Improving Global Surface Soil Moisture Prediction through Physics-Guided Deep Learning and Cluster-Based Regionalization

<p>Surface soil moisture (SSM) is essential in the hydrological cycle and land-atmosphere interactions, and its accurate simulation is crucial for climate prediction and resource management. This study employed an innovative physics-guided deep learning (PGDL) model that integrates the physical knowledge from the Terrestrial Ecosystem Model (TEM) with the temporal learning capacity of the long short-term memory (LSTM) to improve global SSM prediction. By introducing an optimized clustering strategy based on multi-source features, the globe was divided into subregions with consistent characteristics. Within this framework, cluster-specific models were trained using in situ observations and subsequently extended to global predictions and evaluations with satellite data. This clustering approach enhanced model generalization across diverse climatic and geographic conditions, ensuring more robust predictions based on environmentally consistent samples. Overall, this study demonstrates the superiority of the PGDL model for global SSM simulation and highlights the importance of clustering strategies in model construction and evaluation, providing new insights for achieving more accurate and robust SSM predictions across heterogeneous environments.</p>