A team from Wuhan University’s School of Geodesy and Geomatics and the Chinese Antarctic Center of Surveying and Mapping has built a dual-branch attention-fusion Transformer model that integrates data from two GNSS-R satellite missions, Tianmu-1 (TM-1) and Fengyun-3 (FY-3), gridded to a 36-kilometer EASE-Grid 2.0, to sharpen global soil moisture retrieval.
The model folds in supporting data from NASA’s SMAP mission for surface roughness and temperature, MODIS vegetation index data, GTOPO30 elevation data, and SoilGrids clay and silt content, to correct for the many factors that otherwise confound satellite soil-moisture readings.
The integrated dataset achieved 79.7% average global monthly temporal coverage. Validated against SMAP, it reached a correlation of 0.88 and a root-mean-square error of 0.053 cubic meters per cubic meter; against the International Soil Moisture Network, correlation reached 0.67 with an unbiased RMSE of 0.041; and triple-collocation analysis against independent references yielded a correlation of 0.75. The fused product performed best over arid and sparsely vegetated regions, where single-mission satellite retrievals typically struggle most.
The study’s authors note that combining multiple GNSS-R missions “is not simply a matter of collecting more satellite tracks, but of learning how different missions sense the land surface.”
Journal: Satellite Navigation
DOI: 10.1186/s43020-026-00205-z
Article Title: Attention-guided multi-mission GNSS-R integration for enhanced global soil moisture retrieval
Publication Date: 8-Jul-2026
Funding: National Natural Science Foundation of China (42574033, 42425401); Funds for Creative Research Groups of Hubei Province (2025AFA022)
Source: EurekAlert