A seamless global daily soil moisture dataset (2010–2015) harmonized from SMOS observations and SMAP-era assimilation modeling

The consecutive Soil Moisture and Ocean Salinity (SMOS) and Soil Moisture Active Passive (SMAP) satellite missions have enhanced global L-band soil moisture monitoring. Specifically, SMAP Level-4 (SPL4) provides Bias-corrected, spatiotemporally continuous estimates via brightness temperature (Tb) assimilation. However, the temporal gaps between SMOS and SMAP and the limited duration of SPL4 constrain long-term applications. This study proposes a proxy modeling framework to extend SMAP-quality estimates back to 2010, creating a seamless six-year dataset (2010–2015) that bridges pre-SMAP observations with modern products. The framework integrates three components: reconstructing missing SMOS Tb data using the Data-Interpolating Empirical Orthogonal Functions method; training structure-flexible Long Short-Term Memory networks on reconstructed Tb and SPL4 data across six bioclimatic zones; and fusion with Copernicus soil moisture data using a Principal Component Analysis–wavelet approach to enhance consistency and extreme value simulation. Validation against 18 in situ networks showed strong performance (median unbiased root mean square error = 0.04 m³/m³; R > 0.6 in most regions). Compared with benchmark datasets (SGD-SM and NNsm), the Extended Seamless Soil Moisture (ESSM) dataset exhibits improved accuracy, particularly in regions with complex terrain or extreme climates. ESSM provides a robust foundation for hydrological modeling, drought monitoring, and climate variability studies during the critical pre-SMAP period.