Improvement the estimation of reference evapotranspiration by combining different types of meteorological data Using machine learning models

Abstract

Irrigation and the strategic planning thereof play a pivotal role in diverse hydrological inquiries, with reference evapotranspiration (ETo) standing as a paramount variable within this domain. While the equation (FAO-56 PM)is extensively employed for (ETo)estimation, its dependence on numerous weather datas such as solar radiation, temperature, relative humidity, extraterrestrial radiation and wind speed, introduces inherent constraints, the remote computation necessitates a substantial array of sensors, thereby incurring considerable expenses. To surmount this challenge, artificial intelligence methodologies, encompassing various machine learning (ML) models, are harnessed for ETo estimation, requiring only minimal parameters.This investigation scrutinizes the effectiveness of alternative equations (Hargreaves-Samani, Romannenko, Jensen-Haise, ASCE_PM) vis-à-vis (ML) models such as Xgboost,Support Vector Machine (SVM), and Random Forest (RF) in the estimation of ETo across the Meknes region, utilizing diverse permutations of the four measured variables. The study employs an extensive array of hyperparameters in two distinct scenarios: (i) randomization of all data, and (ii) training on one station while validating on another. All methodologies employed in this study yield satisfactory outcomes when juxtaposed against empirical models reliant on minimal meteorological data.