Enhancing streamflow projections under climate change using optimized machine learning: A comparative study in the Babolrood River Basin, Iran

Projecting the hydrological response of river basins to climate change is a fundamental challenge for sustainable water resource management. This study presents a comprehensive framework to assess future streamflow dynamics in the Babolrood River Basin, Iran. First, a rigorous evaluation of eleven CMIP6 global climate models identified the top performers for the region: MRI-ESM2-0 for precipitation, CMCC-ESM2 for minimum temperature, and MPI-ESM1-2-LR for maximum temperature. Their outputs were then downscaled using LARS-WG for 2031-2090 under three climate scenarios. The climate projections revealed a significant warming trend, with monthly minimum and maximum temperatures projected to rise by up to 5.16 and 4.12 C, respectively, alongside extreme fluctuations in precipitation. To translate these projections into hydrological impacts, five machine learning models including K-Nearest Neighbor (KNN), Support Vector Regression (SVR), Long Short-Term Memory (LSTM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) was developed. A key innovation was the systematic optimization of their hyperparameters using both the novel Flow Direction Algorithm (FDA) and Particle Swarm Optimization (PSO). Our comparative analysis identified RF-FDA as the most robust model, achieving outstanding performance in the test phase (R=0.86, RMSE=14.38, NSE=0.79, KGE=0.82). Projections from this optimal model revealed a fundamental shift in the watershed’s hydrology. Projected monthly discharge from optimal model alterations ranges from a decrease of 24.2 m³/s to an increase of 22.4 m³/s relative to the baseline period. This integrated methodology demonstrates high reliability and serves as a powerful tool for developing adaptive strategies in river engineering and water management.