An Interpretable Machine Learning Framework for Unravelling the Dynamics of Surface Soil Moisture Drivers

Understanding the impacts of spatial non-stationarity of environmental factors on surface soil moisture (SSM) in different seasons is crucial for effective environmental management. Yet, our knowledge of this phenomenon remains limited. To fill this gap, we have introduced a framework combining the model agnostic SHapley Additive exPlanations (SHAP) technique with a two-step clustering analysis to provide spatial interpretations for machine learning models. Due to spatial and temporal limitations of in-situ SSM data in Iran, we evaluated the performance of global daily datasets, including SMAP, MERRA-2, and CFSv2, from 2015 to 2023 using the agrometeorological stations to select the best dataset for the machine learning process. Subsequently, SSM values were calculated for Iran’s 609 catchments using the most optimal dataset. Furthermore, we selected a set of climatological, hydrological, topographical, vegetation, and soil factors influencing SSM. While Random Forest (RF) was employed to estimate SSM, the SHAP technique was used for RF spatial interpretation in conjunction with the two-step cluster analysis. The results revealed that among the datasets, SMAP exhibited the highest median correlation and the lowest median root mean square error when compared to in-situ stations. Findings indicated that the RF model can offer SSM estimates with R2 values of 0.89, 0.83, 0.70, and 0.75 for the winter, spring, summer, and autumn seasons, respectively. The results of SHAP and two-step clustering demonstrated that climatic factors primarily influence winter and autumn SSM. In contrast, in spring and summer, SSM is predominantly affected by vegetation and soil characteristics. These findings provide valuable reference points and policy recommendations for effectively managing SSM in various catchments.