Explainable Machine Learning for Evapotranspiration Prediction

Abstract

The current study aims to develop efficient machine learning models that can accurately predict potential evapotranspiration, an essential parameter in agricultural water management. Knowing this value in advance can facilitate proactive irrigation scheduling. Two models, Long Short-Term Memory and eXtreme Gradient Boosting, are evaluated using performance metrics such as mean squared error, mean average error, and root mean squared error. One of the challenges with these models is their lack of interpretability, as they are often referred to as ”black-boxes.” To address this issu, the study provides global explanations for how the best-performing model learns. Additionally, the study incrementally improves the model’s performance based on the provided explanations. Overall, the study contributes to developing more accurate and interpretable machine learning models for predicting potential evapotranspiration, which can improve agricultural water management practices.