Skip to main content
Dryad

Data and code from: Mapping biome boundaries from climate patterns: a machine learning framework

Costa, Gabriel1; Aguirre-Santoro, Julián2; Gehara, Marcelo3; Gonçalves-Souza, Thiago4; Michelangeli, Fabian5

Published Sep 29, 2026 on Dryad. https://doi.org/10.5061/dryad.wh70rxx39

Data files

Sep 29, 2026 version files 2.41 GB

Click names to download individual files

Abstract

This repository includes: (i) the categorical biome reference raster and present and future bioclimatic predictor stacks used in the analysis; (ii) balanced modeling datasets comprising 16,000 observations for each of 14 biomes and the probability Random Forest model fitted with the ranger package; (iii) derived spatial layers describing biome predictions, normalized Shannon entropy, alternative-biome support (1 − p1), uncertainty tiers, cross-model agreement, and projected biome transitions; and (iv) scripts and supporting outputs for nested spatial cross-validation, predictor selection, sampling-sensitivity analysis, individual and grouped permutation importance, accumulated local effects, present-day mapping, and mid-century climate projections. The original WorldClim version 2.1 bioclimatic data and WWF Terrestrial Ecoregions source data are publicly available from the sources documented in the repository.