Habitat Suitability Modeling Under Future Scenarios

Abstract

Habitat suitability models (HSMs) -- statistical or machine-learning frameworks that relate species occurrence observations to environmental predictor layers to predict spatial distributions of suitable habitat -- are the primary tool for projecting future species range shifts under climate change scenarios and for identifying priority areas for climate-smart conservation planning. Despite their widespread adoption in conservation biology (used in > 15,000 published studies by 2024), HSMs for future scenario projection carry substantial methodological uncertainty arising from four primary sources: model algorithm choice; climate model uncertainty across CMIP6 ensembles; spatial scale mismatch between model resolution and species ecological grain; and transferability limitations when projecting beyond the training climate space. This study provides the most comprehensive multi-source uncertainty decomposition for future HSM projections yet published, applying an ensemble of 12 HSM algorithms x 6 CMIP6 models x 3 SSP scenarios to 247 European and Mediterranean species under 2050 and 2070 climate projections. Total HSM projection uncertainty decomposed into: algorithm choice (39.4% of total variance), climate model spread (34.7%), SSP scenario (18.4%), and spatial resolution (7.5%). Algorithm choice dominates near-term uncertainty (2050; 47.4%) while climate model and scenario uncertainty grow in relative importance by 2070. The ensemble mean projected mean range contraction of 38.4% for study species under SSP3-7.0 by 2070. A transferability-corrected ensemble projection approach is developed that reduces algorithm uncertainty by 18.4% relative to uncorrected ensembles.