Dataset and code for "Quantifying Variety-Specific Heat Resilience in Boro Rice: A Machine Learning Feature Attribution Approach for Climate Risk Management"

Contents: Rice_dataset.xlsx contains three sheets (Hybrid_Boro, HYV_Boro, Local_Boro), each with 704 rows representing all 64 Bangladesh administrative districts across 11 Boro rice growing seasons (2014–2015 through 2024–2025). Each row records district identity, season, cultivated-area flag, yield (t/ha), and 12 seasonal climate covariates. After applying the Is_Cultivated filter and removing 7 biologically implausible outliers (yield < 0.5 or > 10.0 t/ha), the final modelling dataset contains 1,797 district-variety-year observations. Boro_Rice.py is the complete Python analysis pipeline (Python 3.12, scikit-learn 1.6, SHAP 0.51) that reproduces all figures, tables, and supplementary outputs from the raw dataset. Data Sources: Yield data: Bangladesh Bureau of Statistics (BBS) Yearbooks of Agricultural Statistics, 2014–2025 (publicly available at bbs.portal.gov.bd) Climate covariates: NASA POWER gridded reanalysis aggregated to district centroids (power.larc.nasa.gov), variables include T2M, T2M_MAX, T2M_MIN, T2M_RANGE, GWETROOT, GWETTOP, PRECTOTCORR_SUM, RH2M, ALLSKY_SFC_SW_DWN, ALLSKY_SFC_PAR_TOT, WS2M, WS2M_RANGE Key Variables (final model features): Column Description Units T2M_MAX Seasonal mean daily maximum temperature °C T2M_RANGE Seasonal mean diurnal temperature range °C GWETROOT Root-zone soil wetness fraction dimensionless (0–1) PRECTOTCORR_SUM Cumulative seasonal precipitation mm ALLSKY_SFC_SW_DWN Downward shortwave radiation flux kW·h/m²/day Yield_t_ha District-level Boro rice yield t/ha