Global error assessment in satellite-based soil moisture data: Harnessing machine learning and triple collocation analysis for SMAP, SMOS, and ASCAT

The accuracy of satellite-based soil moisture (SM) data is a fundamental aspect for a broad range of applications, including climate change modeling, weather prediction, and agricultural planning. While several statistical methods have been proposed for estimating error statistics in large-scale SM datasets, a fully global error map has not yet been achieved due to the inherent limitations in these techniques. Our research takes a novel approach by employing machine learning (ML) to address these limitations and to fill the spatial gaps that have emerged in triple collocation analysis (TCA) results. This method has allowed us to construct detailed error maps for SM data derived from three microwave missions - Soil Moisture Active Passive (SMAP), Soil Moisture and Ocean Salinity (SMOS), and Advanced Scatterometer (ASCAT).In addition, we utilize a model-agnostic interpretation technique, SHapley Additive exPlanations (SHAP) values, to investigate the effect of various environmental factors on the quality of satellite-based SM retrievals. This analysis provides us with a deeper understanding of how and why error varies across different terrains and climates. Applying this ML approach on a global scale, we found that we were able to reconstruct 72.0% of missing error information that was previously unattainable due to data limitations or unfeasibility in TCA. Furthermore, our results revealed that significant portions of the Earth's SM dynamics, 22.7% (a.m.) and 34.2% (p.m.), have not been fully explored across these three satellite missions.Our research represents a substantial step forward in the field of global error mapping for satellite-based SM data. Our method not only overcomes some of the limitations of existing error characterization approaches but also opens new opportunities for enhancing the accuracy and usability of satellite-based SM data in multiple sectors.