Comparative study of reference evapotranspiration estimation models based on machine learning algorithm: a case study of Zhengzhou City

Abstract

Abstract Reference evapotranspiration (ET0) is an important parameter to characterize the hydrological water cycle and energy balance. An extremely heavy rainstorm occurred in Zhengzhou City, Henan Province on 20 July 2021, causing heavy casualties and economic losses. One of the important reasons for this rainstorm was abnormal water circulation. The purpose of this study is to estimate ET0 accurately and avoid extreme disasters caused by abnormal water cycles. This study compared and analyzed the accuracy and robustness of ET0 prediction based on the improved Levenberg–Marquardt (L-M) model based on artificial neural network and the genetic algorithm-backward neural network (GA-BP) model. The model uses seven weather stations in Zhengzhou, including mountain climate and plain climate. By utilizing the Pearson correlation analysis technique, six distinct input scenarios were identified, and the efficacy of the model was assessed using evaluation metrics, including RMSE, MAE, NSE, and SI. The results show that the estimation accuracy of the L-M model is better than that of the GA-BP model; when the number of input meteorological parameters is the same, the combined simulation effect including wind speed is the best; the R2 of L-M3 and L-M4 are 0.9285 and 0.9675, respectively; models can accurately estimate ET0 with limited data.