Assessing the Accuracy of Stochastic and Deterministic Data-Driven Models in Modeling and Predicting River Flow (Case Study: Western and Southwestern Rivers of Lake Urmia)

Simulating and forecasting river flow is crucial for understanding water inflow over different time periods, playing a significant role in water resource management. Therefore, this study evaluates and identifies the most suitable forecasting model for streamflow at seven hydrometric stations located on rivers in the west and southwest of Lake Urmia. The models employed include various stochastic ARIMA patterns as well as artificial intelligence (AI)-based models, including Extreme Learning Machine (ELM), Artificial Neural Networks (ANN), and Support Vector Machine (SVM). The results indicate that ARIMA (1,0,0) and ARIMA (2,0,1) models demonstrated superior performance in modeling streamflow in the studied rivers, with average Nash-Sutcliffe Efficiency (NSE) values of 0.67 and 0.68, respectively. In contrast, the average NSE values for the ANN, SVM, and ELM models were 0.38, 0.47, and 0.21, respectively. Additionally, the Pearson correlation coefficient for the stochastic models (0.83) was significantly higher than that of the best-performing data-driven models (0.61). Thus, stochastic models outperform artificial intelligence models in streamflow modeling and future river flow prediction within the study area, making them the preferred choice for hydrological applications.