Root zone soil moisture (RZSM) is critical for irrigation management, as it directly affects plant water availability, crop growth, and irrigation scheduling. However, modeling RZSM is challenging due to the high variability and nonlinearity of soil moisture patterns. Physically based models, such as those solving Richards’ equation, offer detailed soil dynamics but require extensive hydrological parameters and significant computational resources. In contrast, statistical approaches are more efficient but lack physical interpretability. This study proposes an event-based framework that models soil moisture increases following individual water input events (e.g., precipitation, irrigation). Water balance models are designed to simulate moisture changes within each soil layer, while an XGBoost ensemble captures interlayer interactions. By embedding the machine learning model within physically structured equations, the approach ensures both accuracy and interpretability. The model was applied using soil moisture sensor data collected from weather stations with and without crops in Florida. Results show strong accuracy in event-pattern identification and the modelling of event-scale duration and magnitude behavior across locations.
Zhang et al. (Sun,) studied this question.