Course

Flexible species distribution models based on presences or abundances with flexsdm and adm R packages.

Organizer(s):

Understanding species distribution patterns is critical for effective tropical conservation in our rapidly changing world. Species distribution models (SDMs) combine species presence/absence with environmental variables to predict suitability maps, while abundance-based species distribution models (ASDMs) correlate spatial abundance with environmental data to project abundance maps. SDMs and ASDMs have applications in ecology, conservation, and biogeography, being used to prioritise conservation areas and evaluate climate and land-use changes. SDMs remain popular due to simple data requirements and broad applicability. ASDMs are still underdeveloped compared to traditional SDMs.

 Despite the relative ease of use of both models, their outcomes are sensitive to data inputs and methodological choices. This makes it important to develop robust workflows that can be easily inspected at each step. flexsdm is an R package (https://sjevelazco.github.io/flexsdm/) that enables users to design complete or partial SDM workflows. The adm R package (https://sjevelazco.github.io/adm/) is a novel modeling toolbox designed to streamline the creation of ASDM workflows. adm enables users to build modeling workflows and can be easily integrated with some flexsdm’s functions. Both packages are based on the philosophy of providing multiple functions that can be used as building blocks to create partial or complete modeling workflows. They can also be easily integrated with other packages.

This course will introduce participants to the flexsdm and adm key features and guide them through a complete modeling workflow by providing presence and abundance data of a terrestrial organism from the Neotropics. This session will guide participants through the full protocol of an SDM and ASDM project:

1) Pre-modeling: Prepare model input data, exploring sampling bias approaches, delimitation of training area, selecting pseudo-absences, partitioning data, and addressing predictor collinearity.

2) Modeling: Fit and evaluate models using various machine learning and statistical algorithms (e.g., Random Forest, Support Vector Machine, Deep Neural Network). Learn about key decisions on tuning model hyperparameters.

3) Post-modeling and model output exploration: Refine output maps, evaluate model extrapolation for current and future conditions, and produce exploratory plots such as univariate and bivariate partial dependence curves.

Target Audience: This course is for graduate students, postdocs, and professors at all levels. While covering advanced concepts, the workflow is accessible to those with basic R programming knowledge.