machine learning and deep learning approaches for river flow forecasting

Crucial. this study focuses on forecasting one month-ahead river flow in the zarrineh rood river, a region in northwest iran facing water scarcity. four machine learning models, random forest (rf), m5 model tree (m5), support vector regression (svr), and multilayer perceptron (mlp), were evaluated. a 12-month lag length, informed by the dataset's monthly time frame and the river's 12-month cycle, consistently provided the best results, and no maximum lag length constraint was imposed. the random forest (rf) model excelled in the training dataset, delivering highly accurate predictions with significant correlation coefficients, minimal root mean squared error (rmse), and negligible mean absolute error (mae). conversely, the support vector machine (svm) model outperformed in the test dataset, demonstrating substantial correlation coefficients and low rmse and mae. the study emphasizes the importance of selecting appropriate models and input variables for precise hydrological predictions, underlining the dynamic and nonlinear nature of river flow patterns. by analyzing both training and test datasets, this research contributes insights for improving hydrological models and understanding river flow dynamics