Comparing AquaCrop and Machine Learning Regression Models for Predicting Actual Evapotranspiration in Paddy Crops: Surin Province, Thailand

Optimizing agricultural irrigation water use is critical for crop efficiency and balancing supply and demand, under growing water stress driven by population growth. Crop growth models provide quantitative data for water management, but the development of a robust crop model is challenging due to unknown and uncertain parameters. This study evaluates and compares the predictive performance of six regression-based machine learning (ML) models against AquaCrop simulated actual evapotranspiration (AET) for paddy crops in Surin province, Thailand, using 40 years (1984–2023) of climate data. We compare three distinct scenarios: (1) all climate variables, (2) only highly correlated variables, and (3) only temperature data. Results show that the artificial neural network model effectively predicts AET, with a better performance across all scenarios. Reliable AET predictions through ML models are important for stakeholders in countries like Thailand, where the economy heavily depends on agriculture and paddy crop exports.

Readme Text: To get an output for the study region other than Surin, Thailand, follow the methodology of the manuscript (available at: https://doi.org/10.2166/wpt.2026.325). The stored output folder contains AquaCrop model output - Evapotranspiration (1984-2023), and Machine Learning regression model output (*.MAT files) - 3 different scenarios. 
(1) all climate variables, (2) only highly correlated variables, and (3) only temperature data.

In mat files, there are various variables: val and test are calibration/validation (1984-2018) and testing/prediction (2019-2023) datasets. x and y are the input (predictors) and output (predictand) data obtained from the Aquacrop model (similar to AquaCrop_Data.csv). y1...y6 are various ML models in the following sequence: Multiple Linear Regression (MLR), Decision Tree (DT), Support Vector Machine (SVM), Gaussian Process Regression (GPR), Ensemble of Trees (EoT) , and Artificial Neural Network (ANN).