Accurate prediction of monthly precipitation is crucial for sustainable water resources management in climate-sensitive regions such as the Urmia Lake Basin, northwestern Iran. This study proposes an optimized hybrid forecasting framework integrating Variational Mode Decomposition (VMD), Least Squares Support Vector Machine (LSSVM), Bidirectional Long Short-Term Memory (BiLSTM) networks, and the Harris Hawks Optimization (HHO) algorithm. The normalized monthly precipitation series (1980–2024) was decomposed into five Intrinsic Mode Functions (IMFs) and a residual component using VMD. LSSVM was applied to model the high- and medium-frequency IMFs, while BiLSTM captured the long-term residual trend. The HHO algorithm was employed to optimize model hyperparameters, minimizing reconstruction and prediction errors. Model evaluation revealed outstanding predictive performance, with Nash–Sutcliffe Efficiency (NSE) exceeding 0.97 and Root Mean Square Error (RMSE) below 1.6 mm across both training and testing phases, and strong correlation between observed and simulated rainfall. Bootstrap-based uncertainty analysis confirmed the model’s stability and reliability. The optimized hybrid model was further applied to project monthly precipitation for 2030–2050, indicating an approximate 7 % increase in mean annual rainfall relative to the historical average (1980–2024). These results demonstrate that the VMD–HHO–LSSVM–BiLSTM hybrid framework effectively captures nonlinear, nonstationary rainfall dynamics across multiple temporal scales and provides a robust tool for hydrological forecasting and adaptive water management in the Urmia Lake Basin.
