of free-ranging animals. However, whether restoration of landscape connectivity across roadways can mitigate barriers to movement is insufficiently understood in light of indirect effects of roads on wildlife movement. We GPS-tagged free-ranging Mojave desert tortoises (Gopherus agassizii) to quantify movement behavioral states using hidden Markov models in relation to a major highway and to document use of existing, permeable culverts. We then used the observed movement behaviors to parameterize simulations of tortoise movement to evaluate alternative culvert designs and placements for enhancing connectivity across the roadway. Tortoises were most active during mid-day, in warm temperatures, and when close to the highway. The highway affected transition probabilities between movement states, as females were more likely than males to switch to an energy demanding traveling movement state, remain in that state, and move farther than usual within that state. In contrast, males were more likely than females to continue in the low-energy resting state when close to the highway, but if traveling, to travel farther than usual. We observed two highway crossings by a tagged tortoise, which was a higher rate of crossing than in simulated tortoises. Simulated crossing rates increased with culvert size and culvert density, and size and density appeared more important for crossing than if culverts were placed singly or in pairs. Existing culvert densities across the region appeared potentially sufficient for long-term genetic connectivity, but only if retrofitted to allow for tortoise access and passing. We concluded that existing highway traffic may indirectly depress tortoise populations adjacent to the highway, particularly via negative impacts to female movements, and that existing culverts in washes should be retrofitted to allow for periodic tortoise crossings to improve structural connectivity for occasional passage..
Authors: Perry de Valpine (UC Berkeley)+; Daniel Turek (Lafayette Collete); Wei Zhang (University of Glasgow); Paul van Dam-Bates (Fisheries and Oceans Canada); Benjamin R Goldstein (North Carolina State University); Christopher Paciorek (UC Berkeley).
Session: S-3-4.
Where: G037.
When: 10:30-12:30.
Abstract: NIMBLE (R package nimble, r-nimble.org) is a hierarchical statistical modeling and algorithm tool. Within R it provides an extensible BUGS/JAGS-like language for writing models and an algorithm language for using models in a generic way. NIMBLE was designed to allow many algorithms to operate on the same model object, so that one does not need to use multiple packages and rewrite workflows to try alternative methods. Models and algorithms can be automatically compiled via C++. NIMBLE has been used in hundreds of publications, including many in ecology. The customizability of models and MCMC configurations has enabled efficiency gains of 1-2 orders of magnitude in some problems compared to other tools.
This talk will give an overview of recent advances in NIMBLE with ecological examples. NIMBLE now supports automatic differentiation for any subset of model and/or algorithm calculations, built upon the CppAD package. The nimbleHMC package uses this to implement the NUTS HMC sampler, with the distinct feature that it can be used in combination with other samplers. NIMBLE now includes Laplace approximation and adaptive Gauss-Hermite quadrature to marginalize over continuous random effects and maximize the resulting likelihood. Posterior approximation methods are in development, following integrated nested Laplace approximation (INLA), that explore the parameter likelihood surface based on Laplace integration over random effects. These methods can generally work with the marginalized distributions provided in the nimbleEcology package for capture-recapture, occupancy, N-mixture, and hidden Markov models. Together these advances open up an exciting range of methods and applications in NIMBLE. .
Authors: Bokgyeong Kang (Duke University); Erin Schliep (North Carolina State University); Alan Gelfand (Duke University); Tina Yack (Duke University); Christopher Clark (Cornell University); Robert S Schick (Southall Environmental Associates, Inc)+.
Session: S-4-1.
Where: G029.
When: 14:30-16:30.
Abstract: Sound is the primary mode of communication among many marine species. Studying the recordings of these sounds helps to understand the function of the acoustic signals. With increasing anthropogenic noise in the ocean, understanding its impact on the acoustic behavior of marine mammals is needed. One important type of vocalization is an “up-call,” thought to serve as a contact call between individuals that facilitates social cohesion. Motivated by a dataset recorded by a network of hydrophones in Cape Cod Bay, Massachusetts, USA, utilizing automatically detected upcalls in recordings, we aim to study the communication process of the endangered North Atlantic right whale. We propose novel spatiotemporal excitement modeling consisting of background and countercall processes. The background process describes the intensity of contact calls, providing inference for the impact of diurnal patterns and noise on acoustic behavior of the whales. The countercall process accounts for the potential excitement. Call incidence is found to be clustered in space and time; a call seems to excite more calls nearer to it in time and space. We find evidence that whales make more calls during twilight hours, respond to other whales nearby, and are likely to remain quiet in the presence of increased ambient noise..
Authors: Matthew W Rees (CSIRO)+; Jens Froese (CSIRO).
Session: S-7-2.
Where: G043.
When: 14:30-16:30.
Abstract: To promote data-driven and interrogable estimates of species distributions across broad scales, there is need for reproducible workflows which can make use of all available data. Statistical advancements have allowed the integration of different data types to improve our understanding of species distributions, but their application at broad scales has been limited.
In this project we work closely with Government agencies to update essentially hand-drawn distribution maps of priority pest species using integrated species distribution models. I will present a case study on feral pigs as well as reflect on the technical and social challenges in developing standardised, data-driven distribution models for multiple pest species.
Feral pigs are recognised as one of the most damaging invasive species. Despite being widespread across much of Australia, existing national maps of feral pig distribution are based on incomplete and outdated data. This is largely due to disparate data sources being held by many different stakeholders across jurisdictions. We used 168,376 presence-only records, 46,653 presence-absence surveys, and 34,981 abundance-absence surveys to estimate feral pig relative abundance and distribution across Australia. We fitted an integrated species distribution model using six covariates as fixed effects and a spatial random effect, as well as a sampling effort layer to correct for bias in the presence-only data. Importantly, our reproducible workflow allows estimates to be updated as new information becomes available and demonstrates the power of bringing together diverse datasets for continental-scale estimates of species distributions..
Authors: Maeve McGillycuddy (UNSW Sydney)+; Gordana Popovic (UNSW Sydney); David Warton (UNSW Sydney).
Session: S-3-4.
Where: G037.
When: 10:30-12:30.
Abstract: A random effect with a variance covariance matrix proportional to a known covariance matrix offers a versatile approach to fit a range of models. Two common examples in ecology are models with smoothers defined by basis functions, and models with phylogenetic terms. While software is currently available to fit such models, they are not very flexible when it comes to adding additional random effects. We propose adding a new random effect covariance structure to glmmTMB called propto, which adds a multivariate random effect to the model whose variance-covariance matrix is proportional to a user-specified matrix.
We illustrate that we can combine propto with existing mixed model terms to reflect a broad range of potential study designs. In particular multivariate data with nonlinear relationships. We reanalyse a classic phylogenetic data set of maximal running speed in cursorial mammals from Garland and Janis (1993). Maximal running speed shows a non-linear relationship with body mass. This non-linear trend cannot be easily accounted for using available phylogenetic analysis tools. However, we can fit a smoother to phylogenetically structured data using propto random effects in glmmTMB..
Authors: Fanny R Dupont (UBC)+; Marie Auger Methe (UBC); Marianne Marcoux (DFO); Nigel Hussey (Windsor University).
Session: S-6-2.
Where: G049.
When: 10:30-12:30.
Abstract: Hidden Markov Models (HMMs) are a versatile statistical framework used to analyze time-series. In ecology, they are commonly used to characterize behavioural patterns from animal movement data (e.g., time-series of locations). In HMMs, the observed data is dependent on a finite number of underlying (hidden) true states, generally interpreted as the animal’s unobserved behaviour. The number of states is a crucial parameter in HMMs controlling the trade-off between ecological interpretability of animal behaviours (fewer states) and the goodness of model fit (more states). Common model selection metrics (e.g., AIC and BIC) perform poorly in selecting an appropriate number of states, especially when the models do not completely account for complex processes affecting the data. A stationary maximum penalized likelihood estimation approach (MPLE) has been proposed in HMMs by Hung et al. (2013) in cell adhesion analysis and has shown promising results when models are well-specified. Its application remains unexplored in ecology and is challenging given the potential interest in understanding how time-varying covariates affect transition probabilities. We propose a non-stationary MPLE approach for simultaneous estimation of the order (i.e., number of states) and parameters of hidden Markov models. Using a simulation study, we show that our method more frequently selects the correct number of states under model misspecification than AIC and BIC. Using a narwhal (Monodon monoceros) case study, we demonstrate how it helps our ecological inference: our method selects less states than both information criteria and is thus more relevant for ecological inference..
Authors: Charlotte R Patterson (Queensland University of Technology)+; Xiaotian Zheng (University of Woollongong); Kate Helmstedt (Queensland University of Technology ); Justine Shaw (Queensland University of Technology).
Session: S-1-1.
Where: G043.
When: 14:00-15:00.
Abstract: The rapid expansion of biological data collection and immediate need for biodiversity assessments motivate the development of methods that can harness benefits from multiple data types. Integrated Species Distribution Models (ISDMs) have been shown, under certain conditions, to outperform SDMs fitted to a single data type in the precision and accuracy of parameter estimates and can improve spatial prediction within the sampling domain. However, the limits to benefits arising from data integration using ISDMs are not well explored, particularly when an ISDM is transferred to a spatially or environmentally distant site. Two components of an ISDM might improve spatial model transfer: shared learning of coefficients from different observation datasets, and the incorporation of a shared spatial random effect that can capture underlying ecological processes. We present a case study across Antarctica’s ice-free ‘islands’, where spatially constrained survey design and an urgent need for improved prediction of biodiversity patterns motivate the use of ISDMs. With this case study we pose two questions: (1) Can ISDMs improve spatial transfer relative to single dataset approaches and under what conditions? And (2) How does adding or removing a spatial random effect change the success of a spatial transfer? Using recent data from East Antarctica, we test these questions with ISDMs for presence-only and presence-absence data. We validate model predictions at a separate ice-free site with an independent presence-absence dataset..
Authors: Leslie Skora (University of Massachusetts )+; Tammy Wilson (U.S Geological Survey).
Session: S-8-4.
Where: G029.
When: 11:00-12:45.
Abstract: Traditional mark recapture techniques for estimating wildlife population size and density can be costly and time consuming making it difficult to replicate and monitor populations over time. This is particularly relevant in remote areas such as Katmai National Park and Preserve (Katmai) in southwest Alaska, home to one of the largest brown bear (Ursus arctos) populations in the world. The remote location, harsh weather conditions, and limits to staff and funding make monitoring brown bear population abundance extremely difficult. Although Katmai staff conduct regular aerial counts of bears feeding at salmon spawning streams, the resulting data have been difficult to analyze leaving a gap in our understanding of how the abundance of Katmai’s famous fat bears (Fat Bear Week) has changed through time. Using conventional distance sampling data from three comprehensive aerial surveys conducted in 2004/2005, 2009, and 2022, we estimated brown bear density and population size through time in Katmai National Preserve. In future work we will combine this distance sampling estimator with our annual stream surveys to gain a more complete understanding of how bear populations change through time..
Authors: Julie Vercelloni (Australian Institute of Marine Science)+.
Session: S-7-1.
Where: G049.
When: 14:30-16:30.
Abstract: Long-term monitoring surveys are essential to detect ecological changes across space and time. Spatio-temporal modelling uses monitoring data to estimate spatial patterns while considering the effects of drivers in species responses, and interpolate over a continuous spatial field to predict responses at unobserved locations. This approach allows to increase the volume of information to interpret and detects more robust trends across spatial scales. However, it is complicated to validate model predictions because model outputs are typically specified at different spatial scales than the observational data.
In this study, we develop a quantitative framework to investigate what influence the performances of predictive Bayesian generalized linear mixed models including the sampling design and types of spatio-temporal random effects. To do this, the framework creates synthetic data based on information from sampling design represented by the size of spatial domain, fixed or random sampling, number of surveyed locations and replicated years, presence of hierarchical spatial scales and effects of disturbances. Then, several spatio-temporal models are fit using popular R packages (INLA, sdmTMB, FRK, mgcv and brms) and model performances investigated based on various diagnostics.
We showcase the framework using a case study from the Australian Great Barrier Reef. Synthetic data are created using the sampling design from Australian Institute of Marine Science’s Long-Term Monitoring Program. Spatio-temporal models estimate long-term trajectories of coral cover for every coral reef in a region under the effects of bleaching events and cyclones. We show how performances of the models are influenced by modifying the original sampling design and develop a set of recommendations to optimize their use. The proposed framework is flexible enough that it can be easily adapted for different ecological case studies enabling better understanding of predictive models and ultimately more robust assessments of biodiversity changes across space and time..
Authors: Albert Bonet Bigata (University of Aberdeen)+.
Session: S-2-2.
Where: G043.
When: 15:30-16:30.
Abstract: Invasive mammalian predators are a major leading cause of global defaunation, contributing to over half of all recorded vertebrate extinctions. Eradication efforts of long-established invasive species over large areas are usually prohibitively expensive, leading to long-term population control strategies instead. These operate assuming that suppressing invader’s residual populations below desirable thresholds will reduce their impact to negligible levels. However, to test such assumptions and optimise control strategies, reliable methods are needed to estimate how removal efforts change population residual abundances over time. Removal models have been historically used for these strategies, but classical formulations require strict assumptions (e.g. population closure) and can perform poorly under scenarios of high capture effort and spatial heterogeneity. One of these scenarios is the long-term control of invasive American mink Neogale vison in Scotland, encompassing a wide range of habitats and patchy removal efforts.
By integrating recent developments in Bayesian removal models, we provide the first estimation of spatiotemporal residual population dynamics of the invasive American mink. We use data from 20 years of control with varying intensity led by citizen conservation efforts within a 29,000 km2 area in Scotland. We estimated within- and between-year regional dynamics as a function of environmental factors while accounting for heterogeneous capture effort and probabilities, spatial biases, and missing data patterns common in citizen-science programmes. The results show how spatiotemporal patterns of abundance and capture rates vary along environmental gradients and control efforts as well as mink individual features. Additionally, we show how sex and age distributions of captured and residual invader populations change across time and space depending on habitat features and control effort. The modelling framework and results provide a blueprint tool to reliably estimate residual invader abundances under high environmental and effort heterogeneity and could serve to optimise ongoing and future large-scale invasive predator removal efforts..
Authors: Jing Liu (University of Auckland)+; Rachel Fewster (University of Auckland); Ben Stevenson (University of Auckland).
Session: S-3-3.
Where: G049.
When: 10:30-12:30.
Abstract: Spatial capture-recapture (SCR) models are commonly used to estimate animal population density. These models allow the probability that an animal is detected to vary amongst spatially separated detectors based on the location of its activity centre.
However, activity centres are not observed on a SCR survey and are treated as random effects. Existing methods require sampling or numerical integration over activity centres to obtain the marginal likelihood function.
In this talk, we show that the exact likelihood function for a popular SCR model is available in closed form, and so are the derivatives. This result leads to faster model fitting and the possibility of survey-design recommendations. In addition, one can obtain analytical expressions for some other SCR models by combining this result with Taylor-type approximation..
Authors: Murray B Christian (Centre for Statistics in Ecology, Environment and Conservation, University of Cape Town)+.
Session: S-6-1.
Where: G043.
When: 10:30-12:30.
Abstract: Multievent models were introduced by Pradel (2005) to generalise multistate models to the case where an individual’s state cannot be determined with certainty when captured. To achieve this, they model each capture history as a Hidden Markov Model (HMM) in which the observed capture events arise from possibly hidden biological states. Consequently, multievent models can be fit using HMM techniques, namely marginalised likelihoods computed with the forward algorithm. This HMM formulation is very general and can also be used to fit multistate models. However, for multistate models it is often preferable to use alternative product-multinomial likelihoods for a reduced data representation called an ‘m-array’. The m-array summarises the capture histories by the number of individuals released in each state on each occasion that are next recaptured in a given state at a later occasion (or never recaptured). Fitting the product-multinomial likelihoods is often substantially quicker than HMM likelihoods, and the m-arrays form the basis for diagnostic goodness-of-fit tests for multistate models. Despite these advantages, no m-array has been developed for multievent models to date. In this talk I will describe a product-multinomial likelihood for multievent models that directly generalises the multistate case, and present some results on efficiency gains in model fitting..
Authors: Javier Fernández-López (Institute for Game and Wildlife Research IREC (CSIC-UCLM))+; Olivier Gimenez (CNRS); Pelayo Acevedo (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); José Antonio Blanco-Aguiar (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); Joaquín Vicente (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); Ana Santamaría (WWF-Spain); Tamara Burgos (WWF-Spain); Sonia Illanas (Institute for Game and Wildlife Research (IREC-CSIC-UCLM)); Davide Carniato (Institute for Game and Wildlife Research IREC (CSIC-UCLM)); Fernando Silvestre (Fundación CBD-Hábitat); Sergio Ovidio (Junta de Comunidades de Castilla-La Mancha); Ángeles Sánchez (Junta de Comunidades de Castilla-La Mancha); Llanos Gabaldón (Junta de Comunidades de Castilla-La Mancha); Ramón Pérez de Ayala (WWF-Spain).
Session: S-5-3.
Where: G043.
When: 10:30-12:30.
Abstract: The European rabbit (Oryctolagus cuniculus) is a key species in the Iberian Peninsula which participates in many ecological processes and is involved in numerous human activities (hunting, agriculture, etc.). The Iberconejo project (funded by LIFE program from the European Union) aims to develop and implement a system for monitoring rabbit populations in the Iberian Peninsula, which is essential for a correct species managing and conservation. In a first stage, to analyze the state of rabbit populations in Castilla-La Mancha region (central Spain) we combined distance sampling data (n= 144) with hunting statistics (n= 2720) and latrine counts (n= 58) by using joint likelihood methods in a hierarchical Bayesian framework to develop and mapping spatially explicit models of rabbit abundance. Change-of-support methods were used to reconcile misaligned spatial resolution of data sources and restricted spatial regression approach was implemented in NIMBLE to account for spatial autocorrelation in rabbit distribution and abundance. Maximum rabbit densities were related with crops and agricultural areas in Castilla-La Mancha region, while lower densities were founded in forest areas. Cross validation tests indicated a good performance of distance sampling and harvest data, which agreed with predicted abundance. However, observed and predicted latrine counts did not match, and hence a deeper understanding of observation process for latrine counts is needed. Data integration is a powerful tool in wildlife monitoring programs, specially for widely distributed species in which their management depends on several Administrations (different autonomous regions or countries). However, a strong knowledge about the processes originating each input data set is necessary in order to make the most of each information source..
Authors: Ricardo Carrizo Vergara (Schweizerische Vogelwarte)+; Marc Kéry (Schweizerische Vogelwarte); Trevor Hefley (Kansas State University).
Session: S-8-1.
Where: G043.
When: 11:00-12:45.
Abstract: We present two statistical models for count data for unmarked individuals, constructed through the description of the aggregate of trajectories of moving individuals. While most current models for dynamic abundance, such as ecological diffusion models, invoke independence assumptions over space-time grids, our models take into account a random dynamics of each individual that induces a particular space-time auto-correlation structure on the counts. In both models, the trajectory of individuals are currently supposed to be independent and identically distributed and following an arbitrary continuous-time stochastic process. Our first model, the Snapshot model, assumes that the counting does not affect the movement of the individuals, while in the second , the Capture model, we consider the case where individuals can be captured, retained and liberated, and that the abundance data corresponds to the total number of retained individuals at a given time/space unit. This model follows an axiomatic approach for the behavior of the random variables involved, implying in particular that the distribution of the capture time must have a density which is the solution to a Volterra integral equation of the second kind. For both models we describe some key statistical properties, particularly their mean and covariance structures, and develop simulation methods. Since the explicit likelihood of these models is costly to evaluate,we propose two likelihood-free methods for fitting them to data: a maximum Gaussian likelihood estimator, justified by the central limit theorem, and an adaptive Approximate Bayesian Computation method enjoying our simulation methods. We illustrate our models with an experimental dataset of flies released in a landscape and then repeatedly captured and counted..
Authors: Niklas Moser (University of Jyväskylä)+; Dmitri Finkelshtein (Swansea University); Georgy Chargaziya (Swansea University); Sara Hamis (University of Uppsala); Dagim Tadele (University of Oslo); Otso Ovaskainen (University of Jyväskylä).
Session: S-5-4.
Where: G029.
When: 10:30-12:30.
Abstract: Statistical ecology is facing a mismatch between models and data, resulting in two key modelling challenges: (1) reliable predictions under global environmental change and (2) the inference of mechanisms that drive the observed ecological patterns to broaden system understanding. As most ecological processes are based on discrete interacting agents, agent-based models (ABMs) seem especially well poised to appropriately represent ecological phenomena and thus to overcome these challenges of prediction and inference. However, while ABMs have been shown to be suitable for simulating complex system dynamics, the model behaviour is too complicated in most systems to be analysed mathematically. Thus, due to intractable likelihood equations, the model calibration of most ABMs is bound to heuristic workarounds like pattern-oriented modelling where extensive simulations are compared to predefined criteria or the use of pseudolikelihoods which lack mathematical rigor. In this study we consider a broad range of ABMs that can be defined as spatio-temporal point processes, specifically the reactant-catalyst-product models (RCP-models) that operate in continuous space and time. We mathematically describe the system of interest by spatial and spatio-temporal moments and cumulants of any order without being constrained by heuristic moment closure methods. We apply a perturbation expansion that enables us to include information beyond the mean field approximation. This results in a general, rigorously derived and asymptotically exact expression for the conditional density of any agent type given information about the distribution of other agents. We utilize the conditional density predictions to construct a composite likelihood and a Bayesian parameter estimation framework. We use simulated case studies of both single- and multi-species communities as well as empirical data on the distribution and time evolution of cancer cells to illustrate the general suitability of this approach to parameterize agent-based models based on snapshot or time-series data in ecological research and beyond..
Authors: Bob Douma (Wageningen University)+; Eveline van Woensel (Wageningen University); Stephen Parnell (University of Warwick); Arnold van Vliet (Wageningen University); Wopke van der werf (Wageningen University).
Session: S-6-3.
Where: G029.
When: 10:30-12:30.
Abstract: Countries across the globe get more and more connected through trade and movement of people. This also leads to adverse effects: pests and pathogens hitchhike to new countries and may establish there. Some of these cause great damage to plants and trees in these newly colonized areas. International agreements require countries to monitor and substantiate their freedom of plant pests. It poses a challenge to plant protection organization to monitor the presence or absence of non-native pests, and it requires substantial resources to declare an area free of pests. Citizen science data is collected widely and could be used to help substantiate pest freedom. However, currently it is only used on an informal basis. Inherent spatial and temporal biases in citizen science data and the lack of absence reports make it, however, difficult to extract meaningful information. Here, we present a methodology to calculate the detection probability of insect pests from citizen science data and use citizen science data to complement official surveys. We did this by constructing proxies for sampling effort for citizen science data and subsequently estimating the probability of detection for locations at which a pest was known to occur. The methodology is illustrated with an invasive plant pest in Europe, Popillia japonica, that was first observed in the Italian region of Lombardia in 2014. We conclude that citizen science can help to make usable statements on pest freedom and as such can contribute to the substantiation of pest freedom..
Authors: Anastasia Frantsuzova (BioSS)+.
Session: S-5-1.
Where: G037.
When: 10:30-12:30.
Abstract: The impacts of offshore wind farms on seabird populations are an important issue for offshore renewable development in the UK. Seabirds can be affected through displacement from foraging areas, collisions and barrier effects on migratory routes or regular flight paths.
The ECOWINGS project (Ecosystem Change, Offshore Wind, Net Gain and Seabirds), funded by the [UK] Natural Environment Research Council, The Crown Estate and DEFRA through the EcoWind programme, aims to transform the existing evidence base on the cumulative effects of offshore wind on key seabird species. ECOWINGS aims to gain insight into the potential to compensate for the cumulative effects of offshore wind farms on seabirds and to achieve biodiversity net gain.
One approach to estimate the impact of marine developments on breeding seabirds is through an individual-based model (IBM) called SeabORD (Searle et al. 2014, 2018). SeabORD is an individual-based stochastic simulation model to predict the survival and reproductive outcomes of seabirds from breeding colonies impacted by offshore renewable energy developments. This sophisticated model encompasses parameters such as foraging behaviour, energetics, and reproductive success during the breeding season. However, as is the case with many IBMs, SeabORD is computationally intensive. To improve computational efficiency and facilitate sensitivity analysis, we explore the use of Gaussian stochastic process emulators as a model surrogate for SeabORD. In this talk, we will outline the structure of SeabORD and its complexity and describe Gaussian process emulation and the potential for this approach to predict the impacts on seabird populations that SeabORD would simulate under wind farm scenarios for which it has not been run. Authors: Anastasia Frantsuzova (BioSS) [presenting], Adam Butler (BioSS), Ken Newman (BioSS), Kate Searle (UKCEH), Francis Daunt (UKCEH), Esther Jones (BioSS) .
Authors: Katherine Whyte (Biomathematics and Statistics Scotland (BioSS))+; Ana Couto (BioSS); Charlie Cooper (Scottish Government Marine Directorate); James Dunning (Scottish Government Marine Directorate); Christopher Pollock (UK Centre for Ecology and Hydrology (UKCEH)); Thomas Cornulier (Biomathematics and Statistics Scotland); Adam Butler (Biomathematics and Statistics Scotland); Thomas Regnier (Scottish Government Marine Directorate); Kate Searle (UK Centre for Ecology and Hydrology (UKCEH)); Francis Daunt (UK Centre for Ecology and Hydrology (UKCEH)); Esther L Jones (Biomathematics & Statistics Scotland).
Session: S-6-2.
Where: G049.
When: 10:30-12:30.
Abstract: Interactions between predators and their prey occur at multiple spatial, temporal, and ecological scales. Understanding and quantifying these interactions is important for (1) improving our ecological knowledge of the drivers of animal movement, and (2) increasing our ability to robustly predict how animals will respond to human alterations to the environment.
As part of the PrePARED (Predators and Prey Around Renewable Energy Developments) project concurrent data are being collected on marine top predators and their prey, but there are statistical challenges in using these data to understand predator-prey relationships. First, determining an appropriate spatiotemporal scale at which to analyse these interactions, accounting for data resolution and the scales at which these interactions can meaningfully be examined ecologically. Second, accounting for mismatches in space and/or time since, in practice, exact overlap between predator movements and prey data collection occurs only occasionally. Third, considering how to maximise the benefits of the different spatiotemporal scales of prey data that are available: regional prey surfaces, area-specific prey surfaces, and individual prey school detections.
Here, we discuss the challenges and opportunities of using movement modelling to examine predator-prey interactions, using GPS tracks of individual seabirds (e.g. kittiwakes, auks) and contemporaneous data on fish distributions (e.g. sandeel, clupeids) from dedicated acoustic and trawl surveys. In particular, we consider how we can incorporate prey data directly into Hidden Markov Models describing the movements and at-sea behaviours of different seabird species, allowing us to improve our understanding of the mechanisms driving predator-prey interactions. We show how we are using this work to (1) further understand the drivers of seabird movements and foraging behaviour, (2) develop approaches that account for the variable spatial and temporal scales of ecological processes and observations, and (3) ultimately improve our ability to predict seabird movements in an environment altered by anthropogenic activities..
Authors: Gavin L Simpson (Aarhus University)+.
Session: S-7-1.
Where: G049.
When: 14:30-16:30.
Abstract: The species abundance curve (SAC) and its ranked variant (RAC) provide one of the most fundamental yet basic descriptions of an ecological community. A universal feature of such curves is the paucity of highly abundant taxa and the commonality of rare ones. SACs and RACs have been much studied over the century or so of their use, and the literature is filled with discussion of the mechanisms leading to the variety of shapes taken by these curves. A distinct advantage of using RACs to compare communities is that they allow comparison of communities with few or no species in common, because species identity is ignored.
Recently, Avolio et al (2019, Ecosphere 10(10), e02881) proposed a framework for comparison of RACs for the analysis of ecological community dynamics across space and time. This framework is largely metric-based, with statistical analysis being performed on the values of the derived metrics. In this talk, I propose an alternative approach for fitting models to RACs that is purely phenomenological, providing a direct statistical estimation of variation in RACs across space and time. The approach is similar in spirit to Avolio et al’s framework, but proceeds by direct fitting of a particular statistical model, a hierarchical generalized additive model (HGAM), to the collection of RACs. Taking inspiration from smoothing histograms, I describe how HGAMs with penalized tensor product splines can model the changing shape of RACs over space and time. The motivation behind the proposed approach is to statistically compare ecosystem dynamics in communities with very different species composition, or even those with no species in common.
I will briefly describe the proposed approach and demonstrate its use with examples from a long-term nutrient enrichment experiment, a spatial survey from a serpentine grassland, and a time series of a desert rodent community..
Authors: Florencia Grattarola (Czech University of Life Sciences Prague)+; Gurutzeta Guillera-Arroita (Pyrenean Institute of Ecology, Spanish National Research Council); José J. Lahoz Monfort (Pyrenean Institute of Ecology, Spanish National Research Council); Petr Keil (Czech University of Life Sciences Prague).
Session: S-5-3.
Where: G043.
When: 10:30-12:30.
Abstract: Understanding biotic interactions has become a key topic in the study of species distributions. Unfortunately, species occurrence data are typically available at a coarser grain than the one of biotic interactions. We ask: can fine-scale associations be detected using coarse-grain data or a combination of different coarse-grain data types? We performed a simulation study to assess if we can recover species interactions at fine grain from data of varying spatial grains, combining ideas from two rising popular methods: integrated species distribution models (iSDMs) and joint species distribution models (jSDMs). We simulated two species that are jointly influenced by an environmental covariate and have varying levels of species-species associations (e.g., negative, zero and positive). We assumed an underlying Poisson point pattern and described the species associations with a covariance matrix (residual species–species correlations). We then drew presence-absence and abundance data at three different increasing grain sizes and modelled the species distribution aiming to recover the true parameters influencing the distributions (environment and species associations). We generated 10 replicates for each species-environment pair, grain size and species covariance and ran them in R with JAGS. Our preliminary findings show that as we increase the grain of the species data, we fail to recover the species association. However, our results indicate that we have not yet successfully captured the complexity of these interactions when using point patterns. This study serves as a foundation for future exploration as we plan to assess several modelling variants for data integration. Specifically, we will explore alternative models that can enable us to integrate data, such as downscaling species distribution models..
Authors: Carlina C Feldmann (Bielefeld University)+.
Session: S-6-2.
Where: G049.
When: 10:30-12:30.
Abstract: In movement ecology, measuring individual variation poses a challenge but also opens up unique opportunities for large-scale assessment of behavioral diversity. To gain insight into animals’ behavior from time series data, hidden Markov models (HMMs) provide a flexible and popular modeling framework, allowing inference on behavioral patterns underlying the observed data. While random effects can be included in HMMs to model individual variation and heterogeneity of the population, the resulting models are limited in their flexibility, as only some parameters can be assumed to be individual-specific. Therefore, we propose mixtures of HMMs, which allow all model parameters and even the model structure to vary across subpopulations. The proposed model is used to investigate behavioral diversity within species, aiming to uncover distinct behavioral patterns, such as foraging strategies in Galápagos sea lions. We thereby avoid multi-stage approaches of first clustering the animals into subgroups and subsequently learning about behavioral phenotypes. Instead, the proposed framework integrates these steps and jointly infers group membership as well as behavioral characteristics of each group. Furthermore, it opens avenues for comparative studies across different species, potentially providing valuable insights into species-specific behaviors and their ecological significance..
Authors: Killian GREGORY (CEFE, Univ Montpellier, CNRS, EPHE, IRD, Montpellier)+; Charlotte FRANCESIAZ (OFB, DRAS, Juvignac); Jean-Yves BARNAGAUD (CEFE, Univ Montpellier, CNRS, EPHE-PSL University, IRD, Montpellier); Jean-Dominique LEBRETON (CEFE, Univ Montpellier, CNRS, EPHE, IRD, Montpellier); Pierre-André CROCHET (CEFE, Univ Montpellier, CNRS, EPHE, IRD, Montpellier); Aurélien BESNARD (CEFE, Univ Montpellier, CNRS, EPHE-PSL University, IRD, Montpellier); Julien PAPAÏX (INRAE, Avignon).
Session: S-8-3.
Where: G037.
When: 11:00-12:45.
Abstract: Dispersal is central to metapopulation dynamics as it contributes to the extinction, colonisation and synchronisation of populations. Given its impact on fitness, dispersal between habitat patches relies on information use by individuals. Both personal and social information are known to direct emigration and settlement decisions in a number of species. Yet information use as a driver of dispersal has rarely been included in studies to understand metapopulation dynamics. Here, we incorporate public information likely to drive dispersal decisions into a dynamic occupancy and population dynamics framework to shed light on how and at what spatial scale informed dispersal shapes the dynamics of a metapopulation of a colonial bird species, the Black-Headed Gull (Chroicocephalus ridibundus), occupying a network of continental ponds in central France. We developed a model relating colony persistence probabilities, pond colonisation probabilities, and local growth rate of the colonies to 1) colony size or breeding success of the colonies in the previous year, and 2) colony size or breeding success of neighbouring colonies with a weight depending on their distance from the focal colony. Preliminary results suggest that while the positive effects of colony size and breeding success on colony persistence cannot be dissociated, colonisation depends only on the breeding success of the nearest neighbouring colonies. New colonies are more likely to appear in areas that hosted successful colonies in the previous year, while areas where colonies failed their reproduction are less likely to be colonised. Persistence of colonies could thus result from any combination of local demographics, conspecific attraction and use of personal or public information. However, the redistribution of individuals near successful colonies and the avoidance of unsuccessful colonies is more likely to result from the use of conspecific success as public information at a spatial scale of a few kilometres..
Authors: Jessica A Leivesley (University of Toronto)+; Hunter Chen (University of Toronto); Simone Collier (Ontario Ministry of Natural Resources and Forestry); Lewei Er (University of Toronto); Henrique Giacomini (Ontario Ministry of Natural Resources and Forestry); Ryan Grow (Lakehead University); Jeremy Holden (Ontario Ministry of Natural Resources and Forestry); Victoria Kopf (Ontario Ministry of Natural Resources and Forestry); Yihang Luo (University of Toronto); Scott Milne (Milne Technologies); Michael Rennie (Lakehead University); Alex Ross (Lakehead University); Christine Wang (University of Toronto); Alice Zhang (University of Toronto); Miley Zhang (University of Toronto); Dak de Kerckhove (Ontario Ministry of Natural Resources and Forestry); Vianey Leos Barajas (University of Toronto).
Session: S-3-1.
Where: G029.
When: 10:30-12:30.
Abstract: Canada’s recreational fishery contributed $7.9 billion to the national economy in 2015, and in Ontario alone freshwater recreational and commercial fisheries represent a $2.2 billion industry. To maintain sustainable and resilient fisheries, managers must have accurate information on the current status of stock health, population-size, and fish communities for many water bodies at a given time. Generally, this information is gathered through resource-intensive and lethal sampling methods. Current hydroacoustic methods can assess individual fish sizes but species identities cannot be discerned. The recent development of wideband acoustic transducers which emit a wide range of frequencies in a single ping may allow more information on body form to be extracted and thus may aid in species identification. In this study, we created a labelled dataset of acoustic responses of two fish species by tethering individual fish under a transducer emitting 249 frequencies between 45kHz and 170kHz. We then applied three different bespoke machine learning algorithms (deep, recurrent, and residual neural networks) to acoustic backscatter measures at each frequency and tested their ability to correctly classify the two fish species. We found that on unseen data all three methods had over 85% balanced classification accuracy. Further, extracting SHAP values for the deep neural network showed that there is not a single range of frequencies that are important for distinguishing the species, but rather the most important frequencies are distributed across the range of frequencies used. Eventually, these algorithms can be integrated into current abundance or biomass models and allow users to propagate classification uncertainty into these models. Overall, the use of wideband acoustics in conjunction with machine learning techniques offers the potential to drastically reduce the resources needed and costs associated with monitoring fish stocks..
Authors: Aline M Lee (Norwegian University of Science and Technology)+; Ellen Martin (Vogelwarte); Ragnhild Bjørkås (Norwegian University of Science and Technology); Jonatan Marquez (Institute of Marine Research); Ivar Herfindal (Norwegian University of Science and Technology); Brage Hansen (Norwegian Institute of Nature Research); Marlene Gamelon (Laboratoire de Biométrie et Biologie Évolutive, CNRS); Sondre Aanes (Norwegian Computing Center); Steinar Engen (Norwegian University of Science and Technology); Are Salthaug (Institute of Marine Research); Bernt-Erik Sæther (Norwegian University of Science and Technology).
Session: S-5-1.
Where: G037.
When: 10:30-12:30.
Abstract: Population dynamics of single species are often synchronized over large areas due to spatial autocorrelation in environmental fluctuations – a phenomenon known as intraspecific population synchrony. Similarly, coexisting species that experience the same environmental fluctuations can show synchronized dynamics, known as interspecific population synchrony. However, different species show different levels of synchrony even when they experience the exact same environmental fluctuations, indicating that species-specific responses to the environment are important in determining patterns of population synchrony within and among species.
We hypothesized that this variation among species could be linked to their position along the slow-fast continuum of life histories. Fast-lived species have short generation times and can produce large numbers of offspring at a time, enabling rapid changes in population size as the environment changes. This could be expected to promote more synchronized dynamics. We investigated this in three large data sets (European birds, fish in the Barents Sea and birds in the Serengeti), testing whether levels of population synchrony within species, the distances over which they are synchronized, and levels of population synchrony between species could be linked to generation time.
Each of these data sets and analyses had their own challenges requiring the use of a variety of statistical methods, particularly for dealing with different sources of variability in the large spatio-temporal datasets and dealing with the inherent dependency structure in pairwise correlations within communities. In addition, we built a stochastic simulation model to place these individual studies within a greater theoretical framework and identify other factors, such as the relative sensitivity of individual vital rates to changes in the environment, that might alter our expectations for patterns of population synchrony within and among species. .
Authors: Frédéric Gosselin (INRAE)+; Ghislain Vieilledent (CIRAD); Clément Vallé (MNHN).
Session: S-4-4.
Where: G037.
When: 14:30-16:30.
Abstract: Many joint species distribution models (JSDMs) used in ecology rely on latent factors to simplify the structure of residual correlation between species and render numerical estimation of the model more feasible. Based on simulated data and reference datasets in the JSDM literature, we show that Bayesian JSDMs (fitted with packages boral, hmsc, jsdm or written directly with Nimble) suffer from convergence issues, especially for the species loadings and random factors. We analyze the issue, how frequent it is reported in the literature and the impact on inference. We finally propose potential solutions to solve the problem. These problems and solutions might also be useful for (latent) factor analysis statistical models..
Authors: Cas Retel (Statistics Netherlands)+; Kathryn M. Irvine (United States Geological Survey); Arco J. van Strien (Statistics Netherlands).
Session: S-7-3.
Where: G037.
When: 14:30-16:30.
Abstract: To protect biodiversity, it is important to know its state. Preferably, biodiversity knowledge consists of information on presence (and absence) of species in an area, as well as a quantitative component (how common or rare the species is). For vascular plants a typical quantitative measurement of population density is the projected percentage of ground covered. Such cover measurements are bound between 0 and 1 and are often recorded in discrete ordinal classes of variable bandwidths (cover-abundance scales). This makes them challenging to model. Here we adapted recently developed ordinal zero-augmented beta regression techniques to a longitudinal study design. This enabled us to apply the model to a Dutch national vegetation monitoring dataset, consisting of vegetation-plot releves collected at 15000 fixed plots, surveyed every four years during 1999-2022 (including data on over 2000 vascular plants). We found evidence of cover changes for 318 out of 721 species with sufficient data and sometimes detected changes in cover for species where evidence for changes in occurrence was absent (and vice versa). We discuss some commonalities and differences in the effect size distribution of cover versus occurrence changes. The analysis of quantitative information on top of binomial presence-absence measurements improved our understanding of Dutch vascular plant population dynamics over the past decades. Our methodology and code implemented in a Bayesian framework (JAGS) is publicly available and facilitates the analysis of longitudinal plant cover-abundance scale data, which is regularly collected in large-scale vegetation surveys but not always rigorously analysed. .
Authors: Andrew Stillman (Cornell Lab of Ornithology)+; Gavin Jones (USDA Forest Service); Matt Strimas-Mackey (Cornell Lab of Ornithology); Guillermo Duran (Cornell Lab of Ornithology); Caitlin Andrews (USDA Forest Service); Shawn Ligocki (Cornell Lab of Ornithology); Tom Auer (Cornell Lab of Ornithology); Viviana Ruiz-Gutierrez (Cornell Lab of Ornithology); Sarah Sawyer (USDA Forest Service); Daniel Fink (Cornell University).
Session: S-7-2.
Where: G043.
When: 14:30-16:30.
Abstract: Wildlife responses to local habitat conditions and environmental disturbance are known to vary across space, yet data collection constraints can prevent us from generating enough data for local-scale inference across broad extents. To aid conservation decision-making, we need to understand broad-scale habitat associations while also capturing variation at scales which are tractable for management decisions. Using a case study with six bird species of management interest in the United States, we addressed this challenge by applying a spatiotemporal ensemble modeling approach to eBird participatory science data to measure nonstationary environmental associations. We focused our inference on three covariates describing the recent fire history at each location: fire frequency, time since fire, and fire severity. Our approach used an adaptive spatiotemporal exploratory modeling framework with an ensemble of random forest models to study species-fire relationships using partial dependence and predictor importance statistics. We then summarized information across the ensemble to map the direction, magnitude, and importance of fire associations at 27-km resolution across the range of each focal species. Our findings revealed previously undocumented fine-scale variation in fire regime associations. All but one focal species displayed both positive and negative effects for at least one fire covariate, even after removing estimates with low confidence. Critically, fire regime associations varied widely in magnitude even when directions remained constant. Moving forward, this analytical workflow provides a flexible approach for assessing range-wide habitat associations and disturbance impacts while also providing information at a sufficiently fine grain for resource prioritization and decision-making..
Authors: Philipp H Boersch-Supan (British Trust for Ornithology)+.
Session: S-8-1.
Where: G043.
When: 11:00-12:45.
Abstract: Understanding three-dimensional habitat use of birds (and bats) is a crucial prerequisite for understanding collision risks associated with wind farms and other human made infrastructure. However, obtaining accurate and precise estimates of flight height distributions of animals from observational data remains a considerable challenge, given that most if not all observation approaches suffer from non-negligeable measurement errors and/or imperfect detection. Reconstructing flight height distributions therefore requires density or distribution function estimation from data contaminated with often complex errors, not least because raw flight height measurements are increasingly derived from sensor-based technologies.
I present work towards improved inferences about flight height distributions of seabirds in the context of offshore wind developments, where environmental conditions provide additional observational challenges. In particular, I will present state-space models and deconvolution approaches for reconstructing flight heights from tracking data derived from animal-borne GPS sensors and from computer-vision based approaches, and outline research priorities for improving statistical frameworks for imperfect sampling in 3D space..
Authors: Ephraim M Hanks (Penn State University)+; Frances Buderman (Pennsylvania State University); Viviana Ruiz Gutierrez (Cornell Lab of Ornithology); Jim Russell (Muhlenberg College).
Session: S-4-2.
Where: G049.
When: 14:30-16:30.
Abstract: Animal movement behavior is complex, and there is often significant variation in movement behavior across a species. We introduce a nonparametric approach for modeling variation in migratory behavior among individuals and spatially across the range of a species. This approach models variation in individual migratory paths (i.e., from GPS data on individual animals) through a Bayesian gradient-based clustering model in which individual behavior is seen as a continuous mixture of multiple estimated extreme behaviors. We show that this approach, which is motivated by provides a more accurate and parsimonious description of variation in the population than competing models, including finite mixture models and other nonparametric clustering methods. We also introduce an approach for scaling this individual-level model to a population-level probabilistic model for the spatio-temporal dynamics of the full population throughout the full annual cycle. This allows for joint inference on individual (i.e. GPS) and population-level (i.e. spatio-temporal abundance data) data. We illustrate this method on continental-scale migratory bird data, and use this method to predict the full-annual-cycle, spatio-temporal distribution of animals that winter within a defined management unit, and also predict spatio-temporal risk maps for animals from that same management unit. .
Authors: Brandon D Merriell (Trent University)+; Micheline Manseau (Environment & Climate Change Canada); Paul Wilson (Trent University).
Session: S-5-2.
Where: G049.
When: 10:30-12:30.
Abstract: Abundance estimation is often an objective of wildlife monitoring programs for threatened or otherwise managed populations. While the use of capture-mark-recapture or spatially explicit capture-recapture methods to estimate abundance is commonplace, such approaches require repeated sampling events which can prove logistically challenging for many systems. The recently developed close-kin mark-recapture (CKMR) framework, which uses the number of kinship pairs obtained within a sample to generate an abundance estimate, does not require multiple sampling events, making it an attractive alternative. However, the existing CKMR framework relies on a pseudolikelihood which assumes that the target population is large and subjected to sparse sampling. As a result, the current CKMR framework is ill-suited for relatively small, intensively monitored populations which may be more typical for terrestrial systems of concern. Fortunately, in cases where there are only a few discrete reproductive classes (such as subadult, prime adult, and old adult), we show that the multivariate hypergeometric distribution (or the hypergeometric distribution in some cases) can be used to calculate the true likelihood, eliminating the need for the pseudolikelihood and its assumption of a large, sparsely sampled population, thereby making CKMR more applicable to terrestrial systems of concern. We demonstrate how the (multivariate) hypergeometric distribution can be used for CKMR to generate an abundance estimate under various reproductive systems and show how this approach can accommodate uncertainty, such as uncertain kinship relations or uncertain life stage assignments..
Authors: Todd Arnold (University of Minnesota)+.
Session: S-2-1.
Where: G037.
When: 15:30-16:30.
Abstract: One commonly stated assumption of mark-recapture-recovery models is that all sampling periods are instantaneous, so that intervals between successive encounter periods are consistent (e.g. 1 year) for all sampled individuals. For populations that are marked opportunistically throughout the year, as occurs for many dispersed bird ringing programs, this assumption often causes analysts to discard data to achieve more concise (but never instantaneous) encounter occasions. An alternative is to consider continuous-time models that estimate survival based on the actual release or encounter times, as is routinely done in known-fates survival analysis, but rarely implemented in mark-recapture-recovery analyses, where incomplete encounter histories complicate interpretation because re-encounter data are products of survivorship and numerous nuisance variables. For wildlife species subject to harvest, harvest recoveries provide opportunities to obtain substantial amounts of dead-encounter data, but reporting probabilities are almost never perfect. I developed a Bayesian hierarchical framework for dead-encounter data, but because these data provide (at best) a single observation per individual, encounter probabilities are always the product of survival, harvest, and reporting probabilities. If analysts have access to covariates that affect each of these probabilities uniquely (including covariates that affect more than one probability), and if these covariates will allow at least 2 of these 3 probabilities to be estimated near 1, then all 3 components of the joint encounter probability can be reliably estimated. I used simulations to verify the reliability of this approach. Additionally, I used data from millions of marked North America waterfowl, where releases immediately before or during hunting seasons provide individuals that clearly survived until harvest periods, and monetary reward bands provide individuals where reporting probabilities approach 1, to estimate seasonal survival probabilities throughout the annual cycle..
Authors: Rocio Joo (Global Fishing Watch)+; Laura Osborne (Global Fishing Watch).
Session: S-2-4.
Where: G029.
When: 15:30-16:30.
Abstract: Concerns about labor abuses in fisheries are rising. However, the problem has been much neglected in fisheries management in part due to the limited availability of directly observed data on what happens onboard fishing vessels. Vessels can use a large range of space—the vast ocean—with very few spatial locations that could serve for potential inspections — including ports, very few of which inspect for living and working conditions. Here we combine different sources of data (a few hundred forced labor-related reports and AIS tracking data) to understand the connections between fishing vessels that have been reported as forced labor offenders and other vessels they encounter at sea and the ports they visit. Vessels may use at-sea encounters to potentially to exchange crew, transship fish or get fuel, allowing them to remain at sea for prolonged periods. Stochastic Block Models of vessel encounters and ports visited allowed identifying different levels of connectivity between vessels and ports, associated with the flag State of the vessel and that of the vessel owner, the fishing grounds, RFMO authorization, and seafood markets. These patterns of connectivity may identify specific factors that facilitate labor abuses, and may highlight gaps in legislation. By projecting encounters at sea for all vessels into our networks, we were also able to provide a distance of every active vessel to the forced labor vessels as a measure of their potential proximity to forced labor. While Stochastic Block Models are at the core of this presentation, a set of deep learning and positive-unlabeled learning algorithms were used in previous modeling stages to get vessel classes, fishing activity and forced labor inferences, among others. Using a mix of machine learning and mechanistic statistical approaches on vessel movement data we can get a better understanding of the networks of forced labor at sea and the role ports could play in stopping this practice..
Authors: Savannah A Rogers (University of St Andrews)+.
Session: S-3-3.
Where: G049.
When: 10:30-12:30.
Abstract: Spatial capture -recapture (SCR) is a popular and robust method for estimating population density, abundance, and dynamics. In systems where leaving stationary detectors in the field is not practical, data for SCR analyses is often collected with moving detectors (e.g., photo identification surveys from a boat, scat detection dogs). In these cases, the path of the moving detector is typically discretised and converted, post hoc, into a grid of ‘effective’ traps. This approach can be inefficient if a fine grid containing many detectors is needed to make the approximation suitably accurate. We developed a new SCR method by maximum likelihood estimation specifically for this data type with a likelihood formulated for detectors that move in space and time. We present a simulation study comparing estimated precision and computation times for this new method with that of the discretisation approach. We then apply the model to the motivating case study of boat-based photo-id surveys of a population of bottlenose dolphins in the southeastern USA . This method has wide applicability to species sampled via transect surveys in terrestrial and marine systems..
Authors: Guilhem Sommeria-Klein (University of Turku)+; Benoît Pérez-Lamarque (Institut de Biologie de l’ENS (IBENS), ENS-PSL); Santiago Rosas-Plaza (Universidad Nacional Autónoma de México (UNAM)); Leo M Lahti (University of Turku); Hélène Morlon (Institut de Biologie de l’ENS (IBENS), ENS-PSL).
Session: S-4-4.
Where: G037.
When: 14:30-16:30.
Abstract: How host-associated microbial communities evolve as their hosts diversify remains equivocal: how conserved is their composition? Do microbial taxa covary in abundance over evolutionary time? What was the composition of ancestral microbiota? Multivariate phylogenetic models of trait evolution are key to answering similar questions for complex host phenotypes. We extended these models in the context of host microbiota, thereby providing a powerful approach for estimating phylosymbiosis (the extent to which closely related host species harbour similar microbiota), integration (evolutionary covariations in bacterial abundances) and ancestral microbiota composition. We first apply our model to the gut microbiota of mammals and birds. We find significant phylosymbiosis that is not entirely explained by diet and geographic location, indicating that other evolutionary-conserved traits shape microbiota composition. We also find remarkably consistent evolutionary covariations among bacterial orders in mammals and birds, indicating that some aspects of microbiota composition were conserved over 150 millions years of host evolutionary history. We then ask whether a similar phylosymbiosis pattern could also be found within one host species and over evolutionary timescales 3 orders of magnitude shorter, namely across modern human populations. Indeed, the rapid adaptation of our species to nearly all terrestrial habitats over the last 150,00 years may have been facilitated by the gut microbiota, and in turn this recent evolutionary history probably played a role in shaping the variations in gut microbiota composition observed across modern populations. We compare the gut microbiota composition of 24 representative populations with minimal industrialization influence from across Africa, Asia and the Americas in light of their evolutionary history. To this end, we reconstruct a tree of divergence times between these populations. We find significant phylosymbiosis, and that evolutionary history explains a larger share of microbiota variation between populations than differences in the mode of subsistence..
Authors: Mohcen Menaa (University of Souk Ahras)+; Abdelkader Djouamaa (University of Souk Ahras); Moundji Touarfia (University of Souk Ahras); Kaouther Guellati (University of Souk Ahras); Mohamed Cherif Maazi (University of Souk Ahras).
Session: S-7-4.
Where: G029.
When: 14:30-16:30.
Abstract: Riparian areas are among the most complex, dynamic, and rich ecosystems in terrestrial biomes. These landscapes are typically characterized by a long-lasting history of intensive land use and human disturbances, with major trends having been related to: river regulation, the establishment of dams, and urban development. Protecting and promoting the rehabilitation of riparian areas throughout the world is crucial for maintaining the integrity of river ecosystems; consequently, it is a central issue in applied ecology. In this study, a multivariate synecological approach was involved to understand the relative contributions of environmental drivers in the forest bird community assembly of riparian forest habitats of Medjerda River. We used three complementary methods (Redundancy Canonical Analysis, a variation partitioning approach based on partial RDA, and a multivariate regression tree with indicator species). We conducted the first bird survey in this area using the point count method to analyze the spatial distribution of breeding riparian birds among habitats with respect to habitat variables, summarising woodland physiognomy and landscape-scale variables. A total of 89 species were observed where the avian species richness at each point-count ranged between six and 17 species. We noted 22 protected species, only three endangered species, and five endemic species to the Maghreb and/or to North Africa. The presence of these species with patrimonial value reinforces the importance of the conservation of Medjerda avifauna Multivariate synecological analysis showed that two major patterns of relationships among birds and habitat were traced: the first involved changes in tree structure during their growth (height of tree layer and large timber), the second was related to characteristics associated with shrub layer. We found according to GLM analysis with a Poisson distribution that the diameter of the largest timber was shown to be the initial key component in determining bird diversity, species richness, and abundance..
Authors: Rachel Mawer (Ghent University)+; Ine Pauwels (Research Institute for Nature and Forest (INBO)); Jelger Elings (Ghent University); Stijn Bruneel (Research Institute for Nature and Forest (INBO)); Johan Coeck (Research Institute for Nature and Forest (INBO)); Peter Goethals (Ghent University).
Session: S-6-2.
Where: G049.
When: 10:30-12:30.
Abstract: Interest in combining step selection functions and hidden Markov models is growing, as a comprehensive way to analyse animal movement behaviour and habitat selection. Outputs from combined hidden Markov model and step selection function analyses can later be used to simulate animal movement in an individual based model, enabling the prediction of movement and spatial usage under different conditions. Here, behavioural state-specific habitat selection is analysed in two freshwater fish, the barbel Barbus barbus and grayling Thymallus thymallus, in an area around 400 m downstream of a riverine barrier with a fish pass. Hidden Markov models were used to define two behavioural states and step selection functions were applied to individual fish to determine state-specific habitat selection. Model results were explored to determine whether differences in habitat selection existed between states. Models were then cross validated using simulations to assess the model’s predictive ability. Lastly, outputs were used to parameterise multiple unique agents for simulation in a state-switching individual based model to predict fish spatial usage under different riverine discharges. In the exploratory analysis, little difference existed in habitat selection between states and individual variation in habitat selection was high. Predicted spatial usage of fish in the study site was typically high directly downstream of the barrier under the tested conditions, indicating fish will often be attracted to this area over the fish pass. High usage was also predicted for barbel around the fish pass but less so for grayling, suggesting that fish pass attractiveness needs to be improved for grayling. With high individual variation present in the studied animals, there is a need for individual based approaches for predicting. Combining hidden Markov models and step selection functions provides an accessible route to simulating movement of unique individuals, enabling predictions of animal spatial usage to inform management..
Authors: Rachel Mawer (Ghent University)+; James Campbell (Leibniz-Institute of Freshwater Ecology and Inland Fisheries); Jelger Elings (Ghent University); Ine Pauwels (Research Institute for Nature and Forest (INBO)); Peter Goethals (Ghent University); Stijn Bruneel (Research Institute for Nature and Forest (INBO)).
Session: S-3-2.
Where: G043.
When: 10:30-12:30.
Abstract: Step selection functions (SSFs) are a valuable tool for analysing animal movement patterns, combining animal movement and habitat selection into one framework. With improving telemetry technologies, movement data can be collected on increasingly fine scales e.g. every few seconds. Such scales may be too fine in comparison to available environmental data resolutions while positioning errors may further impact SSF analysis, with greater impacts at fine scales. As such, there is need to explore the interplay between movement scale, environmental scale, and positioning error upon SSF accuracy. Here, we simulated animal movement across several movement scales with defined habitat preferences to explore SSF accuracy at different movement scales relative to the environmental scale. Positioning error was additionally simulated, to mimic positioning errors encountered in acoustic telemetry systems. Conditional logistic regression models were fitted to simulated tracks to estimate habitat preference. True and estimated habitat preference were compared to evaluate SSF accuracy at each movement scale and the impact of positioning error. At finer movement scales relative to the environmental resolution, habitat preference estimates were less accurate and varied greatly. By comparison, coarser movement scales (where mean step lengths were >5 times the environmental resolution) had more precision in habitat preference estimation. Positioning error reduced SSF performance with a stronger impact at finer movement scales. Researchers seeking to apply SSFs need to ensure that their environmental data and movement data are available on compatible scales. In particular, there is a need to ensure the majority of step lengths exceed the size of the environmental resolution. The results of this study illustrate the effect of data format and quality upon SSF outcomes, providing insight to a key aspect to be considered prior to fine-scale habitat selection analysis..
Authors: Niccolo Anceschi (Duke University)+; Federica Stolf (University of Padua); Gleb Tikhonov (University of Helsinki); Otso Ovaskainen (University of Jyväskylä); David Dunson (Duke University).
Session: S-8-2.
Where: G049.
When: 11:00-12:45.
Abstract: Accurate understanding of joint species distributions in ecological communities is an urgent need under the ongoing global environmental change and biodiversity crisis. A particular interest lies in studying the impact on rare and endangered species, many of which are either completely unknown to science or so little studied as to be essentially unknown. However, meaningful statistical inference on such taxonomic units is greatly hindered by their extreme rareness. Prominent Joint Species Distribution Models (JSDMs) rely on latent factor multivariate probit regressions, and address this issue by borrowing information among species and informing clever prior structures with taxonomic or phylogenetic similarities. Despite this, the persisting statistical and computational difficulties associated with the large overall number of taxa and the little information available on rare species often lead to the complete disregard of the latter. To overcome these limitations, we propose a new class of multi-layer JSDMs that enhance borrowing by leveraging more thoroughly taxonomic associations. We sequentially model taxa occurrences at every level of the taxonomic tree, using inferred values of the regression coefficients and factor loadings to inform those of children nodes, while allowing flexible variability. As the pooled information at higher levels of the taxonomic tree is expected to be less noisy, we target a modular posterior that prevents information from flowing from children nodes to parents. Furthermore, we achieve unprecedented computational efficiency by combining recent advances in latent factors pre-estimation and provably accurate approximate Bayesian inference for high-dimensional multivariate probit regression. This allows us to include in the analysis all 27954 different species of the GSSP fungal biodiversity dataset, where state-of-the-art predecessors could only handle a few hundred most common species..
Authors: Dominic P.D. Grainger (University of Sheffield)+; Paul Blackwell (University of Sheffield).
Session: S-1-2.
Where: G049.
When: 14:00-15:00.
Abstract: Movement ecologists tend to model individual-level animal movement data using methods formulated in discrete time, such as the Hidden Markov Model (HMM). Although efficient, such approaches assume that changes in animal behaviour may only occur at some predetermined grid of times (typically observation times). This is problematic when dealing with temporally irregular data, when comparing separate analyses on different timescales, or when faced with an animal which changes behaviour frequently. Models formulated in continuous time, which are scale invariant, avoid such complications but often lack the computational efficiency, and simplicity, of their discrete-time counterparts. This has hindered their uptake with ecologists.
In my talk, I will detail some of my work which seeks to address the inaccessibility of models for continuous-time inference by rigorously approximating existing ‘exact’ methods. I will demonstrate the accuracy and efficiency of my approach for a range of simulated and real data..
Authors: Amber Cowans (University of St Andrews)+; Albert Bonet Bigata (University of Aberdeen); Chris Sutherland (University of St Andrews).
Session: S-4-4.
Where: G037.
When: 14:30-16:30.
Abstract: Multispecies occupancy models have become a popular framework to jointly model species distributions while simultaneously accounting for environmental factors and imperfect detection. However, these models have been recognised to perform poorly using smaller, but realistic, sample sizes due to convergence and estimation issues. This has prompted the implementation of penalised likelihood approaches, introducing small amounts of bias to increase predictive ability. Despite wide application, there currently is no formal evaluation of how multi-species occupancy model performance is influenced by sample size and configuration, nor how the benefit of using penalised likelihood frameworks varies under different sample size scenarios. To investigate this, we conducted an extensive simulation study to test the model’s ability to recover known co-occurrence patterns. We fit co-occurrence models to simulated datasets under varying sample size scenarios, while iteratively increasing model complexity in two parameter dimensions: number of covariates and number of interacting species. For every scenario, we simultaneously implemented both standard and penalised likelihood approaches to explicitly compare the inference-prediction trade-off. Through this, we effectively demonstrate that the ability of co-occurrence models to uncover abiotic and biotic interactions is sensitive to both sample size and model complexity. Even under the simplest parametrization, we obseve high bias and low coverage in natural parameters (those used for inference) for sample sizes below 200 sites. Penalised likelihood generally outperforms log-likelihood below sample sizes of 200-300 sites. For smaller sample sizes, biases in the general parameters or derived quantities (those of predictive interest) are lower than for natural parameters, suggesting that while systematic biases compromise ecological inference, the predictive ability is less affected. We conclude by providing clear user guidelines for model interpretation and the suitability of alternative model fitting frameworks with respect to sample size and model complexity, increasing the utility of multispecies occupancy models for ecologists exploring species co-occurrence..
Authors: Mollie E Brooks (DTU Aqua)+.
Session: S-5-1.
Where: G037.
When: 10:30-12:30.
Abstract: Body size is typically affected by environmental variables such as population density and weather, but forecasting these effects is not always possible. We investigated the potential for forecasting weight-at-age of 27 fish stocks. We investigated mechanistic Von Bertalanffy and Gompertz growth models as well as linear mixed (LMMs) and generalized additive mixed models (GAMMs). We used glmmTMB to fit GAMMs, a new capability of the package. For model selection of LMMs, we tried both AICc and LASSO methods. For model selection of GAMMs, we used AICc. Our forecast evaluation procedure redid model selection for each validation dataset. We compared the model predictions to a naïve 3-year average. We found that with 15 out of 27 fish stocks, at least one method had potential to forecast better than the naïve method. No single method performed the best across all stocks..
Authors: Johanna de Haan-Ward (University of Western Ontario)+; Simon Bonner (University of Western Ontario); Danielle Ethier (Birds Canada); Douglas Woolford (University of Western Ontario).
Session: S-6-4.
Where: G037.
When: 10:30-12:30.
Abstract: Citizen science monitoring programs, such as the Breeding Bird Survey, provide a wealth of data which can improve our understanding of the long-term changes in the abundance and distribution of species. However, the resulting data may contain only a few records of rare species, making it difficult to fit models of habitat preference using traditional approaches if the dataset is large. To overcome this challenge, we propose a new method for fitting occupancy models to large, imbalanced datasets. In our approach, the original dataset is subsampled to produce a smaller dataset allowing for more efficient modelling by keeping all sites with at least one detection in a given season, along with a random sample of the sites with no detections. Occupancy models cannot be fit directly to the subsampled data because the assumption of binomial sampling no longer holds. However, we show that the occupancy probability is adjusted by a simple offset, meaning that inference on the effects of predictors is still valid, and propose a method for fitting the model that uses the expectation-maximization algorithm to estimate the detection and occupancy parameters along with the bootstrap to compute standard errors. We apply this method to observations of Canada warblers (Cardellina canadensis) from the Breeding Bird Survey in Ontario, Canada, where 95% of sites have zero detections of Canada warblers in a given year between 1997 and 2018. By applying the above sampling method, we can accurately estimate the occupancy and detection parameters using just 10% of the original dataset, including estimating the effects of habitat covariates such as the proportion of the site area which is agricultural land and the length of water body borders in the site..
Authors: Bert van der Veen (Norwegian University of Science and Technology)+; Robert B O’Hara (NTNU).
Session: S-8-2.
Where: G049.
When: 11:00-12:45.
Abstract: Joint species distribution models are commonly used to analyze the presence of many species simultaneously, and to quantify species’ environmental niches. For non-binary responses, Generalized Latent Variable Models fulfill the same role. Often there is little information available for many species, so that extra information is needed to successfully estimate species responses. To that end, phylogenetic random effects can help to borrow information from closely related species that occur more frequently, in order to determine the response of less abundant species.
Few software packages exist to quickly and flexibly implement such models, and the ones that do exist most commonly rely on Markov chain Monte Carlo for estimation (and are thus slow). Here, I present an implementation of such models using the gllvm R package, which uses Variational Approximations (VA) for estimation. For many species, VA tends to be faster for model fitting than the Laplace Approximation, due to a few favorable properties that can be exploited here. The implementation further applies a Nearest Neighbour Approximation and uses Template Model Builder for parallel computations to further speed things along..
Authors: Eleanor Jackson (University of Reading); James Bullock (UK Centre for Ecology & Hydrology); Emma Gardner (UK Centre for Ecology & Hydrology); Tord Snäll (Swedish University of Agricultural Sciences); Rebecca Spake (University of Reading)+.
Session: S-6-3.
Where: G029.
When: 10:30-12:30.
Abstract: Addressing the coupled threats of catastrophic climate change and biodiversity loss requires implementing conservation and restoration actions at scale. However, on-the-ground action is hindered by context dependency, the ubiquitous challenge that implementation outcomes vary from place to place, due to complex dependencies among ecological drivers. Policymakers and practitioners recognise the need to tailor solutions to contexts, and target actions to places where they will work effectively. To provide information for decision making, applied ecologists can learn from established precision disciplines of medicine and marketing, which aim to provide healthcare tailored to individual patients, and advertisements targeting individual tastes. These precision disciplines exploit big data and rapidly developing computational advances to predict unit-specific treatment effects, predictions about the causal effects of particular actions on individual units. For ecology, individual units are not people, but sites such as forests, lakes, grasslands, and arable fields. Unit-specific predictions could target restoration actions such as tree planting or assisted regeneration to sites where, for example, the most carbon would be sequestered, or to where the treatment would contribute most to biodiversity conservation and to avoid sites where potentially adverse effects could occur. Here we road-test a promising ‘precision approach’ for ecology - the application of meta-learner algorithms - using simulations of soil carbon responses to alternative management actions across Sweden. We take a virtual ecologist approach and quantify the influence of sampling and modelling decisions on the accuracy of unit-specific predictions, representing the effects of management on soil carbon at individual forest sites. We provide guidance to maximise the accuracy of predictions for different kinds of observational datasets collected under different sampling strategies..
Authors: Ana Couto (BioSS)+; Fergus J Chadwick (University of St Andrews); David L Miller (BioSS/UKCEH); Thomas Cornulier (Biomathematics and Statistics Scotland); Janice Scheffler (UKCEH); Peter Levy (UKCEH); Jackie Potts (BioSS).
Session: S-1-4.
Where: G029.
When: 14:00-15:00.
Abstract: Combining data at differing spatial or temporal scales presents multiple challenges for modellers. The scale and resolution at which data are collected (the “support”), as well as how data are collected (the observation process), influences perceived variation in the systems of interest and our ability to understand the underlying processes. Observation processes can be as complex as the physical or biological processes we are interested in, and when not accounted for it can result in misleading results. Thus, methods are needed that combine data sources observed at different locations or scales, in continuous space or in time, while accounting for the observation process. In this project, we developed a set of models for data fusion using hierarchical Generalized Additive Models (HGAMs) to efficiently combine data recorded at different scales. This class of models provides a rich suite of models that vary in their flexibility, allowing for sharing of information in different ways (grouping, smoothness or both), depending on what is most appropriate for the given situation. Due to the method generality, these models can be applied in various areas in ecology and environmental sciences. These models were tested using realistic simulations (created using an R package, which gives the user flexibility to simulate and sample from realistic spatio-temporal scenarios) and found to be robust to handle sparse data collected at different scales..
Authors: Alison Johnston (University of St Andrews)+; Michael Schrimpf (Cornell University); Wesley M. Hochachka (Cornell Lab of Ornithology); Charles Eaton (University of Manchester); Steve Buckland (University of St Andrews); Charlie Wright (Independent).
Session: S-8-4.
Where: G029.
When: 11:00-12:45.
Abstract: Distance sampling is a commonly used framework to estimate detectability in animal surveys. One of the key assumptions is that animals don’t move and violation of this assumption typically leads to more animals detected and a positively biased estimate of abundance. Strongly directional movement of animals can cause an additional bias as it typically distorts the distribution of observations across detection distances. For example birds flying in a perpendicular trajectory towards the transect line will be often recorded at further distances as they become detectable before they are near the transect line. Different directions and speeds of travel will cause different patterns of distortion to the estimated detection function. Therefore directional movement biases both the number of animals detected and the estimated detection function, likely causing a bias in abundance and density estimates.
We developed an approach to estimate the underlying true detection function, accounting for directional movement of animals. Using information on the angle and speed of animal movement, we developed a maximum likelihood estimator to infer the true underlying detection function in the presence of directional animal movement. This estimator relies on deterministic relationships between the direction and speed of animals and the expected distortion of distances at which animals are observed.
We demonstrate this estimator with simulated data and real data collected from Antarctic seabirds. Seabirds often move directionally and much faster than boats, which violates the assumptions of distance sampling. Seabirds are often surveyed with distance-sampling on one side of a boat and the bird movement is typically ignored or these individuals are removed, both of which bias any inference. We show that our method can be used on these data and that with simulations it uncovers the true underlying detection function. We discuss assumptions and limitations of this approach. .
Authors: Thomas Bartos (Centre for Environment, Fisheries and Aquaculture Science (Cefas))+; Michael A Spence (Cefas); Jon Barry (Centre for Environment, Fisheries and Aquaculture Science (Cefas)); Robert Blackwell (Alan Turing Institute); James Scott (Centre for Environment, Fisheries and Aquaculture Science (Cefas)); Sophie Pitois (Centre for Environment, Fisheries and Aquaculture Science (Cefas)).
Session: S-3-1.
Where: G029.
When: 10:30-12:30.
Abstract: Frequently, machine learning algorithms are used to classify images from environmental monitoring surveys into categories of interest. These predictions are then used to produce counts. However, classification errors by the algorithm can cause bias in the predictions. I will explain recent work using confusion matrices of machine learning image classifiers to obtain estimates of counts of objects of interest. The core of the approach is a Bayesian model for how the true counts are generated and how items are classified by the machine learning algorithm. The Bayesian approach infers parameters of interest from observed data and quantifies the uncertainty in these estimates. I will demonstrate the method using simulated data with different distributional assumptions for the underlying counts in different categories, as well as on an example of real data from a plankton survey in the Celtic Sea and English Channel..
Authors: Natasha J Klappstein (Dalhousie University)+.
Session: S-3-2.
Where: G043.
When: 10:30-12:30.
Abstract: Step selection functions (SSFs) are popular models that can jointly estimate animal movement and habitat selection. SSFs assess how animals make step-by-step decisions, based on both environmental features and movement constraints. Within the SSF framework, it is often of interest to model inter-individual differences, account for unexplained structure, and capture temporally-varying patterns of movement and habitat selection. However, these aspects can be challenging to model. We show how formulating SSFs as generalised additive models provides a unifying solution, which can be conveniently implemented in the mgcv R package. This approach is highly flexible and allows for various non-linear and random effects, and in this talk, we focus on some particularly relevant examples. Spatial random effects can be used to account for missing environmental covariates or unknown centres of attraction. As an alternative to simpler random slopes, hierarchical smooths can capture inter-individual variation in non-linear effects. Lastly, varying coefficients can allow movement or habitat selection to smoothly vary through time, capturing behavioural variability. We showcase the various modelling options with case studies and highlight how these options can improve ecological inferences..
Authors: Daniel Fink (Cornell University)+; Courtney L Davis (Cornell Lab of Ornithology); Tom Auer (Cornell Lab of Ornithology); Matt Strimas-Mackey (Cornell Lab of Ornithology); Alison Johnston (University of St Andrews); Cynthia Crowley (Cornell Lab of Ornithology); Wesley M. Hochachka (Cornell Lab of Ornithology); Shawn Ligocki (Cornell Lab of Ornithology).
Session: S-6-4.
Where: G037.
When: 10:30-12:30.
Abstract: The increasing volumes of species-observation data being collected by participatory science projects around the world have great potential for use monitoring populations and identifying drivers of population change. However, realizing this potential in a big-data setting is challenging because it requires methods that can 1) control for the confounding sources of variation common in large participatory science datasets that lack the structure to ensure consistent sampling in time and space and 2) estimate heterogenous spatial patterns of population change that arise when multiple spatially dependent drivers (e.g. change in land use and climate) affect species populations simultaneously across large spatial extents.
We investigate the use of the R-learner, a novel framework designed to address these challenges using off-the-shelf statistical and machine learning models. The R-learner is a two-step algorithm that first estimates nuisance functions, the mean outcome and propensity scores, to form a loss function that isolates the effect of interest, here, population change. In the second step this loss is optimized to estimate patterns of population change. The R-learner’s unique algorithmic flexibility allows us to harness machine learning models and large feature sets to accurately predict nuisance functions, while using statistical models to meet inferential objectives about the spatial patterns of population change.
In this study, we investigate how accounting for spatial correlation in the R-learner can strengthen inference about drivers of population change. Using simulations, we evaluate the model’s ability to estimate spatially varying patterns of population change, identify the correlates associated with these patterns, and assess the value of accounting for spatial correlation in the effect estimates. Results show how including spatial correlation improves performance estimating population change and identifying the correlates of change. Finally, we illustrate how the approach can be used to study population change using data from the participatory science project eBird.
Dear Reviewers: If our presentation is not accepted for a talk, we would be happy to present as a poster.
.
Authors: Ron R. Togunov (Norwegian University of Science and Technology)+; Sam Perrin (Norwegian University of Science and Technology ); Philip S Mostert (Norwegian University of Science and Technology); Bram Van Moorter (The Norwegian Institute for Nature Research); Manuela Panzacchi (The Norwegian Institute for Nature Research); Anders Finstad (Norwegian University of Science and Technology ); Rannveig Margrete Jacobsen (The Norwegian Institute for Nature Research ); Joseph Chipperfield (The Norwegian Institute for Nature Research ); Kwaku Peprah Adjei (Norwegian Institute for Nature Research); Robert B O’Hara (NTNU).
Session: S-3-2.
Where: G043.
When: 10:30-12:30.
Abstract: Accurately predicting at-risk species’ habitats is crucial for effective conservation and management. Traditional approaches to habitat modelling using occurrence data, often the most readily available data type for many species, rely solely on species distribution models (SDMs). However, these may not adequately account for landscape connectivity and environmental barriers that fragment habitats. Connectivity modeling has proven to be particularly effective in identifying functional habitats—those both suitable and well-connected. However, this approach relies on accurately quantifying habitat suitability and environmental resistance to movement, which is difficult using occurrence data. This limitation is further exacerbated by the sampling bias inherent in most occurrence data sets such as citizen science, and dearth of data for rare species. To address these challenges, we propose merging integrated SDMs (iSDMs) with Randomised Shortest Path (RSP) connectivity models for predicting functional habitat using occurrence data. We apply our combined iSDM and RSP model to model habitat connectivity of saproxylic Coleoptera (wood-associated beetles) in Norway using opportunistic and systematic occurrence data. Our methodology leverages iSDMs for modelling occurrence data from multiple sources and across several species and RSP models for integrating the effects of landscape structure on habitat function. We believe our approach can provide more refined predictions of functional habitats and help facilitate targeted conservation of critical ecosystems and the species they support. .
Authors: Thomas Cornulier (Biomathematics and Statistics Scotland)+; David L Miller (BioSS/UKCEH); Kate R. Searle (UKCEH); Charlotte Regan (UKCEH); Maria Bogdanova (UKCEH).
Session: S-4-3.
Where: G043.
When: 14:30-16:30.
Abstract: Inference about spatial or temporal processes, e.g. how much predictors may explain spatio-temporal variation in a system of interest, is often complicated by spatial and/or temporal scale mismatches. Mismatches can occur either in the scale at which response and predictor data have been gathered, or in the scale at which ecological and environmental processes take place. For example, local responses of interest may integrate the influence of environmental variables over a broader range of spatial and/or temporal scales. In this presentation, we introduce how signal (vector on scalar) regression provides a helpful framework for identifying the spatial or temporal extent of the “zone of influence” of predictors on local response. We then show how this simple framework can be extended to ask questions about more complex lagged processes, including multi-level, non-linear or spatio-temporal lagged effects, and different types of lagged predictor interactions. We illustrate the application of such model structures to understanding how seabird breeding productivity may be explained by the influence of environmental variables (such as prey availability proxys), integrated over a-priori unknown areas of sea and temporal windows..
Authors: Michael J Thomson (Centre for Environment, Fisheries and Aquaculture Science)+; Michael A Spence (Cefas).
Session: S-3-4.
Where: G037.
When: 10:30-12:30.
Abstract: It is often the case that multiple complex ecosystem models exist of a system of interest where each model can produce significantly different outputs. However, while some models can capture parts of the ecosystem better than others typically no one model is uniformly better than the others. Making management decisions based on a single model is sensitive to the model chosen. This calls for a means to combine predictions from the ensemble of available models, to learn from the strengths and weaknesses of each, to obtain the best estimates of the ecosystem processes of interest. EcoEnsemble is an R package that implements a flexible ensemble modelling approach. A statistical model describes the relationship between the models in the ensemble, the true underlying ecosystem variables, and the observations. The statistical model exploits the strengths and discounts the weaknesses of the individual models to produce a single prediction without making the strong assumption that any single model is correct. The model is fitted using Bayesian methodology allowing for rigorous quantification of uncertainty. However, to use the Bayesian approach requires the definition of prior distributions. The need to define priors for the complex statistical model implemented in EcoEnsemble creates a barrier to wider use of the ensemble modelling approach. We present the results of a comprehensive simulation study, testing the effect of different prior specifications in different scenarios, to determine well-defined default priors and guidelines for prior choice for users of the EcoEnsemble package. These default priors and guidance allow for easier uptake of the ensemble modelling approach and EcoEnsemble package for new users and systems. The use of these guidelines is demonstrated applied to an ensemble of multispecies ecosystem models for North Sea fisheries..
Authors: Robert B O’Hara (NTNU)+; Bert van der Veen (Norwegian University of Science and Technology).
Session: S-8-2.
Where: G049.
When: 11:00-12:45.
Abstract: Model-based ordination has developed over the last decade to become a viable platform for the analysis of ecological communities. These models extend generalised linear model by assuming a residual covariance that can be written as a sum of a smaller number of latent variables. Here we extend the approach to incorporate covariates that influence both site and species effects. This means we can efficiently model trait and environmental effects as influencing latent variables rather than species individually.
Fitting these models can be difficult. But because, conditional on either the site or species effects, the model is a GLM, we have developed an approach that uses repeated calls to INLA to fit the models efficiently. We will describe the models and the fitting approach, and indicate how it might be extended..
Authors: Fergus J Chadwick (University of St Andrews)+; Daniel Haydon (University of Glasgowe); Dirk Husmeier (University of Glasgow); Jason Matthiopoulos (University of Glasgow); Otso Ovaskainen (University of Jyväskylä).
Session: S-6-4.
Where: G037.
When: 10:30-12:30.
Abstract: Citizen science data often contain high levels of species misclassification that can bias inference and conservation decisions. Current approaches to address mislabelling rely on expert taxonomists validating every record or treating labels as completely unknown. This approach makes intensive use of a scarce resource and reduces the role of the citizen scientist. 2. Species, however, are not confused at random. If two species appear more similar, it is probable they will be more easily confused than two highly distinctive species. Identification guides are intended to use these patterns to aid correct classification, but misclassifications still occur due to user-error and imperfect guidebook design. Statistical models should be able to exploit this non-randomness to learn confusion patterns from small validation data-sets provided by expert taxonomists, yielding a much-needed reduction in expert workload. Here, we use a variety of Bayesian multivariate hierarchical models to probabilistically classify species based on the species-label provided by the citizen scientist. We also explore the utility of guidebooks provided by the citizen science schemes as a prior for species similarity, and hence draw conclusions for their future improvement. 3. We find that the species-label assigned to a record by a citizen scientist, even when incorrect, contains useful information about the true species-identity. The citizen scientists correctly identify the species in around 58% of records. Using models trained on only 10% of these records (validated by experts), we can correctly predict species-identity for 69 (90%CI: 64-73)% of records when the guidebook is used, vs 64 (58-69)% for models that do not use the guidebook. The fact that misclassifications can be predicted systematically indicates that improvements could be made to the guidebook to reduce misclassification. 4. By using Bayesian, hierarchical models we can greatly reduce the workload for experts by providing a probabilistic correction to citizen science records, rather than requiring manual review. This is increasingly important as the number of citizen science schemes grows and the relative number of taxonomists shrinks. By learning confusion patterns statistically, we open up future avenues of research to identify what causes these confusionsand how to better address them..
Authors: Adam Butler (Biomathematics and Statistics Scotland)+; Anastasia Frantsuzova (BioSS); Kate Searle (UK Centre for Ecology and Hydrology); Esther L Jones (Biomathematics & Statistics Scotland); Francis Daunt (UK Centre for Ecology and Hydrology); Eleanor Skeate (ABPmer); Bob Furness (MacArthur Green); Annette Fayet (NINA); Maria Bogdanova (UK Centre for Ecology and Hydrology); Tone Reiertsen (NINA); Ana Couto (BioSS); Charlotte Regan (UK Centre for Ecology and Hydrology); Oliver Leedham (UK Centre for Ecology and Hydrology).
Session: S-1-3.
Where: G037.
When: 14:00-15:00.
Abstract: Energy production via offshore wind is expanding rapidly, with the UK and other countries having set ambitious targets for future growth of the sector, motivated by policies to ensure energy security and mitigate climate change. Offshore windfarms have the potential to impact seabird populations through both direct effects (collision) and indirect effects (displacement) and as the number of windfarms increases, there is potential for the cumulative impacts of these developments to be substantial. Potential strategic conservation measures that could compensate for these cumulative effects have been proposed. Existing empirical evidence of the effectiveness of these measures is limited, creating a key knowledge gap around the extent to which these measures are capable of compensating for the effects of offshore windfarms, and of achieving biodiversity net gain.
Expert elicitation is a widely used approach in situations where direct empirical evidence is limited. This talk describes an elicitation of the effectiveness of potential compensatory measures conducted as part of the ECOWINGS project (Ecosystem Change, Offshore Wind, Net Gain and Seabirds) funded through the EcoWind programme. This elicitation involved a number of interesting non-standard features, requiring bespoke solutions, including (a) the objectives of the elicitation requiring the experts to identify the list of potential measures to be considered, as well as quantifying their impact, with the result that the number of questions was unknown at the outset; (b) ranges of the values being elicited depending on unknown information, and (c) complex dependencies between responses to different questions. We outline the objectives and structure of the expert elicitation exercise and the approaches we took to address the non-standard features of it, before presenting the results and outlining their relevance to decision making. We also highlight the wider relevance of the work to the implementation of expert elicitation in the context of complex ecological problems..
Authors: Abinand Reddy Kodi (CREEM, University of St Andrews)+.
Session: S-3-3.
Where: G049.
When: 10:30-12:30.
Abstract: Spatial capture recapture (SCR) has established itself as the gold standard for abundance estimation of animals. With respect to abundance estimation, the model has proven itself to be flexible to a variety of problems and robust to even poor model specification. Using SCR for inference on additional population processes such as resource selection, habitat use, and movement are active areas of research. Many of these methods, however, are limited by the sparseness of data generated by SCR surveys and need to be bolstered by auxiliary data such as telemetry tags on individuals within the surveyed population. While recent studies have developed models to harness these opportunities , there is a lack of a general framework that allows for researchers to build SCR models integrating multiple data types. Notably, much of these are fit using Bayesian methods. In this study we develop a general framework for integrating multiple data sources in SCR within a maximum likelihood framework and contextualise existing models as a subset within the developed framework. Finally, we present a case study using an integrated model to investigate dispersal of snow leopard cubs within a landscape. Our results show that adding telemetry data yields little benefit to inference on density estimates but greatly improves our understanding of space use of individuals to more accurately estimate the overlap of individuals’ home ranges..
Authors: Marwan Naciri (CEFE)+; Jon Aars (Norwegian Polar Institute); Sarah Cubaynes (EPHE).
Session: S-4-1.
Where: G029.
When: 14:30-16:30.
Abstract: Making use of as much of the available data as possible is critical when studying wild populations, as limited data is often a constraint for analysis, inference, and management decision-making. Combining several data types can be beneficial as it may not only improve the precision and accuracy of estimations, but also make new parameters accessible. In polar bears, data from tracking collars and geolocation tags allow to detect maternity denning and thus provide information of reproductive status. Here, we build a custom multi-event capture-recapture model that makes use of this data on reproductive status (‘remote captures’, along with physical captures) in a Bayesian framework. This model allows to separate breeding probability into denning probability and early litter survival. Using simulations, we quantify the benefits of integrating remote captures, then we apply our model to a dataset on Svalbard polar bears from 1987-2023. Specifically, we estimate age-specific vital rates and assess the existence of a cost of reproduction in this species with extended parental care. We find that virtually all vital rates increase in early life and decrease in late life, with most four-year old females not even attempting to den. Further, we find that females who attempted to reproduce the previous year have a lower denning probability compared to females that were previously non reproducing, indicating a cost of reproduction. However, we also find that females who complete their three-year reproductive cycle have a denning probability even higher than previously non-reproducing females. This suggests marked individual heterogeneity in quality, with some females performing consistently better than others in their reproduction.
.
Authors: Nilanjan Chatterjee (senckenberg biodiversity and climate research centre)+; Christen Fleming (University of Central Florida); Jesse Alston (University of Arizona); Justin Calabrese (CASUS); John Fieberg (University of Minnesota).
Session: S-3-2.
Where: G043.
When: 10:30-12:30.
Abstract: Habitat selection stands as a pivotal aspect of animal ecology, crucially informing species conservation management decisions. Technological advancements present significant opportunities for collecting both animal location data and remotely sensed environmental data at increasingly fine scales. Resource Selection Functions (RSFs) represent one of the most commonly employed statistical tools for inferring relationships between animal locations and environmental data. However, finer temporal scales in location datasets often lead to higher levels of autocorrelation and pose significant challenges for approximating the background habitat availability. In this study, we explore how the selection strength and spatial heterogeneity of covariates influence the approximation error for RSFs. We hypothesize that a higher number of available points would be required for approximating the likelihood with covariates exhibiting greater spatial heterogeneity and lower selection. We simulated four sets of individual tracks characterized by varying levels of selection (high and low) and spatial heterogeneity of covariates (high and low). We estimated the likelihood within a specified error margin leveraging the recently developed model for weighing the autocorrelation in RSF for continuous-time movement models. Our findings indicate that the variance of the log-likelihood decreases with increase in available points. Simulation results demonstrate that rasters with high spatial autocorrelation and increased habitat selection magnitude require a greater number of available points for the likelihood approximation. Moreover, the number of available points’ estimated from this study indicates that prior research utilizing RSF analysis often samples too few available points.These guidance regarding how many integration points are needed to approximate the likelihood within a specified error margin will serve as crucial steps toward generating robust inferences from resource selection studies..
Authors: Esther L Jones (Biomathematics & Statistics Scotland)+; Ana Couto (BioSS); Lila Buckingham (NINA); Maria Bogdanova (UKCEH); Kate Searle (UKCEH); Francis Daunt (UKCEH); Adam Butler (Biomathematics and Statistics Scotland).
Session: S-7-1.
Where: G049.
When: 14:30-16:30.
Abstract: Deployments of offshore renewable developments are rapidly increasing around the UK, in line with ambitious energy security targets. Developments exist or are planned in both near and offshore areas of habitat within the marine environment that can be used by seabirds throughout the year. Seabirds can be impacted through displacement or collision with renewable developments and Special Protection Areas (SPAs) protect breeding colonies. Although tools for attributing space use at-sea in relation to colony provenance (apportioning) are well-established in the breeding season, the current method for apportioning during non-breeding does not account for variability in distribution and does not always make use of the most recent available data. Funded by the Offshore Renewables Joint Industry Programme, we developed a novel tool for apportioning birds to populations in the non-breeding season.
From 2017-2020, a bird-borne geolocation study was conducted on the non-breeding season distributions of guillemot and razorbill at twelve breeding colonies around the north of the UK. Geolocators have large locational uncertainty as daily locations are derived from light levels, saltwater immersion, and sea-surface temperature. Monthly estimates of spatial distributions of guillemot and razorbill from tracked colonies were produced and a simulation approach was used to characterise and propagate locational uncertainty. We developed an approach to apportion impacts of developments within the UK EEZ to SPAs, accounting for colonies that were not tracked in the study and allowing for direct comparison of results with the current approach. The tool was developed in R using a simple interactive interface to allow users flexibility in inputs, and output reports are automatically generated to include visual outputs. Here, we present the analysis and tool, and contextualise the approach to address the challenge of propagating uncertainty through analyses, which is particularly important when at least one source of error is known to be large..
Authors: Miranda Tilberg (Travelers Insurance); Philip M Dixon (Iowa State University)+.
Session: S-3-2.
Where: G043.
When: 10:30-12:30.
Abstract: Are there groups of animals with similar home ranges? This is a clustering question, but the objects to be clustered are bivariate distributions, not scalars or vectors. Although there is a large literature on estimating home ranges for an individual, and some literature on estimating similarity between pairs of home ranges, relatively little work has been done on clustering home ranges. Our key insight is that the Bhattacharyya distance, one estimator of the similarity of a pair of home ranges, is roughly a squared distance. Hence, clustering methods designed to minimize squared distances within clusters can be adapted to cluster home ranges. We develop two new clustering methods for home ranges. One uses Ward’s agglomerative hierarchical clustering; the other uses a variation in k-means clustering. These are illustrated using data on 25 bottlenose dolphins, Tursiops truncatus, tracked in Barataria Bay, Louisiana USA. The quality of the clustering was evaluated using silhouette statistics, the general overlap index, and the probabilistic extension of the general overlap index. The two clustering methods generate similar clusters, which differ somewhat from a previous visual clustering. The cluster evaluation statistics also suggest more clusters (four) than the previous visual assessment (three clusters). We briefly discuss extensions for other pairwise similarity measures..
Authors: Alexander Schindler (University of Saskatchewan)+; Anthony Fox (Aarhus University); Christopher K Wikle (University of Missouri); Bart Ballard (Caesar Kleberg Wildlife Research Institute; Texas A&M University-Kingsville); Alyn Walsh (National Parks and Wildlife Service); Seán Kelly (National Parks and Wildlife Service); Larry Griffin (Wildfowl and Wetlands Trust; ECO-LG Limited); Mitch Weegman (University of Saskatchewan).
Session: S-8-1.
Where: G043.
When: 11:00-12:45.
Abstract: Migratory animals make decisions throughout the year to balance the energetic requirements of movement, survival, and reproduction, but these decision-making processes and the spatio-temporal scales at which they operate are relatively unknown. Technological advances to record an animal’s movements, behaviours, and environment allow unprecedented opportunities for fine-scale study of animal decision-making throughout the annual cycle, improving our understanding of behavioural ecology and conservation management. We combined five years of high-resolution animal movement and behaviour data with remotely sensed weather and habitat data in a Bayesian hierarchical model to quantify the fitness consequences of animal decision-making throughout the full annual cycle in a migratory bird of conservation concern, the Greenland white-fronted goose (Anser albifrons flavirostris). We found that weather and habitat use explained variation in time spent feeding and energy expenditure in Greenland white-fronted geese during spring and autumn. Geese that expended less energy and fed longer during spring were more likely to successfully reproduce. Further, individuals with offspring expended more energy and fed for less time when providing parental care during autumn, potentially representing an adverse fitness consequence of breeding. The strong relationships among environmental conditions, movements and behaviour, and reproductive metrics suggest that this population is vulnerable to continued land use and climate change at multiple phases of the annual cycle. Our analysis provides a statistical framework for hypothesis tests about complex relationships among land use, climate change, and animal decision-making in a variety of taxa, empowering practitioners to incorporate spatially- and temporally-rich information for new perspectives in animal ecology..
Authors: Benjamin M Bolker (McMaster University)+.
Session: S-3-4.
Where: G037.
When: 10:30-12:30.
Abstract: Shape-constrained smooth functions, e.g. enforcing monotonicity or concavity of a function, represent a natural connection between theoretical and statistical ecology. For example, the shape of a a predator’s functional response is of theoretical interest; using families of shape-constrained curves, rather than fitting specific alternatives functions, allows us to test fundamental hypotheses about shapes without being restricted by particular mathematical forms. Such smooth, shape-constrained curves can also be embedded in more complex (multilevel or dynamical) models. I will illustrate these approaches with a re-analysis of data on size-dependent functional response models of red-eyed treefrog tadpoles (from McCoy et al 2011). .
Authors: Alexander Schindler (University of Saskatchewan)+; Anthony Fox (Aarhus University); Alyn Walsh (National Parks and Wildlife Service); Larry Griffin (Wildfowl and Wetlands Trust, ECO-LG Limited); Mitch Weegman (University of Saskatchewan).
Session: S-1-1.
Where: G043.
When: 14:00-15:00.
Abstract: Assessing the impacts of ongoing environmental changes on species of conservation concern requires understanding of population dynamics. By estimating demographic rates of animal populations, researchers can directly test hypothesised mechanisms of how weather and habitat conditions throughout the year affect annual survival, reproduction, dispersal, and ultimately, changes in population size. However, our understanding of population dynamics is complicated when animals aggregate in spatially discrete subpopulations, resulting in demographic heterogeneity among a collection of subpopulations (i.e., a “metapopulation”). By integrating multiple data sets in a single population model, researchers and conservation practitioners can obtain more accurate and precise estimates of demographic rates than by analysing each data type separately. We developed an integrated metapopulation model that incorporated annual capture-resighting, productivity, and population size information from 1983 to 2022 to study environmental drivers of metapopulation dynamics of a declining migratory bird, the Greenland white-fronted goose (Anser albifrons flavirostris). We found that low and declining reproductive rates were the primary mechanism of population decline across the entire metapopulation, and that earlier spring vegetation phenology on staging areas and increased snow on breeding areas decreased reproductive rates. Survival in both adults and juvenile geese was high and stable among subpopulations and years, and adult survival was higher following hunting protection on spring staging areas. We also found that population sizes of large subpopulations were maintained by high numbers of immigrating geese each year, while declines in small subpopulations were accelerated by high annual emigration rates. Extending commonly used integrated population models to a metapopulation framework provided a more complete understanding of population dynamics of this species of conservation concern. Our study provides a reproducible modelling framework for testing hypotheses of environmental drivers of population declines within and among subpopulations across the life history gradient in animal ecology..
Authors: Juan M Morales (UNIVERSITY OF GLASGOW)+; Claudio Bracho Estévanez (Universidad de Cádiz); Juan Pedro González-Varo (Universidad de Cádiz).
Session: S-5-3.
Where: G043.
When: 10:30-12:30.
Abstract: Seed dispersal is fundamental for plant population and community dynamics. Many plant species are dispersed by birds that consume fruits and move the seeds with them for certain time. There has been great progress in data collection and movement modelling of birds foraging for fruits and dispersing seeds, but we still need a better understanding of seed retention times to make predictions about how far and where seeds can go. We use a Bayesian joint species approach to model retention times in different bird and plant species. We used a combination of detailed data from experiments and summary statistics from the literature. Seed size and bird size are the most important variables related to retention time. Seed size also determines whether seeds are regurgitated or defecated, thus strongly impacting on retention time. In general, smaller seeds have longer retention times. We explore how models varying in realism and detail perform in predicting seed retention times for unobserved plant species. Models of intermediate complexity/realism seem to provide the best predicting capacity. Our model of seed retention time could be used in combination with movement models in order to predict seed dispersal distance and properties of the seed rain. However. we still lack data about several physiological processes affecting retention times..
Authors: Wendy Leuenberger (Michigan State University)+; Jeff Doser (Michigan State University); Mike Belitz (Michigan State Univerisity); Leslie Ries (Georgetown University); Nick Haddad (Michigan State University); Wayne Thogmartin (USGS); Elise Zipkin (Michigan State University).
Session: S-7-4.
Where: G029.
When: 14:30-16:30.
Abstract: Insects play critical functional roles in ecosystems and recent work indicates that many insect species are declining. However, determining population and community trends is challenging as data on many species are limited and rates of decline are difficult to measure, especially over long time periods and broad spatial extents. Our objective was to evaluate butterfly species and community dynamics across the Midwestern United States for the last 32 years by leveraging five long-running volunteer-based monitoring programs (Illinois, Ohio, Iowa, and Michigan structured monitoring surveys, and North American Butterfly Association counts) to estimate butterfly abundance indices. We used >4.3 million observations of 136 species to develop integrated community models that estimated spatially-explicit, species-specific activity curves for each year (i.e., weekly abundance through the summer season), treating species parameters as random effects drawn from community-level distributions. The community-level distributions are jointly estimated through a unified, integrated analysis of the available data sources. We generated a yearly relative abundance index for each species by calculating the area under the species and year-specific activity curves and used post hoc linear regression to estimate temporal trends over the three decade time period of our analysis. We also calculated local-level species richness and evenness through time to evaluate changes in community diversity. Our approach provides estimates of both species-level dynamics (abundance, distribution) and community metrics (richness, diversity), delivering guidance on which species and communities of butterflies are declining and if any traits (e.g., migratory status, voltinism, host plant breadth) are associated with these declines. Our findings help inform species status assessments and other management decisions aimed at mitigating insect losses..
Authors: Hannah Worthington (University of St Andrews)+; Emily Dennis (Butterfly Conservation); Byron Morgan (University of Kent); Takis Besbeas (University of Kent).
Session: S-8-3.
Where: G037.
When: 11:00-12:45.
Abstract: Over recent decades, three-quarters of the butterfly species that can be found in the UK have shown declines in distribution, population, or both. These trends are identifiable thanks to long-term monitoring schemes and the increasing availability of large-scale citizen science data. These invaluable sources of information offer the chance to measure changes in species’ occupancy, and butterflies in particular are recognised as valuable indicators of biodiversity. These opportunistic datasets are of the magnitude of several hundred thousand sites, with records spanning several decades. Simple dynamic occupancy models, which account for imperfect detection, are a popular analysis option but often run into difficulties when applied to large datasets such as those collected at Butterfly Conservation. We explore the potential for more efficient model fitting to presence-absence data when standard packages and functions appear to encounter computational limitations. A hidden Markov model (HMM) formulation offers the desired computational efficiency, with the additional bonus of the simplicity in how the large amounts of missing data are handled. We apply the HMM approach to several UK butterfly species covering differing rarities and some that are showing changes in spatial distribution. We provide estimates of species persistence and movement, visualise the range contraction or expansion of species over time, as well as potentially revealing some effects on species distributions linked to climate change..
Authors: Sougata Sadhukhan (Bharati Vidyapeeth (Deemed to be University) Institute of Environment Education and Research)+.
Session: S-8-4.
Where: G029.
When: 11:00-12:45.
Abstract: The traditional line transect method is a cornerstone in wildlife ecology for estimating prey density. However, its application is often limited to theoretical understanding among students due to restricted access to natural habitats like National Parks. To bridge this gap, we developed a novel approach utilising vehicle safaris in Indian tiger reserves as a means of data collection. This method employs a simple Epicollect form with pre-classified distance classes (0-20, 21-50, 51-100, 101-200, 201-400, 401-600, 601-800, 801-1000, 1500, 2000, >2000 meters) to facilitate ease of data recording during safaris. This study outlines the implementation of the vehicle transect method in Kanha (2023) and Pench Tiger Reserves (2024), where students collected substantial data over 2045 km of safari routes, resulting in 2867 prey encounters. The data was analysed using the Distance software, with the predefined distance classes serving as cut points. The resulting prey density estimates were comparable to the Phase IV data of the All-India Tiger Reserve. The potential of this vehicle transect method is multifaceted. It offers practicality and accessibility, enabling students to conduct fieldwork within the constraints of a safari, thereby gaining valuable hands-on experience in data collection and analysis. This innovative approach not only facilitates a practical understanding of detection probability concepts but also serves as a scalable and replicable model for wildlife surveys in protected areas. Additionally, the collected data is suitable for occupancy analysis and species distribution modelling, enhancing the method’s utility as an educational tool. The Epicollect platform’s ease of data collection and validation makes this method an invaluable asset for educational and scientific endeavours in wildlife ecology..
Authors: Andrea Havron (NOAA Fisheries Office of Science & Technology)+; Christine Stawitz (NOAA Fisheries Office of Science & Technology); Matthew Supernaw (NOAA Fisheries Office of Science & Technology); Bai Li (ECS Federal in support of NOAA Fisheries Office of Science & Technology); Kathryn Doering (NOAA Fisheries Office of Science & Technology); Patrick Lynch (NOAA Fisheries Office of Science & Technology); Richard Methot (NOAA Fisheries Office of the Assistant Administrator); FIMS Implementation Team (NOAA Fisheries).
Session: S-3-4.
Where: G037.
When: 10:30-12:30.
Abstract: Open science promotes transparency, innovation, and collaboration, and science agencies around the globe have started prioritizing the need to establish open science policies and practices. One such effort out of NOAA Fisheries is the Fisheries Integrated Modeling System (FIMS), a software development project aiming to centralize and modernize U.S. fishery stock assessment software. Fisheries stock assessments apply integrated population models to project fishery catch available for future harvest and are used worldwide to provide crucial management advice. Therefore, updating current decentralized models to a standardized system reliant on identified best practices in population modeling, statistics, open science, and computer science is a national priority. To meet this need, FIMS is being developed in an open science framework by a team of regional experts working with dedicated programming staff to ensure the system meets regional needs while relying on common standards in software design and development. FIMS is delivered as an R package with compiled C++ code and a portable dependency on TMB. Development follows an agile cycle, iterating between phases of prioritization, development, maintenance, and outreach, incorporating feedback from reviewers and stakeholders from the beginning. The team relies on open source development tools, such as GitHub, for asynchronous project management and Codespaces for real time virtual programming. Through training and small coding teams, the group develops knowledge of the FIMS architecture and C++ programming skills, improving team cohesion and code base maintainability. Here, we present our strategies and tools used to implement this statistical software project with an emphasis on lessons learned when working with a distributed team on open science software design and development..
Authors: Eduard Campillo-Funollet (Lancaster University)+; James Van Yperen (University of Sussex).
Session: S-5-1.
Where: G037.
When: 10:30-12:30.
Abstract: The hares-eat-lynx paradox is a phenomenon observed in a well-established Canadian hares and lynx dataset. Rather than following the classical predator-prey dynamics as expected from these species, in some occasions the prey peak follows the predator peak. The dataset was not collected in a structured survey, but rather as secondary data from the accounts of a fur trading company. We will show that the predator-prey underlying model is still applicable if we incorporate the observation process, and we will discuss the identifiablity of the model parameters.
Our results show that the observed dynamics can still be modelled by allowing time dependent parameters, and we will present a robust inference scheme to model the data. The inference scheme exploits a reformulation of the model to combine expectation-maximisation and coordinate-descent approaches. We will show that our approach allows us to identify changes in the population dynamics, by means of changes in the model parameters, that are difficult to detect without an underlying dynamical model.
.
Authors: Maxime Fajgenblat (KU Leuven & Hasselt University)+; Thomas Neyens (Hasselt University & KU Leuven).
Session: S-6-4.
Where: G037.
When: 10:30-12:30.
Abstract: Natural history collections and opportunistic sightings provide a unique glimpse into species communities of the past. Often, such data are too sparse to directly quantify changes in species assembly over time, with considerable unbalance across space, time and species. Furthermore, opportunistically collected data feature several limitations linked to imperfect and heterogeneous detection. We developed a Bayesian hierarchical model that ordinates species communities through space and time, using a latent factor approach in which the site-loadings are spatiotemporally structured by means of a mixture of Gaussian processes. As such, information is efficiently shared across space, time and species, mediating data sparseness. To flexibly accommodate opportunistically collected data, we embedded this construct within an occupancy-detection model structure to address false positives arising from imperfect and heterogeneous detection, e.g. due to seasonal or observer-induced effects. We applied our model to a large dataset of historical butterfly records in Flanders (Northern Belgium), enabling us to recover changes in butterfly community composition at high spatial resolution in over a century’s time. In addition to visualising the ordinated community trajectories through time, we also quantify the speed of community change and we explore mechanisms driving community change..
Authors: Danielle Harris (University of St Andrews)+; Kerri Seger (Applied Ocean Sciences, LLC); David Mellinger (Oregon State University); Jennifer Miksis-Olds (University of New Hampshire).
Session: S-4-1.
Where: G029.
When: 14:30-16:30.
Abstract: Increasingly, passive acoustic monitoring (PAM) is being used to monitor both terrestrial and marine wildlife. PAM equipment, particularly in the marine environment, can be costly and a closely-spaced array of instruments is often required to estimate the location of vocalizing animals. Therefore, there is often a trade-off at the survey design stage between the number of monitoring stations that can be created across the study area, and the spatial information that can be estimated about the target species. This can often lead to survey designs where single instruments are placed at each monitoring station. Where absolute abundance or density is the goal, such designs typically cannot use standard distance sampling (DS) or spatial capture-recapture (SCR), and other methods are required to estimate detection probability. Here, options for abundance/density estimation are discussed, using examples from recent marine studies. Monte Carlo simulation-based methods have been implemented for blue (Balaenoptera musculus) and fin (B. physalus) whales in the Pacific and Indian Oceans, as part of an ongoing research project “CORTADO”, which is focusing on using opportunistic deployments of marine instruments for marine mammal monitoring. The simulations use sound propagation models, noise level measurements and the assumed source level distribution of the target calls to estimate detection probability. The sensitivity of the results to input parameters such as call source levels and assumed animal distribution have been explored. Future research directions for abundance/density estimation from sparsely distributed instruments will also be discussed, which include (1) the use of “hybrid” survey designs, where only some monitoring stations are comprised of instrument arrays capable of estimating detection probability via DS or SCR and (2) integrating data from another type of monitoring platform to enable detection probabilities of the single instruments to be estimated..
Authors: Tomáš Telenský (Center for Theoretical Study, Charles University in Prague)+; David Storch (Center for Theoretical Study, Charles University in Prague); Petr Klvana (Bird Ringing Centre, National Museum, Prague); Jiří Reif (Institute for Environmental Studies, Faculty of Science, Charles University, Prague).
Session: S-5-2.
Where: G049.
When: 10:30-12:30.
Abstract: Decomposition of population growth rate into demographic parameters is an insightful tool which can significantly shift our understanding of population dynamics. Integrated Population Models are often the best tool to achieve this, decomposing the population growth rate into a full Leslie matrix. This, however, usually requires data which are not often available (e.g. dead recoveries). On the other hand, a simple Capture-Mark-Recapture dataset, much more commonly available, often at continental-scale (e.g. the citizen science projects like EuroCES in Europe or MAPS in North America), can provide a simple decomposition of population growth rate into survival and recruitment. This can be done using the famous Pradel model, which has been used for many taxa and cited in over 600 studies. However, Pradel model has a known limitation – it cannot handle transient individuals (i.e. those nonresident in the area, just passing through). This phenomenon is not negligible, as it is prevalent in many taxa. We present an extension of the Pradel model that accounts for transient individuals. We also present a novel way of parametrizing transience. In CJS models, it is parametrized as a proportion of newly captured individuals. We parametrize it as a proportion of all captured individuals, directly in the model. This, in contrast to the previous parametrization, facilitates biological interpretation and thus opens a new possibility to study transience as a biological phenomenon. Estimating all the parameters in a single model allows to account for their correlated error distributions in follow-up analyses. Our model thus unlocks new possibilities to explore continental-scale CMR datasets, where transience is prevalent, and can serve as a good basic building block for more advanced analyses (e.g. transient LTRE) or in IPMs. We will demonstrate the possibilities of the model on several case studies..
Authors: Margarita M. Rincón (Spanish Institute of Oceanography, National Spanish Research Council (IEO-CSIC))+; Jamie Lentin (Shuttlethread); María José Zúñiga (Spanish Institute of Oceanography, National Spanish Research Council (IEO-CSIC)); Alfonso Pérez (Spanish Institute of Oceanography, National Spanish Research Council (IEO-CSIC)).
Session: S-2-4.
Where: G029.
When: 15:30-16:30.
Abstract: Fish stock assessments are a critical component of sustainable fishery management. Accurate estimates of stock size and mortality are essential for making informed management decisions. However, obtaining these estimates can be challenging, and even after developing a statistical model with accurate estimates of stock size, retrospective patterns and residuals autocorrelation can still arise, leading to biased estimates and inaccurate management decisions. One approach to addressing these challenges is to have alternative models or to combine the results of multiple models in a more robust ensemble approach. However, implementing ensemble modeling can be difficult due to the time and resources required to input, calibrate and analyze data for multiple models. The first obstacle for this implementation is often the data input process, which can be time-consuming and prone to errors. To deal with this difficulty, we present a web tool that optimizes the data input process for models and facilitates transitions between various input formats. It includes a user-friendly interface that guides the user through the data input process and automates several time-consuming tasks, such as data cleaning and formatting, considering that the models that allow to include several data sources have remarkable differences in the data input files format. The tool will be tested on a range of datasets and models, as a starting point with data input for very complex and highly-configurable data rich models like Stock synthesis and Gadget, which are widely used stochastic models for stock assessment around the world.
By automating the data input process and freeing up valuable time for scientists to focus on analyzing results and testing trade-offs, the tool can improve the efficiency, transparency and effectiveness of stock assessment workflows, ultimately contributing to more sustainable fishery management..
Authors: Tiago Marques (University of St Andrews / CEAUL / DBA / FCUL)+.
Session: S-4-1.
Where: G029.
When: 14:30-16:30.
Abstract: There has been a recent surge in new technologies for monitoring wildlife abundance. For sound producing taxa, passive acoustic monitoring (PAM) has been used as an alternative to visual surveys, leading to PAM density estimation (DE). As cetaceans are acoustically active and easily detectable on hydrophones, they have been at the methods development forefront, also due to them spending long times submerged, and the more conventional visual surveys being heavily dependent on suitable viewing conditions. Cetacean PAMDE often involves cue counting, where animal density is obtained scaling cue density via appropriate multipliers. One fundamental multiplier is the cue rate, which, for PAMDE, is the number of sounds produced per animal per unit time. Animal borne tags with acoustic sensors are the best way to obtain cue rates. Project ACCURATE investigates cetacean sound production rates and factors affecting them to inform PAMDE exercises. Using a sperm whale dataset (113 DTAGs, 8 sites, 15 years), we report their regular echolocation click production rates, and corresponding precision, for the available location/times. Above and beyond that, we wanted to derive an estimate for other, unobserved, places and times. This talk describes that quest, from GLMs to GLMMs, from lme4 to NIMBLE, settling on a Gamma response Bayesian Mixed effects model. I describe the process and the lessons learnt along the way. The average cue rate estimate and corresponding precision incorporating the variability associated with location and year random effects is 0.95 clicks per second (95% credible interval 0.58-1.37). Main giveaways include (1) cue rates remain a considerable hurdle for PAMDE, (2) estimating the magnitude of random effects with a Gamma response might be harder than expected, and (3) predicting for unobserved levels of random effects is not as easy as one might think at first.
This research was conducted under the ACCURATE project, funded by the US Navy Living Marine Resources program (contract no. N3943019C2176). We are deeply grateful to all the sperm whale DTAG data providers that allowed this work to be implemented (Patrick Miller, Mark Johnson, Peter Tyack, Peter Madsen, Shane Gero, Pernille Tønnesen, Mónica Silva, Claudia Oliveira, Rui Prieto, Leigh Hickmott and Michael Moore). TAM thanks partial support by CEAUL (funded by FCT - Fundação para a Ciência e a Tecnologia, Portugal DOI: 10.54499/UIDB/00006/2020). .
Authors: Nicole Barbour+, Allicia Kelly, Eliezer Gurarie.
Session: S-8-1.
Where: G043.
When: 11:00-12:45.
Abstract: Barriers to animal movement are often artificial, with those with the greatest detrimental impact often being linear features, such as roads. Roads can cause habitat or movement path fragmentation, avoidance behavior, and collision-caused mortalities. Quantifying the permeability of a given road is essential for effective land-use and conservation planning. Previous quantitative methods have utilized complex, taxa-specific models that may not be designed to handle large datasets of highly correlated animal movements. We present an R package, “permeability”, that uses maximum likelihood estimation with animal biotelemetry data to quantify the permeability of a linear barrier of interest. Package functions take an input set of movement tracks with similar sampling rates, along with a given linear barrier. Maximum likelihood estimation is then used to determine an overall permeability score, together with confidence intervals. Values close to 0 represent a road that is impermeable and values close to 1 represent a highly permeable road to animal movement. We demonstrate this method for a large GPS-tracking dataset of boreal woodland caribou (Rangifer tarandus caribou), which move through areas bisected by two main highways (Highway 1 and Highway 3) in the Northwest Territories, Canada. We expected both highways to have low permeability scores, due to the well-known general avoidance of roads exhibited by R. tarandus. We further expected that Highway 3 would have a lower permeability score than the western portion of Highway 1, which experiences considerably lower vehicle traffic, and a similar score to the eastern portion of Highway 1, which has similar vehicle traffic to Highway 3. Our results supported our hypotheses, providing concrete, quantitative comparisons with precision estimates, and having immediate implications for management of this threatened caribou subspecies. We provide this R package as a broadly applicable tool for quantifying barrier permeability for a variety of migratory species..
