ONLINE COURSE – Species distribution modelling with Bayesian statistics in R (SDMB01) This course will be delivered live
7 September 2020 - 11 September 2020£500.00
This course will now be delivered live by video link in light of travel restrictions due to the COVID-19 (Coronavirus) outbreak.
This is a ‘LIVE COURSE’ – the instructor will be delivering lectures and coaching attendees through the accompanying computer practical’s via video link, a good internet connection is essential.
Please feel free to email firstname.lastname@example.org with any questions, full course detials below.
Bayesian Additive Regression Trees (BART) are a powerful machine learning technique with very promising potential applications in ecology and biogeography in general, and in species distribution modelling (SDM) in particular. Unlike most other SDM methods, BART models can generally provide a well-balanced performance regarding both main aspects of predictive accuracy, namely discrimination (i.e. distinguishing presence from absence localities) and calibration (i.e., having predicted probabilities reflect the species’ gradual occurrence frequencies). BART can generate accurate predictions without overfitting to noise or to particular cases in the data. As it is a cutting-edge technique in this field, BART is not yet routinely included in SDM workflows or in ensemble modelling packages. This course will include 1) an introduction or refresher on the essentials of the R language; 2) an introduction or refresher on species distribution modelling; 3) an overview of SDM methods of different complexity, including regression-based and machine-learning (both Bayesian and non-Bayesian) methods; 4) SDM building and block cross-validation focused on different aspects of model performance, including discrimination, classification, and calibration or reliability. We will use R packages ’embarcadero’, ‘fuzzySim’ and ‘modEvA’ to see how BART can perform well when all these aspects are equally important, as well as to identify relevant predictors, map prediction uncertainty, plot partial dependence curves with credible intervals, and map relative favourability regarding combined or individual predictors. Students will apply all these techniques to their own species distribution data, or to example data that will be provided during the course.
Any researchers (PhD and MSc students, post-docs, primary investigators) and environmental professionals who are interested in implementing best practices and state-of-the-art methods for modelling species’ distributions or ecological niches, with applications to biogeography, spatial ecology, biodiversity conservation and related disciplines.
Venue – Delivered remotely
Duration – 5 days
Contact hours – Approx. 35 hours
ECT’s – Equal to 3 ECT’s
Language – English
PLEASE READ – CANCELLATION POLICY: Cancellations are accepted up to 28 days before the course start date subject to a 25% cancellation fee. Cancellations later than this may be considered, contact email@example.com. Failure to attend will result in the full cost of the course being charged. In the unfortunate event that a course is cancelled due to unforeseen circumstances a full refund of the course fees will be credited.
Each day will consist of approximately 3 hours of interactive live online sessions (in the Western European afternoon), which will include theoretical lectures, discussion, and general practical guidance; and approximately 4 hours of practicals that each participant will do on their own schedule / time zone, based on annotated self-explanatory R scripts. The instructor will be available for questions and help during Western European working hours and a bit beyond that, depending on the participants’ time zones. Data sets for the practicals will be provided, although participants are also encouraged to use their own species distribution data.
Assumed quantitative knowledge
A basic understanding of what species distribution / ecological niche models are.
Assumed computer background
Basic knowledge and experience with R is not strictly mandatory (as the basics will be provided), but it will make the practicals much less of a struggle.
Equipment and software requirements
A laptop/personal computer with any operating system (Linux, Windows, MacOS) and with recent versions of R (https://cran.r-project.org) and RStudio (https://www.rstudio.com) installed; both are freely available as open-source software. Also a working webcam if at all possible, for enhanced interactivity during the live sessions.
UNSURE ABOUT SUITABLILITY THEN PLEASE ASK firstname.lastname@example.org
Monday 7th – Classes from 14:30 to 17:30
An introduction / refresher on base R language
Species distribution modelling: basic concepts
Species distributions: data types and sources
Predictor variables: data types and sources
Defining the modelling region: extent and resolution
Tuesday 8th – Classes from 14:30 to 17:30
Overview of methods and R packages for species distribution modelling
Presence-absence vs. presence-background modelling methods
Regression and machine-learning methods: GLM, GAM, Maxent, Random Forests, Bayesian Additive Regression Trees (BART)
Wednesday 9th – Classes from 14:30 to 17:30
Model evaluation and validation: overview of performance metrics
Different facets of model performance: discrimination, classification, calibration
Splitting the study area for block-cross-validation
Comparing the performance of regression, machine-learning and Bayesian methods
Making predictions comparable across species, regions and time periods: probability and favourability
Thursday 10th – Classes from 14:30 to 17:30
Selecting relevant predictors with BART
Mapping prediction uncertainty with BART
Plotting partial dependence curves with Bayesian credible intervals
Mapping relative favourability regarding specific predictor variables
Friday 11th – Classes from 14:30 to 17:30
Final discussion and outlook