R for geospatial predictive mapping

R for geospatial predictive mapping: practical workflows for reliable spatial predictions

This talk introduces practical workflows for building reliable spatial predictions in R, illustrated with a real-world case study on predicting plant species richness across South America. We’ll start with classic interpolation methods like IDW and kriging, then move to machine learning approaches using Random Forests. Along the way, we will discuss common pitfalls, such as unrealistic predictions and spatial bias, and explore practical solutions like the Area of Applicability (AoA) and prediction-domain adaptive cross-validation methods (kNNDM). We will also look at R’s spatial machine learning ecosystem, including CAST, caret, mlr3, and tidymodels, along with key steps in feature engineering, hyperparameter tuning, and uncertainty estimation. All examples include reproducible R code, making it easy to adapt the workflows to different spatial prediction tasks.