ACTA GEOPHYSICA, cilt.70, sa.2, ss.897-915, 2022 (SCI-Expanded)
Hydrological modelling is essential for improving water management and planning efficiency and sustainability. In this study, lumped conceptual models [i.e., Genie Rural a 4 parametres Journalier (GR4J), Genie Rural a 6 parametres Journalier (GR6J)] and wavelet-based data-driven models [Wavelet-Genetic algorithm-Artificial neural network (WGANN), Wavelet-based support vector regression (WSVR)] were used for daily rainfall-runoff modelling by using three gauging stations, namely caydere Egirdir Gol Giris, Kargi c. Turkler and Naras D. Siseler, in semi-arid and humid areas of Antalya basin, Turkey. The Nash Sutcliffe efficiency (NSE), index of agreement (d) and root mean square error (RMSE) were used to evaluate the model performance. Although conceptual and data-driven models yielded a good performance, data-driven models could be more helpful, especially in semi-arid and small basins, challenging for conceptual models due to nonlinearity and complexity. The best runoff forecasting performance improvement was observed in caydere Egirdir Gol Giris with the WGANN (NSE = 0.96, d = 0.99, RMSE = 0.5 mm/d), WSVR (NSE = 0.95, d = 0.99, RMSE = 0.6 mm/d) against the GR4J (NSE = 0.53, d = 0.79, RMSE = 1.8 mm/d) and the GR6J (NSE = 0.49, d = 0.78, RMSE = 1.8 mm/d). It was also found that the GR4J and GR6J yielded a similar performance. Data denoising via wavelet transformation and input selection had a significant role in developing performance for the data-driven models. Data-driven models yielded better results for the forecasting of extreme flows. In this regard, using and integrating the useful parts of the conceptual and data-driven models could be more favourable.