Predicting hydrological dynamics in semi-arid regions is challenging due to climate instability. This study proposes a hybrid modeling framework integrating Seasonal-Trend decomposition (STL), SARIMA, and Long Short-Term Memory (LSTM) networks to capture seasonal and nonlinear residual patterns. A modular Intelligent Information System (IIS) architecture is designed to operationalize this approach using real-time Central Asian data. Results demonstrate the hybrid model significantly outperforms standalone methods, achieving Nash-Sutcliffe Efficiency (NSE) scores of 0.85–0.96 and reducing RMSE by 18–35%. This framework provides a robust operational tool for data-driven water resource management.
Rustamjon Bakhtiyorjon ugli Nasridinov (Sun,) studied this question.