Multi-site multi-objective calibration of SWAT model using a large dataset for improved performance in Ethiopia

In hydrological studies, Soil and Water Assessment Tool (SWAT) model is usually calibrated for the data of a single site, which is often the watershed outlet, employing a single objective function describing the error criterion. It is, however, quite possible that a basin is gauged at multiple sites, and therefore, the model can be more reliably calibrated using a larger set of data as input besides allowing for compensating the instrumental/human errors associated with the measurements at one site by the data of other site(s). In addition, depending on the purpose of the study, i.e., whether high or low flows, water availability, low flows, and so on, a set of error criteria can be identified and employed in model calibration. Following this concept, the popular SWAT model was calibrated/validated using the data of Hombole, Weyb, and Gidabo watersheds located in Ethiopia and measured for discharge at Hombole and Melka Kutire stations in Hombole watershed, Denbel and Alem Kerem stations in Weyb watershed, and Measa and Aposto stations in Gidabo watershed. To this end, the most influential parameters were selected first using the global sensitivity function of the SWAT CUP routine. Then, the model was calibrated following (i) simultaneous multi-site calibration (SMSC), where data of measuring stations were used simultaneously in a single calibration; and (ii) conventional calibration (CC), i.e., individual calibration in sequence from upstream gauging station to downstream gauging station. The calibration used five objective functions, viz., statistical error criteria involving NSE, KGE, PBIAS, RSR, and R2. The SMSC technique performed better than CC in all applications to three watersheds. Additionally, the effect of the wet and dry seasons on SWAT performance was also assessed. The SWAT model performed well for the wet period and poorly for the dry period.