An evaluation of the performance and subsequent calibration of two reference evapotranspiration estimation models for Gweru, Zimbabwe

Abstract

This study evaluated the performances and subsequently calibrated the Hargreaves and Blanney Criddle reference evapotranspiration (ETo) estimation models for the semi arid Gweru climate in Zimbabwe. The FAO –56 Penman Monteith model was used as the standard method for the evaluation. The ETo calculator (Version 2.3) was used for the computation of daily ETo for all models using observed maximum temperatures, minimum temperatures, mean temperatures, mean relative humidity and wind speed at 2 m height as input data. The Hargreaves and Blanney Criddle models were evaluated using the Model Efficiency (EF), the Bias (B), the Root Mean Square Error (RMSE), the Mean Absolute Prediction Error (Err) and the coefficient of determination (R2). The Blanney Criddle model was found to be highly accurate (Errof 6.807%), a B of -0.303 mm day-1, a MSE of 0.404 mm day-1, an EF of 0.926 and an R2 of 0.970. The Hargreaves model had a less accurate but rather good performance as evidenced by an Errof 12.06%, a RMSE of 0.855 mm day-1, a B value of -0.211 mm day-1, EF of 0.66 and an R2 of 0.751. After calibration, the performance of the Blanney Criddle model was within acceptable limits of the evaluation methods used. The Err dropped to only 3.925%, the RMSE to 0.269 mm day-1and the B value to -0.00012 mm day-1. The EF also increased to 0.965. There was little change in the performance of the Hargreaves model even after calibration. The EF increased to 0.703 and the B value decreased to -0.081 mm day-1. The Err, RMSE and the R2 remained largely unchanged. From these observations, it can be concluded that the Blanney Criddle model is the best method to use for daily ETo estimation for the semi arid region of Gweru because of its highly accurate prediction accuracy. For this site, it is highly recommended to use 0.303 as the value of the constant a in the equation instead of the default 0 value. The site specific k value of the Hargreaves model is 0.0025 instead of the default 0.0023.