Spatial habitat suitability modelling of Astraeus Hygrometricus using explainable machine learning and GeoAI approaches

Astraeus hygrometricus, a wild mushroom, is an important nutrition-rich edible non-timber forest resource, threatened by environmental and climatic change. This study developed an explainable GeoAI framework to model its habitat suitability in the lateritic forests of eastern India. Using 514 field observations and 32 geo-environmental predictors, five machine-learning models were optimised through Bayesian optimisation and evaluated using spatial cross-validation. Explainability and feature selection were assessed using SHAP, Boruta, and nature-inspired algorithms, while CMIP6 scenarios were used to evaluate future suitability. Model performance was consistently high (AUC = 0.979–0.993), with XGB and GBM achieving the highest AUC (0.993). Very-high suitability declined from 26.47% under baseline conditions to 23.03% under SSP245 and 21.10% under SSP585. LULC, radar backscatter, soil temperature, and elevation were important predictors. The framework provides a robust approach for identifying climate-sensitive habitats and supporting conservation and sustainable management of this valuable forest resource.