Training
Skills Showcase
- A practical tutorial on residual diagnostics for hierarchical (multi-level/mixed) regression models with DHARMa (Florian Hartig, University of Regensburg)
- An introduction to multilevel and phylogenetic meta-analysis using metafor (Shinichi Nakagawa, UNSW, Sydney)
- Bayesian modelling with greta (Nick Golding, University of Melbourne)
- CaPow! Online simulation and power analysis for open-population capture-recapture (Robin Aldridge-Sutton and Rachel Fewster, University of Auckland)
- Deep learning: opening the black box (Jennifer Hoeting, Colorado State University)
- Getting started with hidden Markov models (Vianey Leos-Barajas, North Carolina State University)
- Going beyond 2D and 3D to visualise higher dimensions, for ordination, clustering and other models (Dianne Cook, Monash University)
- Hierarchical modelling of species communities - accounting for environment, traits and phylogeny in space and time (Otso Ovaskainen, University of Helsinki)
- Inferences from passive acoustics: detection, classification, localisation, tracking and density estimation (Len Thomas, University of St Andrews)
- Modelling ecological data as thinned point processes with
inlabru (Janine Illian, University of Glasgow)
Short Courses
The weekend before the conference (June 20th-21st) featured the below short courses. Each course ran over both days, in one of the two conference sessions: Session 1 (8-11am GMT), or Session 2 (9pm-12am GMT).
Sorry, all short courses full!
| Session 1 (8-11am GMT). Session is fully subscribed. | Session 2 (9pm-12am GMT). Session is fully subscribed. |
|---|---|
| Level-up your R package | A statistical view of deep learning in ecology |
| Multivariate modelling in ecology and joint species distribution models | With great power comes great responsibility: Stan for modern ecological modelling |
| Spatial modelling and visualization of species distribution and disease risk using R and INLA |
Level-up your R package
Nick Golding (University of Melbourne) and Saras Windecker (University of Melbourne) cover the essential tools and strategies for making your R package the best it can be.
Saturday 20th-Sunday 21st June, Session 1 (8-11am GMT)
Level-up your R package
It's easier than ever to write and distribute an R package to implement your new method. But it's much harder to make sure your new R package is bug-free, easy to use, and easy for you to maintain and extend. This workshop will teach you some tools and strategies for making your R package the best it can be.
We'll cover how to:
- Design and implement a simple and intuitive user interface
- Plan the internal architecture of your package code so it's easy to extend and maintain
- Create high-quality documentation, including vignettes, a website and forum
- Write good unit and integration tests to find and fix bugs
- Set up automated package testing, code style-checking, and documentation spell-checking
This will be a hands-on workshop. You'll gain experience reviewing other people's packages and working together to design new package interfaces. You're welcome to bring along your own package, work on it during the workshop, and get feedback on how to improve it.
Required background:
- Experience writing functions in R
- Interest in creating R packages
- Opinions about what you like/dislike about the R packages you use
Nick Golding and Saras Windecker
Nick and Saras are ecologists, research software engineers, and R obsessives. They have written over 20 R packages (some wonderful, some dreadful) and reviewed more than 30 more for journals like Methods in Ecology and Evolution, and Journal of Open Source Software.
Multivariate modelling in ecology and joint species distribution models
Scott Foster (CSIRO), Otso Ovaskainen (University of Helsinki), Gordana Popovic (UNSW Sydney), David Warton (UNSW Sydney) and Skip Woolley (CSIRO) introduce the latest tools developed for handling multivariate data in ecology.
Saturday 20th-Sunday 21st June, Session 1 (8-11am GMT)
Multivariate modelling in ecology and joint species distribution models
Multivariate analysis of abundance or presence/absence data in ecology is a challenging problem, for which analysis techniques have been developing rapidly in recent years. Historically these sorts of data were analysed using algorithms based on pairwise dissimilarity metrics, but a modern approach involves specifying a joint statistical model for the data, sometimes called a joint species distribution model. This approach has a number of advantages, including in statistical properties, interpretability, and functionality. This short course will give an introduction to a range of tools that have recently been developed for multivariate data in ecology, including methods for hypothesis testing, ordination, trait modelling, prediction, classification, and studying causes of co-occurrence. Packages discussed include mvabund, HMSC, gllvm, SpeciesMix and ecoCopula.
Scott Foster, Otso Ovaskainen, Gordana Popovic, David Warton and Skip Woolley
The presenters are developers of new statistical approaches to multivariate analysis in ecology, whose multivariate software is having increasing impact in ecology.
Spatial modeling and visualization of species distribution and disease risk using R and INLA
Paula Moraga (University of Bath) introduces how to develop spatial geostatistical models using the R-INLA package to predict species distribution, estimate disease risk and quantify risk factors. She also shows several R packages to create visually informative and interactive reports, dashboards, and Shiny web applications.
Saturday 20th-Sunday 21st June, Session 1 (8-11am GMT)
Spatial modeling and visualization of species distribution and disease risk using R and INLA
In this course we will learn how to develop spatial geostatistical models using the R-INLA package to predict species distribution, estimate disease risk, and quantify risk factors. We will also learn how to create data visualizations such as static and interactive maps, and introduce presentation options such as interactive dashboards and Shiny web applications that facilitate the communication of insights to collaborators and policy makers. We will work through several fully reproducible examples of ecology and disease mapping applications using real-world data such as sloths in Latin America and malaria in The Gambia. The examples will provide clear descriptions of the R code for data importing, manipulation, modeling and visualization, as well as the interpretation of the results. We will cover the following topics:
- Manipulate and transform geostatistical and raster data using spatial packages
- Query and collect species occurrence data from several sources including the Global Biodiversity Information Facility (GBIF) and the Atlas of Living Australia (ALA) using the spocc package
- Retrieve high resolution spatially referenced environmental data using the raster package
- Model species distribution, disease risk, and risk factors in different settings
- Fit and interpret spatial models using Integrated Nested Laplace Approximations (INLA) and the Stochastic Partial Differential Equation (SPDE) approaches
- Create static and interactive visualizations using leaflet and ggplot2
- Communicate results with reproducible R Markdown reports, interactive dashboards and Shiny web applications
The course materials are based on the book 'Geospatial Health Data: Modeling and Visualization with R-INLA and Shiny' by Paula Moraga (2019, Chapman & Hall/CRC Biostatistics Series).
Paula Moraga
Paula Moraga (@Paula_Moraga_) is a Lecturer in the Department of Mathematical Sciences at the University of Bath, UK. She develops innovative statistical methods and open-source software for disease surveillance including R packages for spatio-temporal modeling, detection of clusters, and travel-related spread of disease. Her work has directly informed strategic policy in reducing the burden of diseases such as malaria and cancer in several countries.
A statistical view of deep learning in ecology
Jennifer Hoeting (Colorado State University) describes deep learning from a statistical viewpoint and its applications in ecology.
Saturday 20th-Sunday 21st June, Session 2 (9pm-12am GMT)
A statistical view of deep learning in ecology
The goal of this short course is to introduce neural networks and deep learning from a statistical viewpoint. The focus will be on explaining deep learning for statistical ecologists and ecological statisticians. Many conceptual explanations and cartoon sketches of deep learning are available, but deep learning is rarely translated into the mathematical framework required by most statisticians to understand the topic. In addition to presenting deep learning from a statistical viewpoint, we will consider where deep learning is useful in ecological applications. Students will gain experience with latest interface for deep learning programs within R (no Python required)!
Jennifer Hoeting
Jennifer Hoeting is a Professor of Statistics at Colorado State University, where she has been based since 1994. Her honours include being a fellow of the American Statistical Association and a Distinguished Achievement Medal from the Section of Statistics and The Environment of the American Statistical Association. She has broad research interests, including model selection and uncertainty, spatial statistics, Bayesian statistics, and more recently, the interface between statistics and machine learning.
With great power comes great responsibility: Stan for modern ecological modelling
Daniel Simpson (University of Toronto) and Andrew MacDonald (Université de Montréal) demonstrate how to specify and infer statistical models using the RStan package to appropriately represent the data and process at hand.
Saturday 20th-Sunday 21st June, 9pm-12am GMT
With great power comes great responsibility: Stan for modern ecological modelling
Contemporary ecological models are growing more complex, capturing not only ecological processes but also other sources of variation, such as sampling noise and measurement error. At the same time, ecological data is growing not only more available, but also more highly detailed. How can we create models that capture all this complexity, while confronting the unavoidable spectre of model misspecification? It is useful to turn to specialized programming languages like Stan, which aims to be a language for specifying probabilistic models.
Stan allows users to specify and infer complex, bespoke, statistical models that are built to appropriately represent the data and process at hand. While this extra power allows scientists to get the most out of their data, we must keep in mind the mantra of Spiderman: "With great power comes great responsibility".
In this course we will cover three main topics:
- Building bespoke models for ecological data in Stan, including appropriate prior modelling and model checking
- Inferring models using the Stan language
- Post-inference model checking, model criticism, and model selection
Assumed knowledge:
- Generalized linear models
- R
- Generalized linear mixed models (or anything with random effects) is recommended
Outcomes:
- Prior construction in ecological models
- Prior and posterior model checking
- Basic understanding of the Stan language
Participants are encouraged to bring their laptops with R and the RStan package installed.
Daniel Simpson and Andrew MacDonald
Daniel Simpson is a Professor in the Department of Statistical Sciences at University of Toronto.
Andrew MacDonald is a Professor in the Department of Biological Sciences at Université de Montréal.
