Pakistan monsoon rainfall for climate adaptation using generalized linear models

This is the authors' abstract. We don't add key points for this paper.

Study region Pakistan, where highly variable summer June, July, August, and September (JJAS) monsoon rainfall drives both water security and flood risks across diverse topography. Study focus To support climate adaptation and sustainable water governance, this study develops a transparent Generalized Linear Modeling (GLM) framework for concurrent statistical estimation of JJAS rainfall totals. Using CHIRPS observations (1981–2020), two nested models were evaluated by leave-one-year-out cross-validation across Pakistan and four hydro-climatic subregions. The precipitation-free atmospheric-teleconnection model (M1) uses winds, thermodynamic variables, and climate indices; the full model (M2) additionally includes concurrent ERA5 seasonal precipitation. New hydrological insights for the region At the Pakistan scale, precipitation-free M1 retained cross-validated R² = 0.43 (RMSE = 30.8 mm), whereas M2 with ERA5 precipitation achieved R² = 0.74 (RMSE = 20.9 mm). M1 also retained substantial skill in LIB (R² = 0.76) and BCP (R² = 0.55), but not in NHM or UIB. Raw ERA5–CHIRPS rainfall correlations of 0.62–0.91 demonstrate strong concurrent product agreement and indicate that the two precipitation estimates should not be treated as practically independent. Relative to an ERA5-only model, the remaining predictors produced a statistically resolved RMSE reduction only in LIB (+8.34 mm; 95% CI 1.61–14.33 mm; positive favors M2). M2 is therefore interpreted as a concurrent multi-product reconstruction, while M1 quantifies the stand-alone atmospheric and teleconnection signal. Comparison with ridge regression and random forest showed comparable ridge performance and no consistent random-forest improvement, supporting the GLM as an interpretable regional baseline.