Climate modeling for South Asia: statistical and deep learning for rainfall and temperature prediction

We benchmark four approaches for hydrometeorological forecasting—SARIMA, TDNN, LSTM, and XGBoost—within one reproducible pipeline for monthly rainfall and temperature across seven South Asian countries using a centennial dataset from 1901 to 2023 with 1,476 months per series. The novelty lies in the joint, cross-country comparison of classical and deep learning models on century-scale data under unified preprocessing, early stopping, and cross-validated evaluation with RMSE, MAPE, R², and CV-RMSE. Skill varies by variable and region. Rainfall often favors TDNN or XGBoost due to nonlinear dynamics. Minimum temperature favors LSTM, reflecting long memory. Stable seasonal series suit SARIMA. We convert these patterns into a hybrid selection guide that maps variable and location to the best model. The framework produces 2024–2025 forecasts with uncertainty bands to inform crop calendars, irrigation scheduling, flood readiness, heat warnings, and water allocation. The dataset scale and diagnostics support replication and policy use.