Evaluating CFSR and WATCH Data as Input to SWAT for the Estimation of the Potential Evapotranspiration in a Data-Scarce Eastern-African Catchment

Abstract

One of the key inputs of a hydrological model is the potential evapotranspiration (PET), which sets an upper limit to evapotranspirative water demand. However, limited data availability often challenges the choice of a PET estimation method, which in turn affects the PET estimates as well as the water balance (WB) components. The objectives of this research are (1) to evaluate the use of different sources of weather input data to derive PET: Climate Forecast System Reanalysis (CFSR) data, Water and Global Change (WATCH) data, and data generated by the weather generator of the soil and water assessment tool (SWAT) (SWAT–WG); and (2) to investigate the effects of the Penman–Monteith and Hargreaves (HG) methods on WB components using a SWAT-based model for the Upper Mara Catchment (Kenya). It is shown that PET estimations using the CFSR, WATCH, and SWAT–WG data sets compare well with the average annual and seasonal PET estimates from local observations over a period of 20 years. This shows the potential of global reanalysis climate data sources for the computation of PET in data-limited catchments. The SWAT models forced by these data sets and by gauged rainfall show a modified Nash–Sutcliffe (NSm) efficiency ranging from 0.6 to 0.72 for the simulation of the flow, depending on the selected PET estimation method. Unlike water yield, the other WB components simulated by the SWAT models (ET, deep aquifer loss, and reevaporation from the shallow aquifer) vary in magnitude, depending on the data and methods being used. The sensitivity analysis and the calibration results show that the model parameters are sensitive to the choice of the PET estimation method. Therefore, for catchments where parameterization is a challenge owing to data scarcity, it is crucial to consider the appropriate PET estimation method for a realistic modeling of the hydrological processes. In our case study, the HG method gives more robust and realistic ET estimations.