Influences of Potential Evapotranspiration Estimation Methods on SWAT's Hydrologic Simulation in a Northwestern Minnesota Watershed

Abstract

The Soil and Water Assessment Tool (SWAT), a widely used watershed hydrology and water quality model, provides three different methods (Hargreaves, Priestley-Taylor, and Penman-Monteith) for estimating potential evapotranspiration (PET) and the corresponding actual evapotranspiration (AET). Although these methods have been extensively tested, the effects of using them within SWAT's framework are largely unknown. The objective of this study was to test the three PET methods within SWAT's framework using data collected in the Wild Rice River watershed, located in northwestern Minnesota. The performance of the SWAT models was measured using three statistics: the Nash-Sutcliffe coefficient (Ej2), coefficient of determination (R2), and performance virtue (PVk). The three models were independently calibrated and validated using the observed daily stream flows at two USGS gauging stations. The simulated stream discharges were compared with the corresponding observed values and the estimated evapotranspiration examined in accordance with the wet-environment areal evapotranspiration (ETW) derived from the evaporation data for Williams Lake, located about 100 km southeast of the study watershed. The use of the three PET methods resulted in different values for two calibration parameters, namely the soil evaporation compensation factor and SCS curve number. At the lower station, which is near the watershed outlet, the observed annual mean discharge (8.33 m3/s) during the model validation period was predicted to be 10.25, 10.87, and 9.69 m3/s by SWAT-Penman, SWAT-Priestley, and SWAT-Hargreaves, respectively. The annual mean discharge (10.83 m3/s) was more accurately predicted during the model calibration period, with an absolute error of less than 0.5 m3/s. The prediction errors for the upper station were comparable with those for the lower station. In addition, all three models exhibited good performance when simulating the monthly, seasonal, and annual mean discharges (Ej2 >0.75 and PVk >0.80) and satisfactory performance when predicting the daily stream flows (Ej2 >0.36 and PVk >0.70). In estimating evapotranspiration for the study watershed, SWAT-Hargreaves seemed to be slightly superior to the other two models, while SWAT-Priestley might be more appropriate for an ETW value greater than 8.0 mm/d. Nevertheless, the AET values estimated by the three models shared a concurrent spatial pattern and temporal trend, and were insignificantly different from each other at a 5% significance level (p-values > 0.05). The results indicated that after calibration, using the three ET methods within SWAT produced very similar hydrologic (AET and discharge) predictions for the study watershed.