Verification data and evaluation scripts for BPDP V2.0C 3-hourly QPF over Brazil (April–May 2026)

This dataset supports the manuscript submitted to the Journal of Hydrology on a GEFS-ensemble-driven dual-gate machine-learning system for 3-hourly quantitative precipitation forecasting over Brazil. Contents:- Verification scorecards (JSON) for V2.0C and comparison configurations (raw GFS, GEFS ensemble mean, V2.0B-II, single-gate ablation, no-GEFS ablation) against IMERG Late Run, April–May 2026.- INMET dual-truth cross-verification scorecards (April 2026).- Daily aggregated statistics and 95% day-block bootstrap confidence intervals.- Per-lead-time, spatial, reliability/ROC, and case-study tabular data underlying manuscript figures.- Diurnal-cycle and regional-regime stratification CSVs (v2 addition).- Python scripts to reproduce scorecards, bootstrap CIs, calibration demo, diurnal/regional analysis, and paper figures (including fig9). Raw NWP inputs (GFS/GEFS) and IMERG/INMET observations are openly available from NOAA/NCEP, NASA GES DISC, and INMET BDMEP; this deposit contains only derived verification products and evaluation code. Version note (v2): adds diurnal_2026.csv, regional_2026.csv, diurnal_regional_2026.py, and an updated make_paper_figures.py for the diurnal/regional analysis added to the manuscript.