Himalayan Fire — RF Hurdle Model (MAM) Reproducibility Package

Code package to reproduce the two-stage Random Forest hurdle modelling workflow for Himalayan forest-fire activity in the pre-monsoon season (MAM; March–April–May). The workflow fits (i) a fire-occurrence model ϕ(X) using RandomForestClassifier and spatially blocked GroupKFold cross-validation (block_id_spatial), and (ii) a conditional intensity model μ(X) using RandomForestRegressor trained on positive fire months (fire_cnt > 0). Out-of-fold predictions, cross-validation metrics, final trained pipelines, and run-configuration snapshots are produced. The script defaults to MAM (manuscript scope); optional subsets (ALL/DJF/JJA/SON) are available for exploration.