Evaluating SWAT model for streamflow estimation in the semi-arid Okavango-Omatako catchment, Namibia

Abstract

A semi-distributed hydrological model was used for runoff estimation in the Okavango-Omatako catchment in Namibia. The model was configured for a 31-year period from 1985 to 2015. Subsequently, calibration and validation processes followed using the SUFI-2 algorithm. For evaluating catchment simulation, two methods were used: i. model prediction uncertainty measured by P-factor and R-factor and ii. Model performance indicators, that is, Nash-Sutcliffe Efficiency (NSE), Coefficient of determination (R 2 ), Percent bias (PBIAS), and Residual variation (RSR). The P-factor achieved 0.77 and 0.68 while R-factor attained 1.31 and 1.82 for calibration and validation, respectively. The following indicators were used to evaluate the model performance through calibration and validation results respectively; NSE with 0.82 and 0.80, R 2 with 0.84 and 0.89, PBIAS achieving -20≤PBIAS≤-1.1 and RSR performing 0.42 and 0.44. All performance indices achieved very good ratings apart from PBIAS validation which rated as satisfactory. The semi-arid characteristics together with relatively flat terrain features justified the need for the evaluation of model performance using discharge data in our study region. SWAT demonstrated reasonable results in modelling semi-arid streamflow with high and low flows adequately captured. Consequently, this evaluation was necessary for further investigations into impacts of climate change on scarce water resources highlighting the challenges of SWAT model applications in our study area climatic regime and other similar regions globally for further model improvements.