Evaluating machine learning models for estimating evapotranspiration in Colombia’s Cauca River Valley

Water loss in soil and plants occurs through processes such as evaporation and transpiration, giving rise to evapotranspiration. In water resource management and precision agriculture, it is important to have accurate estimates of water loss in order to optimize agricultural irrigation and improve crop productivity. To estimate evapotranspiration data, this study used different machine learning methods such as multiple linear regression, polynomial regression, and neural networks using libraries such as Scikit-learn, Statsmodels, TensorFlow, and Keras. Climate data gathered by the Cenicaña weather station located in Florida, department of Valle del Cauca, Colombia, were used to train the machine learning models and later compare evapotranspiration estimates based on the FAO 56 Penman-Monteith method. To ensure the best estimate, 13 different data combinations were used, each combination involving three of the six parameters studied. Combinations involving relative humidity and minimum, maximum or mean temperatures showed the lowest precision, presenting values such as R = 0.7, R² = 0.51 and RMSE = 0.57, as compared with combinations that considered net radiation and wind speed, which presented the most accurate predictions with values of R = 0.99, R² = 0.98 and RMSE = 0.73. These results highlight the relevance of machine learning models in estimating evapotranspiration, particularly in areas with limited availability of climate data.