Document Type : Original Article
Authors
1
ASD
2
Department of Civil Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran
3
Associate Professor, Department of Range and Watershed Management, Urmia University, Urmia, Iran
4
Professor, Department of Water Engineering, Center of Excellence in Hydroinformatics, Faculty of Civil Engineering, University of Tabriz, Tabriz, Iran Farazab Co. (Consulting Engineers), Research and Writing Capacity Enhancement Affairs, Tabriz, Iran.
Abstract
With the increasing complexity and dynamics of hydrological systems, accurate and reliable river stream flow prediction is necessary for sustainable water resource management. This research utilized 20 years (from 2001 to 2021) of daily precipitation, river discharge, and mean air temperature data from the Dehgolan basin in Kurdistan Province. To select the optimal combination and model scenarios, Pearson's correlation coefficient was employed using precipitation (Pt), mean temperature (Tt), and river discharge with one to three days of lag (Qt-1 to Qt-3). The Pearson correlation coefficient (PCC) was used to select optimal scenarios and model combinations, establish the relationship between input and output variables, and subsequently choose the model and scenario combinations. For streamflow prediction, we utilized hybrid models including the Artificial Neural Network-Tasmanian Devil Optimizer (ANN-TDO), Support Vector Regression-Red Tailed Hawk (SVR-RTH), and the deep learning model Long Short-Term Memory-Marine Predators Algorithm (LSTM-MPA). Model evaluation involved the following metrics: Root Mean Square Error (RMSE), Coefficient of Determination (R2), and Kling-Gupta Efficiency (KGE). Among the hybrid models, ANN TDO consistently demonstrated the best performance. Its R2 values for the testing phase were frequently above 0. 8, and KGE values reached up to 0. 915 in some scenarios, indicating a very high correlation with observed data. The LSTM-MPA model also delivered very good performance. Although it performed slightly below ANN-TDO in some scenarios, its R2 and KGE values (often above 0. 7 and 0. 8), along with low MAE and RMSE values, demonstrated this model's high capability in time series modeling.
Keywords
Main Subjects
)