Machine learning methods for estimating reference evapotranspiration

Abstract

Precise estimation of evapotranspiration is crucial for optimizing crop water uses particularly in the context of agriculture and horticultural production. In this study, various machine learning techniques was used to determine reference evapotranspiration by leveraging historical weather data. The models tested include artificial neural networks (ANN), Lasso, Ridge, Random Forest, LGBM regressor, and Gradient boosting regressor. LGBM regressor emerged as the top-performing model, exhibiting exceptional accuracy with a testing R-squared of 1.0. ANN also demonstrated notable performance, achieving a testing R-squared of 0.99. Moreover, the Random Forest and Gradient boosting regressor models showcased strong predictive capabilities, with R2 values of 0.99 and 0.98, respectively. These models offer valuable alternatives for estimating evapotranspiration, providing robustness and adaptability to diverse environmental datasets.