Towards an explainable irrigation scheduling approach by predicting soil moisture and evapotranspiration via multi-target regression.

The article focuses on the development and evaluation of a Multi-Target Regression model for estimating soil moisture (Volumetric Water Content, VWC) and evapotranspiration (ET) to improve irrigation scheduling. The proposed MTR-SMET model jointly predicts ET and VWC using meteorological and soil data, enabling calculation of daily irrigation needs through two established methods—ET-based and soil moisture-based irrigation—while also detecting inconsistencies such as sensor errors. To address the interpretability challenge of machine learning in agriculture, the study introduces an explainable AI extension (xMTR-SMET) that provides visual explanations of predictions to assist farmers in understanding irrigation decisions. Experimental results demonstrate that the RF-PCT-based MTR-SMET model achieves high accuracy (MSE = 0.00015, RMSE = 0.0039, MAE = 0.002, R² = 0.9676) and that the combined irrigation approach aligns well with expert assessments, with the explainability module offering both global and local insights into model behavior and prediction drivers.