Prediction of daily reference crop evapotranspiration in different Chinese climate zones: Combined application of key meteorological factors and Elman algorithm

Reference crop evapotranspiration (ETO) is a key factor for estimating crop water requirements, which guide agricultural irrigation. To improve the accuracy of predicting ETO in different climate zones in China that lack meteorological data, an ETO hybrid model based on K-nearest neighbour (KNN) machine learning algorithm for extracting factor contribution rates is proposed in this study. Meteorological factors with large contribution rates were selected as input, and a prediction model for ETO was established using the Elman daily ETO prediction model. The ETO prediction model was optimised using three optimisation algorithms [Genetic optimization algorithm (GA), Cuckoo optimization algorithm (CS) and Whale optimization algorithm (WOA)] to improve the accuracy of ETO prediction. The results revealed that surface radiation (Rs) is the most important factor in estimating ETO (contribution rate = 0.392-0.626), followed by temperature factors (T; including maximum, minimum, and average temperatures). And each model has the highest accuracy with the input combination of Rs and T. For the different machine learning models, the CS-Elman model had the highest accuracy (RMSE = 0.468-2.235, R-2 = 0.567-0.928, MAE = 0.363-1.343, and NSE = 0.345-0.923), and the machine learning model had higher accuracy than the experience model. The CS-Elman model performed more favourably in tropical monsoon and subtropical monsoon regions than that in other areas, and the model performed best at the junction of two climatic zones. The results can provide a theoretical basis for high-precision prediction of ETO in different climate zones in China.