Machine Learning-Based Classification of Convective and Stratiform Precipitation from GPM-DPR Dual-Frequency Signatures in the Amazon Basin — Supporting Data, Code and Saved Models

Abstract

This record provides supporting data, Python code, saved models and numerical results for the Amazon Basin GPM-DPR precipitation-type classification study. The release comprises a ZIP archive and a separate compressed CSV containing 360,000 observations and 227 columns. The ZIP includes 18 preserved study scripts, a portable main-analysis launcher, 20 saved model objects, test features and predictions, numerical evaluation results, SHAP summaries, bootstrap outputs, processed surface-rainfall context collocations, recorded software versions and an analysis-specific seed register. Model training uses observations from 2021–2023, model and decision-threshold selection uses 2024, and locked temporal testing uses 2025. The classification target is the operational GPM-DPR convective/stratiform precipitation-type label. Surface-rainfall products provide contextual comparisons rather than independent precipitation-type ground truth. Original third-party source archives are not redistributed. Archived local model checks are documented; a fresh clean-environment installation and full retraining were not performed during release preparation. The included numerical SHAP matrix was recomputed for verification and is distinct from the original matrix underlying the historical beeswarm plot. The supporting files are publicly available for research and reproducibility checks.