Recurrent neural networks and sensitivity analysis for accurate monthly evapotranspiration estimation in the region of Fez, Morocco

Abstract

AbstractGood water management is essential, including addressing water scarcity, which is exacerbated by climate change and increasing food demand due to population growth. In modern era of high technology, artificial intelligence and the Internet of Things have an important role to play in decision‐making approaches. According to hydro‐agrological studies, an accurate assessment of evapotranspiration is necessary for irrigation control, and to have an in‐depth understanding of the environmental factor interactions, a sensitivity analysis should be carried out. One of the tools, alongside empirical equations, is artificial intelligence for time‐series prediction. Thus, we compared several models for Penman–Monteith ET0 estimation to see which one performs better in the arid area of Morocco using meteorological data from the region of Fes. They were evaluated according to the MSE, RMSE, MAE, MAPE, and R2 and the variance of error distribution to show the performances of linear regression, K‐Nearest Neighbor, decision tree, random forest, support vector regression, long short‐term memory, and artificial neural network. According to the findings, LSTM outperformed the models with satisfactory results and an accuracy rate of 99.82%. An underlying mechanism of sensitivity analysis was also introduced to find the contribution of each element and estimate the target with a limited dataset. The findings show satisfactory results of R2 = 98.97%. The distribution and reliability of the prediction were proven using the Taylor diagram and Kruskal–Wallis test for the effectiveness of the study. This research demonstrates the potential of employing data‐driven techniques for evapotranspiration estimation to improve the efficacy of water management strategies. This will help to address present issues and establish sustainable water practices in the face of changing environmental conditions.