A novel hybrid machine learning approach for suspended sediment load forecasting: A case study of Mazandaran rivers basins

Estimating the suspended sediment load (SSL) in watersheds is a fundamental challenge in surface water resource management and hydraulic structure design. This study utilized data from the Haraz, Babol Rud, Talar, and Neka Rud watersheds in Mazandaran province, Iran, encompassing discharge, daily sediment load, and key physical characteristics. These data were organized in three different combinations as inputs to several machine learning models. Among these, the input scenario M ₃ , which combines streamflow data, sediment load, and the physical characteristics of the basins, demonstrated the best performance across all models. The machine learning models employed were Support Vector Regression (SVR), Long-Short Term Memory (LSTM), Random Forest (RF) and Extreme Gradient Boosting (XGBoost) the performance of which was enhanced using two optimization algorithms: Particle Swarm Optimization (PSO), and Flow Direction Algorithm (FDA). While both optimization methods produced acceptable results, the FDA algorithm demonstrated superior performance. The RF-FDA hybrid model achieved the highest accuracy, yielding the following statistical metrics during the training phase: RMSE = 2.45, MAE = 1.98, R = 0.95, KGE = 0.87, and NSE = 0.90. The corresponding metrics in the testing phase were 3.26, 2.44, 0.92, 0.88, and 0.84, respectively. The findings of this study have significant implications for engineers and policymakers in the design of hydraulic structures and water resources management.