A Novel Merging Framework for Generating High-Accuracy Global Soil Moisture by Error Decomposition Through Multiple Collocation Analysis

Surface soil moisture (SSM) information from different sources exhibits distinct error characteristics. Data merging offers an effective method to combine the advantages of multisource data and to enhance SSM estimation accuracy. The triple collocation (TC)-based merging approach under the linear weighted average (LWA) scheme was widely used for SSM data combination. However, it can only consider three datasets simultaneously and the accurate estimation of errors and merging weights are generally affected by nonzero error cross-correlations (ECCs). This study proposes a novel merging framework based on the multiple collocation (MC) analysis. The MC technique can estimate the errors of an arbitrary number of datasets simultaneously by solving space distances under the constraint of the Pythagorean theorem in Hilbert space. It addresses nonzero ECCs by the separation of structural and nonstructural errors. The MC-based merging weights were derived for seven individual SSM datasets from active and passive microwave satellites and land surface models (LSMs). In addition, the random forest (RF)-based gap filling was applied to different land cover types to improve the spatial coverage of the merged dataset. Validations against ground observations showed that the MC-based merged SSM dataset achieved significant performance improvements compared with parent SSM datasets. The MC-based merged dataset had comparable overall accuracy in detecting ground SSM temporal dynamics, compared with the TC-based merged datasets derived from two optimal triplet screening scenarios. However, it yielded higher signal noise ratio (SNR) over densely vegetated (the vegetation optical depth (VOD) $\ge 0.7$ ) areas, and tropical, cold, and polar climates. The RF-based gap filling notably improved the merged data spatial coverage and showed minimal impact on data merging quality. The gap-filled merged dataset exhibited significantly higher temporal correlations with ground observations, compared with the official blended products from the European Space Agency (ESA) Climate Change Initiative (CCI) and the National Oceanic and Atmospheric Administration (NOAA) Soil Moisture Products System (SMOPS). The proposed MC-based merging framework could effectively improve the accuracy of SSM information without requiring complex parameter tuning, massive auxiliary data, and high computational costs, providing valuable insights into multisource SSM data merging.