Spaceborne GNSS-R Soil Moisture Retrieval over Expansive Soils Using an Attention-Enhanced Spatio-Temporal Graph Convolution Network

Expansive soils are rich in hydrophilic clay minerals, and repeated wetting–drying cycles can induce deformation that threatens infrastructure safety. Therefore, accurate monitoring of soil moisture (SM) dynamics is essential for understanding hydro-mechanical processes and assessing related geohazards. In this study, spaceborne Global Navigation Satellite System Reflectometry (GNSS-R) is applied to expansive SM monitoring, and an Attention-Enhanced Spatio-Temporal Graph Convolution Network (ASTGCNet) is proposed for SM retrieval. The Texas coastal region, where Beaumont clay is widely distributed, was selected as the study area. The ASTGCNet-derived SM showed consistency with the Soil Moisture Active Passive (SMAP) reference product, with an overall correlation coefficient of 0.92, an RMSE of 0.035 m3/m3, and a bias of 0.006 m3/m3. Validation against in situ observations showed that ASTGCNet provided more accurate SM estimates than the Cyclone Global Navigation Satellite System (CYGNSS) L3 SM product. Extended triple collocation analysis further indicated that ASTGCNet achieved the lowest standard deviation of 0.020 m3/m3 and the highest signal-to-noise ratio of 7.33. Compared with non-expansive soils, expansive soils exhibited stronger water absorption and moisture retention behavior. By integrating GNSS vertical displacement observations, the retrieved SM revealed a nonlinear SM–deformation response that was mainly observed in shallow expansive soils. Drying-induced SM decreases corresponded to pronounced subsidence, while subsequent wetting led to ground rebound; this behavior was not clearly observed in non-expansive soils. This study demonstrates the potential of GNSS-R for expansive SM monitoring and provides new insights into the coupling between SM dynamics and deformation.