Daily reference evapotranspiration estimation utilizing deep learning models with varied combinations of weather data

Abstract

Effective irrigation planning pivots on the meticulous monitoring of ETo (the reference evapotranspiration), a fundamental variable in diverse studies. The go-to method for approximate ETo, the FAO-56 Penman-Monteith (FAO-56 PM) equation, demands an array of weather data, encompassing relative humidity, temperature, solar radiation, and wind speed. However, this data-intensive requirement presents challenges in situations where such information is limited, and artificial intelligence is being used to address this challenge, come into play to estimate ET0 with a streamlined set of parameters. The study begins with a comprehensive analysis, comparing the performance of Penman-Monteith (FAO-56 PM) and (ASCE_PM) with deep learning models such as artificial neural networks (ANN) and one-dimensional convolutional neural networks (CNN 1d).The principal aim is to estimate daily reference evapotranspiration (ETo) in the region of Morocco, specifically Meknes, employing a minimal set of meteorological variables across various combinations of measured data on the fundamental variables that constitute ETo. These combinations encompass scenarios involving all four variables, different combinations of three, two, and each variable in isolation. Two implementation scenarios are considered: (i) cross-validation across all datasets and (ii) training with one station and validating with another. Across these varied techniques, commendable results emerge, portraying a favourable comparison against empirical models reliant on minimal meteorological data.