همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Comparative Assessment of Process-Based and Deep Learning Models for Runoff Simulation: A Case Study of the Zayandehrood River Basin

نویسندگان :

Mohamad Saeed Zarkhan¹ Azadeh Ahamdi²

1- Shahid Beheshti University 2- Shahid Beheshti University

کلمات کلیدی :

Runoff Simulation،SWAT،LSTM،CNN،Upstream Zayandeh-Rood Basin

چکیده :

Effective management of water resources requires accurate hydrological modeling. This study aims to compare the runoff simulation performance of the Soil and Water Assessment Tool (SWAT) and two Deep Learning (DL) models, Long-Short Term Memory (LSTM) and Convolutional Neural Network (CNN), in the upstream Zayandehrood River Basin (ZRB). The outputs of these three models were evaluated at the Eskandari and Ghalehshahrokh hydrometric stations using five statistical criteria: Nash-Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Correlation Coefficient (CC), and Percent Bias (PBIAS). Results indicated that DL models demonstrated higher accuracy (NSE: 0.79-0.91) compared to SWAT (NSE: 0.71-0.78) in the training period. Nevertheless, the performance of DL models degraded sharply in the testing period (NSE: 0.49-057). The CNN model specifically exhibited a Pbias of -21.23 that reflected underestimation. In contrast, SWAT illustrates greater stability in the testing phase, maintaining an NSE of 0.71, which suggests more reliable generalizability for runoff simulation.