Coupling a Bat Algorithm with XGBoost to Estimate Reference Evapotranspiration in the Arid and Semiarid Regions of China

Abstract

Accurate estimation of reference evapotranspiration (ETo) is key to agricultural irrigation scheduling and water resources management in arid and semiarid areas. This study evaluates the capability of coupling a Bat algorithm with the XGBoost method (i.e., the BAXGB model) for estimating monthly ETo in the arid and semiarid regions of China. Meteorological data from three stations (Datong, Yinchuan, and Taiyuan) during 1991–2015 were used to build the BAXGB model, the multivariate adaptive regression splines (MARS), and the gaussian process regression (GPR) model. Six input combinations with different sets of meteorological parameters were applied for model training and testing, which included mean air temperature (Tmean), maximum air temperature (Tmax), minimum air temperature (Tmin), wind speed (U), relative humidity (RH), and solar radiation (Rs) or extraterrestrial radiation (Ra, MJ m−2·d−1). The results indicated that BAXGB models (RMSE = 0.114–0.412 mm·d−1, MAE = 0.087–0.302 mm·d−1, and R2 = 0.937–0.996) were more accurate than either MARS (RMSE = 0.146–0.512 mm·d−1, MAE = 0.112–0.37 mm·d−1, and R2 = 0.935–0.994) or GPR (RMSE = 0.289–0.714 mm·d−1, MAE = 0.197–0.564 mm·d−1, and R2 = 0.817–0.980) model for estimating ETo. Findings of this study would be helpful for agricultural irrigation scheduling in the arid and semiarid regions and may be used as reference in other regions where accurate models for improving local water management are needed.