Estimated Daily Reference Evapotranspiration Using Machine Learning and Deep Learning Based on Various Combinations of Meteorological Data

Abstract

The agricultural sector is the largest consumer of water resources compared to all other sectors. For effective planning of irrigation systems in agriculture, plant water requirements are a factor that must be calculated accurately. The evapotranspiration reference (ET0) is used to calculate the real need for water of a given plant. The calculation of ET0 poses a challenge given the limited availability of meteorological data and the consequent number of variables that enter into its equation. In order to estimate the ET0 using only a few parameters, we have developed machine learning models such as SVR, RF and deep learning, namely long-term memory (LSTM), based on a time series of climate data. In this present work we will present the result of the best model -the LSTM-. The data used in this study were obtained from our weather station, including maximum air temperature (Tmax), minimum air temperature (Tmin), average relative humidity (RH), average solar radiation (SR) and average wind speed. (u2). Our proposed model was trained and tested on two global scenarios, A and B. Scenario A consists of 8 different combinations which are generated from the collected data, while Scenario B additionally contains combinations of data from Scenario A the day of the year as a new variable. ET0 calculated using the FAO Penman Monteith approach was taken as the output of our model. Good results were obtained for all the combinations of data used in this study, in particular those of scenario B.