Streamflow Prediction Under Climate Change Scenarios (CMIP6) Using Improved Machine Learning Technics

Climate change is one of the most significant factors affecting global water resources and has become a primary global challenge in recent decades. In this study, the MRI-ESM2-0, CMCC-ESM2, and MPI-ESM1-2-LR models were selected from 11 CMIP6 GCMs based on their performance in simulating daily minimum/maximum temperatures and precipitation, achieving average Kling-Gupta Efficiency (KGE) values of -0.058, 0.826, and 0.564, respectively. Subsequently, the LARS-WG model was used to downscale and generate future climate data for three periods (2031-2050, 2051-2070, and 2071-2090) under three Shared Socioeconomic Pathways (SSPs): SSP1-2.6, SSP2-4.5, and SSP5-8.5. Relative to the baseline period (1997-2016), projected monthly changes are expected to range from 0.69 to 5.16 °C for minimum temperature, 0.86 to 4.12 °C for maximum temperature, and -102.5 to +225.9 mm for precipitation. This study employs several machine learning models—Support Vector Regression (SVR), Long Short-Term Memory (LSTM), and Random Forest (RF)—enhanced with the Dragonfly Algorithm (FDA) and Particle Swarm Optimization (PSO) to predict future streamflow. Seven input feature combinations were assessed using K-fold cross-validation to identify the optimal feature subset, with the first combination consistently demonstrating the best performance across all models. Furthermore, the optimized RF-FDA model outperformed all other models, achieving R, RMSE, and MAE values of 0.96, 6.11, and 2.29 during training, and 0.86, 9.55, and 8.34 during testing, respectively. The projected monthly discharge variation at the Koshtargah hydrometric station ranges from -24.2 to +22.4 m 3 /s relative to the baseline period. This research demonstrates that integrating optimized machine learning models with CMIP6 GCMs provides a powerful and reliable tool for assessing climate change impacts on a watershed’s hydrological system, particularly for applications in river engineering, water resource management, and flood management.