Global multi-scale surface soil moisture retrieval coupling physical mechanisms and machine learning in the cloud environment

Surface soil moisture (SSM) is a critical state variable for water cycle research, and the advances in satellite remote sensing technology have provided a novel means for acquiring large-scale SSM data. While satellite microwave remote sensing-based SSM retrieval has emerged as the dominant approach for global SSM product development, offering numerous advantages, it still faces significant challenges. These include the trade-off between model accuracy and generalizability, the limitations of applying uniform retrieval models across diverse environments, and the inherent complexity and computational demands of the retrieval process. To address these common issues in microwave remote sensing-based SSM retrieval studies, this study proposed a cloud-based intelligent retrieval framework for global high-accuracy SSM estimation. This framework integrated physical mechanisms with machine learning models to ensure robust generalization and high retrieval accuracy; additionally, a model selection module was incorporated to enhance the overall retrieval accuracy by providing environment-specific retrieval models. In an assessment based on global validation sites for 1-km resolution SSM retrieval, the proposed framework performed well, with an R value of 0.851 and an ubRMSE of 0.058 m³·m⁻³. Furthermore, to mitigate the computational resource demands and time-consuming of the retrieval process, the SSM retrieval framework was implemented in a cloud environment utilizing Google Earth Engine, Drive, and Colab, thereby enabling seamless online operation of the entire retrieval process. This cloud-based intelligent retrieval framework facilitates real-time point-scale SSM retrieval on a global scale and rapid production of high-accuracy SSM products at the regional scale (SSM products for China at 1 km resolution can be accessed via https://tinyurl.com/SSMproduct). The SSM retrieval framework can significantly contribute to agricultural, environmental, and other related fields, and serve as a reference for the retrieval of other environmental variables.