Data- and Model-Based Discharge Hindcasting over a Subtropical River Basin

This study aims to evaluate the performance of Soil and Water Assessment Tool (SWAT), a simple Auto-Regressive with eXogenous input (ARX) model, gene expression programming (GEP)-based model in one-day-ahead discharge prediction for upper Kentucky River Basin. Calibration models were carried out period 2002–2005 using daily flow at stream gauging station unaffected by regulation. Validation calibrated executed 2008–2010 same along another 88 km downstream. GEP provided best calibration (coefficient determination (R) value 0.94 Nash-Sutcliffe Efficiency (NSE) 0.88) validation (R values 0.93 0.93, NSE 0.87 0.87, respectively) results two stations. While SWAT performed reasonably well 0.85 0.72), its somewhat degraded 0.82, 0.65 0.65, stations). ARX very 0.92, 0.82) 0.88 0.76 0.85) Research suggest that sophisticated hydrological could be outperformed data-driven has advantage generate functional relationships allows investigation complex nonlinear interrelationships among variables.