Understanding Seasonal Variability of Soil Moisture Patterns with Explainable Machine Learning

Soil moisture plays a critical role in regulating the water and energy flux between land and atmosphere, influencing hydrological processes, vegetation dynamics, and climate feedbacks. The traditional data-driven approaches such as correlation and regression have been widely used for understanding the relationships between soil moisture and its controlling drivers, but these methods are limited in capturing complex nonlinear interactions between the multiple drivers, and not much robust enough to quantify the dominant key drivers that control soil moisture patterns. Machine learning algorithms can overcome this limitation by learning the complex non-linear interaction directly from observational datasets. However, the black box nature of the model limits its explainability, which can be improved by the integration of explainable artificial intelligence (XAI). The present study aims to understand the climatic and vegetation drivers controlling the seasonal variability of soil moisture patterns at a remote sensing scale across India, using a random forest modelling framework integrated with XAI. Satellite-derived soil moisture and other hydroclimatic and vegetation drivers were analysed at a large scale (36 km) across seasons. The result shows that the model performs well in capturing the grid-wise temporal variability of soil moisture based on seasons. The XAI based interpretation identifies precipitation as the dominant controlling driver during the monsoon season, covering nearly 70 percent of the areal extent across semi-arid and sub-humid regions. The diurnal land surface temperature and evaporative fraction are identified as the dominant drivers across the arid regions during the non-monsoon seasons. Our findings with the aid of model with XAI integrated explainability techniques, helps in understanding the complex drivers affecting soil moisture patterns across seasons in India, which is essential for improving weather and climate forecasting models and better preparedness for extreme events, including drought.