INLA AND BEYOND: BAYESIAN HIERARCHICAL MODELLING WITH INLABRU* (1-day)

Jafet Belmont Osuna*, University of Glasgow

INLA, a powerful tool for efficiently fitting complex spatial models, comes with a steep learning curve. This workshop introduces inlabru – a user-friendly interface to INLA. Learn to specify models intuitively, handle spatial data efficiently, and focus on data structures and interpretation relevant to ecology, rather than computational complexity!

CO-OCCURRENCE MODELLING IN PRACTICE – UNDERSTANDING FLEXIBLE MODELS FOR SPECIES CO-OCCURRENCE IN FREQUENTIST AND BAYESIAN FRAMEWORKS* (1-day)

Albert Bonet Bigata, University of Aberdeen
Amber Cowans, University of St Andrews
Chris Sutherland, University of St Andrews

Co-occurrence models are a powerful and increasingly popular approach for assessing how interacting species respond to environmental features and management actions. However, these models can be statistically demanding and require careful consideration of sample size requirements and model assumptions. These challenges are especially relevant for early-career researchers developing their statistical toolkits.

In this workshop, participants will gain both conceptual understanding and practical experience with co-occurrence models. We will begin with an accessible introduction to the theoretical foundations of occupancy models, focusing on how they account for imperfect detection in species distribution estimation. We will then extend these concepts to multi-species co-occurrence models, highlighting how they can be used to explore ecological interactions.

Participants will engage in hands-on exercises to simulate, fit, and evaluate co-occurrence models using R, working within both frequentist and Bayesian frameworks. Practical guidance will be provided on how to interpret model output, draw ecological inferences and evaluate the feasibility and reliability of applying these models in their own research.
Building on these foundations, the second part of the workshop will introduce novel non-linear extensions of co-occurrence models that incorporate semi-parametric additive regression approaches. These models will allow researchers to capture more realistic non-linear species interactions, moving beyond the limitations of standard linear assumptions. Through real-world case studies and interactive coding sessions, participants will learn how to simulate and fit non-linear co-occurrence models and critically evaluate their performance.

By the end of the workshop, participants will be comfortable fitting and interpreting both classical and additive co-occurrence modelling approaches. In addition, participants will have the foundational material to apply them to their own ecological research questions.

BAYESIAN DISEASE ECOLOGY: MECHANISTIC MODELS FOR EPIDEMICS, INVASIVE SPECIES AND BEYOND (1-day)

Dr. Rob Deardon, University of Calgary
Caitlin Ward, University of Minnesota

Understanding and modelling transmission processes are central to wildlife ecology, where infectious diseases play a critical role in shaping population dynamics and community structure. In ecological systems, the challenges of modelling disease spread are compounded by partial observations, latent processes (e.g., unobserved infection or recovery times), and the strong spatial and behavioural heterogeneity inherent to wildlife populations. Individuals rarely mix homogeneously, with contact structures are influenced by social grouping, habitat use, and landscape connectivity, often necessitating models that incorporate spatial or network-based transmission dynamics.

This workshop will introduce participants to the formulation and Bayesian inference of such transmission models in an ecological context. Beginning with the classic SIR framework, we will extend to individual-level models that explicitly account for spatial, environmental, and network-driven heterogeneities typical of wildlife disease systems. Emphasis will be placed on data augmentation for latent variables and on implementing Bayesian Markov chain Monte Carlo (MCMC) inference. Examples will include both real disease data (e.g., avian influenza, leptospirosis, chronic wasting disease) and simulated outbreaks. Practical sessions will demonstrate model fitting and simulation using the R packages deSolve, NIMBLE and EpiILM.

RESIDUAL DIAGNOSTICS OF GENERALIZED LINEAR (MIXED-EFFECTS) MODELS WITH DHARMA* (1-day)

Melina de Souza Leite, University of Regensburg
Florian Hartig, University of Regensburg

Generalized linear (mixed) models (GLMs and GLMMs) are some of the most widely used methods in ecological data analysis. While the benefits of these modeling approaches have been abundantly discussed, one disadvantage is the lack of general and consistent residual diagnostics, which raises concern because model misspecification may easily be overlooked. 
In this workshop, we will use the powerful R package DHARMa, together with the most commonly used packages for GLMs/GLMMs, to enhance our understanding of model diagnostics for GLMMs.

We will focus on (1) discussing the importance of diagnosing GLMMs and the consequences of interpreting poorly fitted models, and (2) detecting the main problems of assumptions violation in GLMMs through residual analysis (tests and graphical inspection), such as: non-independence (grouping variables, temporal, spatial, or phylogenetic autocorrelation), heteroscedasticity, overdispersion, and zero-inflation. Ultimately, we discuss possible modeling solutions to the most common problems.

BEST PRACTICES FOR SPECIES DISTRIBUTION MODELING USING PARTICIPATORY SCIENCE DATA* (1-day)

Matt Strimas-Mackey, Cornell University
Anna Lello-Smith, Wildlife Conservation Society, Mesoamerica & Western Caribbean
Andrew Stillman, Cornell University

Robust inference using data collected through participatory science programs requires careful processing and specialized analytical techniques to address the spatial, temporal, and taxonomic biases frequently present in these datasets. The participatory science project eBird provides an unparalleled database of over 2 billion bird observations collected by volunteer observers. The broad spatial, temporal, and taxonomic coverage of his dataset make it will suited to a wide range of applications in conservation, management, and ecological research. In this workshop, we will introduce attendees to the structure of the eBird dataset, provide an overview of the challenges associated with working with participatory science data, and highlight a range of existing applications of the eBird dataset. Through hands-ons programming exercises in R, attendees will learn a suite of best practices for accessing eBird data, pre-processing these data to facilitate robust analysis, and using the eBird dataset to estimate species distributions, including encounter rate and relative abundance. Attendees should have some experience with R, including data manipulation and plotting using Tidyverse R packages, and working with spatial data using the terra and sf packages.

JOINT SPECIES DISTRIBUTION MODELLING FOR DNA METABARCODING DATA WITH HMSC* (1-day)

Otso Ovaskainen, University of Helsinki
Nerea Abrego, University of Jyväskylä
Jennifer Kampe, University of Jyväskylä
Brendan Furneaux, University of Jyväskylä
Sten Anslan, University of Jyväskylä

In this workshop, participants will 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 or abundances as a function of environmental, spatial and temporal predictors, as well as species traits and taxonomies or phylogenies. Compared to previous HMSC courses, this course places especial emphasis on analyzing high-dimensional DNA metabarcoding data with taxonomically/phylogenetically informed models. The workshop begins with brief lectures introducing the conceptual and statistical background of HMSC. The main focus is on hands-on computer demonstrations showing how to apply Hmsc-R to high-dimensional metabarcoding data.

DEEP SPECIES DISTRIBUTION MODELING AND OTHER REGRESSION MODELS WITH THE CITO R PACKAGE (1-day)

Maximilian Pichler, University of Regensburg
Florian Hartig, University of Regensburg

Species distribution models (SDMs) relate the occurrence or abundance of species to a set of (environmental) predictors. The technical backbone of an SDM is a regression that relates a response to a number of predictors. As such, they share many commonalities with related ecological analyses methods, such as resource and step selection models in wildlife ecology, or demographic models relating growth, fecundity or mortality to environmental predictors.

Traditionally, the main modelling approach for these models was the use of GLMMs or GAMs and more recently also machine learning algorithms such as random forest or boosted regression trees. However, all these algorithms require that predictors to be scalar (i.e. a numeric value), which means that complex data such as satellite pictures, time series etc. need first be reduced to summary statistics (e.g. bioclim variables) before they can be used by the model. These algorithms also offer limited flexibility for modeling different multiple responses at the same time.

This situation changes with the emergence of deep learning models, which can process complex data such as pictures or time series directly. Models such as convolutional neural networks (CNNs) essentially create their own summary statistics of images during training. One would therefore hope that they can extract more information from complex data than traditional approaches. Deep learning models can also model multiple responses jointly, thereby potentially increasing the predictive performance of each other by sharing information.

In this workshop, we will give an introduction to “deep regression modeling” in R, using the cito package and various ecological examples from an SDM and movement ecology context. Participants will learn to run simple deep regression models for typical ecological tasks in R, how to tune and evaluate these models, and how to interpret outputs as well as model behavior using explainable AI (xAI) methods. 

INTEGRATED MOVEMENT MODELS FOR TELEMETRY AND SPECIES DISTRIBUTION DATA* (Half-day)

Frances Buderman, Pennsylvania State University
Ephraim Hanks, Pennsylvania State University
Dave Miller, Pennsylvania State University
Viviana Ruiz-Gutierrez, Cornell Lab of Ornithology

The quantity, quality, and variety of movement data has increased, but methods for using movement data together with population-level data to make population- and species-level inference are still underdeveloped. We have developed a formal integrated movement model (IMM) framework for combining individual-level movement and population-level distribution data. This framework allows estimation of movement behavior and variation in that behavior across the range of the species that is informed by both individual-level movement data and species level distribution data. From this model, researchers can quantify population-level metrics of migratory connectivity, such as the spatially explicit probability of individual being present at a specific location at any stage of its annual life cycle given prior knowledge of the overall region of any life cycle stage (e.g., overwintering grounds). 

In this 1/2-day workshop, we will provide an overview of the IMM framework and present an R package that processes telemetry and species distribution data, fits multiple IMMs appropriate for different analyses, and calculates quantities of management interest from the resulting fitted IMM. The examples in this workshop will be focused on using eBird and telemetry data from migrating individuals, but the general approach could be applied to combining telemetry data with other distribution-level data. Proficiency with R and some experience with animal movement modeling is a pre-requisite.

RELIABLY INFERRING BIODIVERSITY METRICS FROM CONTINUOUSLY COLLECTED MONITORING DATA CLASSIFIED WITH AI (1-day)

Aimée Freiberg, University of Fribourg
Daniel Wegmann, University of Fribourg

Although collecting conservation monitoring data has been largely automated through devices such as camera traps, acoustic recorders, etc, their analysis is often a major bottle neck and causes massive delays in informing conservation practitioners. This workshop will demonstrate how to build a seamless workflow that goes from continuously collected conservation monitoring data to reliable species occupancy estimates. We will show the steps to prepare the data, train an AI species classifier and discuss how to deal with the unique challenges of monitoring data so that participants will be able to set up their own workflow. We will then discuss how to quantify classification errors and how to account for them in downstream analyses. We will focus on our tool NoisyCamTrap (https://bitbucket.org/wegmannlab/noisycamtrap), a python package that jointly learns occupancy and parameters regarding classification errors so that they can be effectively integrated out. We will finally discuss how to learn and account for classification errors more broadly and how one could combine the idea with other ecological models (e.g. for activity patterns, species abundance, etc.). The workshop will have a strong practical focus in which participants will learn how to run all discussed tools themselves and how to interpret and visualize their output.

AN INTRODUCTION TO CLOSE-KIN MARK-RECAPTURE FOR FISHERIES APPLICATIONS (Half-day)

Joanna Mills Flemming, Dalhousie University

This half-day workshop provides a practical introduction to Close-Kin Mark-Recapture (CKMR), an emerging genetic approach for estimating abundance, survival, and demographic rates in fisheries populations. CKMR combines genetic relatedness information (e.g., parent–offspring and sibling matches) with statistical models, enabling population assessment in cases where direct tagging is difficult or impossible.

Participants will gain an understanding of:
• The core concepts behind CKMR and how it differs from classical mark-recapture.
• Types of genetic data and kinship assignments used in CKMR.
• How CKMR models translate kinship counts into abundance and demographic estimates.
• Key considerations in study design, sampling, and interpretation of uncertainty.

The workshop blends conceptual explanation with worked examples and discussion of real-world case studies. No prior experience with genetic methods is required, though familiarity with basic statistical or population modelling concepts will be helpful.

BAYESIAN HIDDEN MARKOV MODELS FOR ANIMAL MOVEMENT TIME SERIES* (1-day)

Vianey Leos Barajas, University of Toronto
Marco Gallegos Herrada, University of Toronto
Arturo Esquivel, University of Toronto

Hidden Markov models (HMMs) are a popular class of time series models commonly applied to animal movement datasets. HMMs assume that the movements we observe are a product of an unobserved behavioural process. As such, they can be used to identify behavioural states and also allow for incorporation of variables that may affect the behavioural process and movement dynamics. We can further extend HMMs to account for individual variability via random effects and multivariate processes. However, fitting HMMs to data can be challenging as the likelihood often has multiple local maxima, which can make frequentist inference challenging, and can lead to multimodal posterior distributions, making standard Bayesian inference non-trivial. 

In this workshop, we will show how to fit HMMs to animal movement time series in a fully Bayesian framework using the probabilistic programming language Stan as well as the Julia package Pigeons. We will demonstrate that using a Bayesian approach allows for two major advantages: (i) setting appropriate prior distributions for the parameters of the HMM stabilizes inference and (ii) we can directly quantify uncertainty over a multimodal posterior via parallel tempering. 

PARTIAL OBSERVABILITY AND MANAGEMENT OF ECOLOGICAL SYSTEMS (Half-day)

Cassie Speakman, Centre for the Synthesis and Analysis of Biodiversity
Olivier Gimenez, Centre national de la recherche scientifique

In conservation and applied ecology, effective decision-making for managing dynamic systems ideally relies on a complete understanding of the system’s state. However, the true state of a system is often difficult to determine or monitor. Dynamic optimization methods, such as Markov decision processes (MDPs), are particularly well-suited for identifying the optimal sequence of decisions to achieve ecological management objectives. Yet, when the state of the system is unknown or only partially known, partially observable Markov decision processes (POMDPs) extend MDPs by incorporating uncertainty about the system’s state. “In the language of statistics, POMDPs are models that describe optimal control of hidden Markov models (HMMs). In this framing, the partially observed states of the model are latent variables that are inferred from the (fully observable) observation states” (Chadès et al., 2021). These models can be applied to certain types of management problems. The first is when managers need to decide between using current knowledge of a system or reducing state uncertainty, usually through monitoring. The second is when managers need to manage systems under imperfect detection. And lastly, for adaptive management problems where the state is unobservable.

POMDPs provide a powerful tool for solving sequential decision-making problems when the decision-maker has incomplete knowledge about the current state of the system. This approach has been successfully applied in a range of ecological contexts, including conservation, invasive species management, and resource management. In this workshop, we will introduce participants to MDPs and explore when POMDPs become necessary. We will discuss the theoretical challenges of solving POMDPs and demonstrate practical solutions through case studies in ecology. Our goal is to provide a valuable entry point for understanding and applying MDPs and POMDPs for ecological decision-making.

NIMBLE FOR ECOLOGY USING FLEXIBLE MCMC SPECIFICATION AND QUADRATURE-BASED APPROXIMATIONS (1-day)

Christopher Paciorek, UC Berkeley 
Perry de Valpine, UC Berkeley 
Daniel Turek, Lafayette College 
Paul van dam-Bates, Canada Fisheries and Oceans

NIMBLE (https://r-nimble.org) is a system in R for writing statistical algorithms, especially computationally intensive methods, for general model structures. NIMBLE adopts the modeling language used by BUGS and JAGS but opens it for extensions with user-defined distributions and functions written from R. NIMBLE provides an algorithm library that includes Markov chain Monte Carlo (MCMC) samplers, sequential Monte Carlo (particle filtering), and quadrature-based approximations. Both models and algorithms are automatically compiled via custom-generated C++. 

In this proposed workshop, we plan to cover the following components:

1. An introduction to NIMBLE, which will prepare everyone, especially new users, 
for the remainder of the workshop.
2. Introduction to the nimbleEcology package, where many common ecological
models (such as capture-recapture, occupancy, N-mixture) are provided that can be
used directly in NIMBLE model code.
3. Use of NIMBLE’s MCMC system (including HMC) in the context of ecological case studies, emphasizing the flexibility to choose samplers and add user-defined sampling strategies. 
4. Use of Laplace approximation and NIMBLE’s new INLA-like nested approximation for fitting models efficiently without MCMC, but using the same model specification. 
5. An overview of recent changes to the NIMBLE compiler and model structure/processing and how these increase ease of use and efficiency. 

METHODS FOR ESTIMATING DENSITY OF UNMARKED INDIVIDUALS FROM CAMERA TRAP DATA (Half-day)

Matthew Gonnerman, University of Maryland

An understanding of animal densities is a crucial component of proper wildlife management, but it is often difficult to obtain accurate estimates of density without expending a considerable amount of effort, time, and/or money. Following efforts to develop cost saving methods for monitoring animal populations, camera traps have become a commonly used tools, allowing persistent monitoring at an array of locations simultaneously. Camera trap data can be analyzed in a number of ways, often requiring differentiating between individual animals to estimate density. However, for many species it is impossible to consistently differentiate individuals by their external traits, and marking efforts may be beyond what available project resources will allow. In such cases, a set of modeling tools have been developed to estimate density from photos of unmarked individuals. 

These methods differ in multiple ways, including how photos are prepared and processed, how cameras are distributed on the landscape, and how estimates are interpreted. For example, N-mixture models use the count of animals within the picture window to estimate relative density, but no true estimates of abundance can be produced. In comparison, random encounter models (REM), random encounter and staying time models (REST), and distance sampling based on camera traps (CT-DS) all produce true abundance estimates but require additional information on where the animal is in relation to the camera or for how long it remains within the detection area of the camera, which expends considerably more effort processing photos. Our goal is for all participants to have a grasp on the unique advantages and limitations of each method so that they can make informed decisions when deciding on methodology.

We will divide this workshop into five parts, beginning with a general introduction to camera trap setup, sampling design, photo processing, and preparing code for running models in JAGS. The remainder of the workshop will be split between the four estimation methods. For each, we will provide a review of the current literature, highlighting practical concepts such as model assumptions, requirements of the sampling design, and limitations to inference. We will then walkthrough the general steps of the sampling process, from camera setup to photo processing, using a real dataset collected from mammal communities in urban parks. Finally we will walk through translating model equations into JAGS code and review how to troubleshoot and interpret model outputs for further applications. 

From this workshop, we will provide participants with a practical understanding of the available options for estimating animal density from photos of unmarked animals. We will summarize the workshop into a written step-by-step guide, describing how to determine study design and camera placement, how to prepare data and run models in R using JAGS, and how to interpret model outputs. We will be providing participants with all code and materials used within the workshop to enable their future work.