Global Seamless Retrospective SMAP Soil Moisture (2010–2015)

Continuous, accurate, and long-term global surface soil moisture observations are key sources in understanding the Earth system science. As the most advanced L-band passive microwave soil moisture monitoring missions, both SMOS and SMAP offer comparable sensing depths and observational sensitivities, yet differ in temporal coverage and instrument configurations. To bridge this temporal gap and align SMAP-derived soil moisture with the longer SMOS records, this study proposes a spatiotemporal extension framework to reconstruct SMAP-like soil moisture back to 2010 through the integration of multi-source passive microwave data. The framework begins by reconstructing missing SMOS brightness temperature (Tb) using an empirical orthogonal function (EOF)-based method. Subsequently, structure-flexible Long Short-Term Memory (sf-LSTM) models are trained across distinct bioclimatic zones using SMOS Tb and SMAP soil moisture data to produce temporally extended estimates. A post-processing fusion procedure is then applied, incorporating a composite satellite product to enhance seasonal consistency and suppress residual inconsistencies. After validation against in-situ networks, the generated dataset achieves a median ubRMSD and correlation coefficients of less than 0.04 m³/m³ and over 0.5 in most regions. The proposed framework performs robustly across diverse environments and enables the generation of long-term, physically consistent soil moisture records for Earth system and climate studies.