Modeling, evaluation and forecasting of suspended sediment load in Kal-e Shur River, Sabzevar Basin, in northeast of Iran

Studying sediment transport to rivers is crucial for effective river management, engineering, and environmental preservation. Neglecting this aspect can lead to significant harm to natural ecosystems. This research aims to estimate suspended sediment levels in the Kal-e Shur Sabzevar River using various machine learning algorithms, which have gained popularity in recent years due to their high accuracy and reliability. The study employs ensemble Bagging algorithms, the gradient boosting machine (GBM), genetic algorithm, Naïve Bayes algorithm, gradient boosting decision trees, and extremely randomized trees. These algorithms provide a coherent framework that can serve as a standard for evaluating and comparing models in future research. Initially, data from 354 sediment measurement stations, including flow discharge, sediment discharge, and precipitation, were collected. After validating data homogeneity using the double mass method, 70% of the data were allocated for training, and 30% for testing. The algorithms were trained with this data, and their performance was evaluated using the coefficient of determination (R²), root mean square error (RMSE), and Nash–Sutcliffe efficiency (NSE) statistics. Additionally, a partial least squares (PLS) regression model was employed to identify the most influential factors affecting suspended sediment load in the basin. The results demonstrate that the gradient boosting machine (GBM) model outperforms other algorithms, exhibiting R² values of 0.95, RMSE values of 0.019, and NSE values of 0.78. The PLS model identified geological factors and slope as primary determinants of suspended sediment load in the region. Lastly, the algorithms predicted sediment levels, with the GBM algorithm estimating a sediment concentration of 8955 mg/liter with a relative error of 8.54%, indicating strong alignment with the total sediment load in the region.