Rainfall-runoff modelling is pivotal for effective water resources management, particularly in subtropical regions characterized by complex hydrological dynamics and climatic variability. This research investigates the application of six advanced machine learning (ML) methods—Long Short-Term Memory networks (LSTMs), Support Vector Machines (SVMs), Gaussian Process Regression (GPR), LASSO Regression (LR), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM)—to simulate monthly streamflow in the subtropical sub-basin of the Chinab River, Jammu province. Utilizing historical hydro-meteorological data (precipitation, temperature, evapotranspiration, and streamflow records) from 1980–2020, models were trained, validated, and tested to predict streamflow. Performance was evaluated using statistical metrics: Nash-Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R²). Results indicate that LSTM and XGBoost outperformed other models, with LSTM achieving the highest NSE (0.92) and R² (0.93) in validation, demonstrating superior capability in capturing temporal dependencies and nonlinearities. GPR and LightGBM also showed robust performance, while SVM and LASSO regression exhibited limitations in handling complex seasonal patterns. This study underscores the potential of ensemble and deep learning approaches in improving hydrological predictions in subtropical basins, offering insights for sustainable water management and flood forecasting in the Jammu region.
This research presents a comprehensive comparative evaluation of six advanced machine learning methods for monthly streamflow prediction in the subtropical Chinab River basin of Jammu & Kashmir, India. Using 40 years of hydro-meteorological data (1980–2020), LSTM and XGBoost emerged as the best-performing models, achieving Nash-Sutcliffe Efficiency values of 0.91 and 0.89 respectively, substantially outperforming traditional linear and kernel-based approaches. The study demonstrates that deep learning and ensemble methods are particularly suited to subtropical basins characterized by pronounced monsoon variability and complex nonlinear hydrological interactions. These findings support the integration of advanced ML techniques into operational water resources management systems for the Jammu region.