Predictive Modeling and Comparative Analysis of Reference Evapotranspiration with Machine Learning Algorithms

Abstract

Accurate estimation of reference evapotranspiration (ET0) is crucial for a multitude of applications, encompassing drought detection, irrigation scheduling, water resource management, and disaster risk reduction. This investigation utilized the FAO-PM equation for ET0 estimation and subsequently incorporated meteorological variables as input variables with machine learning (ML) models to enhance ET0 predictions. The dataset was bifurcated into training and testing data segments. Four distinct machine learning models were deployed in this study, namely Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting Machine (GBM), and Linear Regression (LR). The performance of these models was evaluated using various statistical indices, including Mean Absolute Error (MAE), Mean Sum of Error (MSE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and the coefficient of determination (R2), to pinpoint the most efficacious ML algorithm. After conducting a comprehensive analysis involving both training and testing data, the results unequivocally identify GBM with MAE values of 0.054 and 0.077, MSE values of 0.005 and 0.011, MAPE values of 0.014 and 0.022, RMSE values of 0.072 and 0.107, and an R2 value of 0.096 and 0.092 during training and testing, respectively. This model has been selected as the optimal choice for precise ET0 estimation within the study region. Subsequently, SVM, RF, and LR follow as alternatives in terms of performance, in descending order.