Authors
1
Department of Civil Engineering, Faculty of Engineering, Suleyman Demirel University, Isparta, Turkiye
2
Department of Civil Engineering, Faculty of Engineering, Suleyman Demirel University, Isparta, Türkiye
3
Department of Civil Engineering, Faculty of Engineering, Antalya Bilim University, Antalya, Türkiye
10.22034/ceej.2025.69880.2462
Abstract
Despite the complex hydrology of karst systems, which typically necessitates specialized modeling approaches due to rapid flow dynamics and multiple flow regimes, hydrological models may yield adequate results in basins with limited data resolution. This study evaluates the performance of three hydrological models LSTM, Mapshed and HBV in the karstic Kucuk Aksu basin in Turkey’s western Mediterranean coast. The research examined nine years (2013-2021) of daily meteorological and streamflow data to identify dominant hydrological elements and evaluate the accuracy of the model. The snow cover indices derived from satellite imagery analysis were used in the training of the LSTM model. The results indicate that LSTM model produced overall better performance metrics compare to two conceptual models. The results have been evaluated according to Nash-Sutcliffe Efficiency (NSE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R2 performance indicators. During calibration, the NSE values for the LSTM, HBV, and Mapshed models were calculated as 0.82, 0.74, and 0.72, respectively, whilst during validation, the NSE values were calculated as 0.90, 0.75, and 0.50, in the same order. The study demonstrates that, despite karstic geology, complex Mediterranean climate patterns and low temporal resolution of the available data, LSTM and HBV models can provide reliable river flow forecast.
Keywords
Main Subjects
)