Advanced reference crop evapotranspiration prediction: a novel framework combining neural nets, bee optimization algorithm, and mode decomposition

Abstract

Various critical applications, spanning from watershed management to agricultural planning and ecological sustainability, hinge upon the accurate prediction of reference evapotranspiration (ET_o). In this context, our study aimed to enhance the accuracy of ET_o prediction models by combining a variety of signal decomposition techniques with an Artificial Bee Colony (ABC)–artificial neural network (ANN) (codename: ABC–ANN). To this end, historical (1979–2014) daily climate variables, including maximum temperature, minimum temperature, mean temperature, wind speed, relative humidity, solar radiation, and precipitation from four arid and semi-arid regions in Egypt: Al-Qalyubiyah, Cairo, Damietta, and Port Said, were used. Six techniques, namely, Empirical Mode Decomposition, Variational Mode Decomposition, Ensemble Empirical Mode Decomposition, Local Mean Decomposition, Complete Ensemble Empirical Mode Decomposition with Adaptive Noise, and Empirical Wavelet Transform were used to evaluate signal decomposition efficiency in ET_o prediction. Our results showed that the highest ET_o prediction accuracy was obtained with ABC-ANN (Train R²: 0.990 and Test R²: 0.989), (Train R²: 0.986 and Test R²: 0.986), (Train R²: 0.991 and Test R²: 0.989) and (Train R²: 0.988 and Test R²: 0.987) for Al-Qalyubiyah, Cairo, Damietta, and Port Said, respectively. The impressive results of our hybrid model attest to its importance as a powerful tool for tackling the problems associated with ET_o prediction.