From heterogeneous signals to consistent wind fields: a unified framework for sea surface wind speed retrieval with multi-GNSS reflections

Global Navigation Satellite System Reflectometry (GNSS-R) provides a low-cost approach for observing sea surface winds with dense spatial and temporal sampling. The growing availability of multi-system GNSS-R observations further improves sampling capability, but differences among missions, transmitted signals, and receiving platforms still limit consistent multi-source retrieval and fusion. This study develops a sea surface wind speed retrieval and fusion framework using the Cyclone Global Navigation Satellite System (CYGNSS), FengYun-3 (FY-3), and Tianmu-1 (TM) observations. To reduce intra-system heterogeneity, a grouped Random Forest (RF) strategy is adopted. Observations are modeled separately according to the transmitting GNSS constellation and receiving GNSS-R satellite, and the subgroup results are then combined at the system level. The grouped models achieve Root Mean Square Errors (RMSEs) of 1.15, 1.26, and 1.10 m/s against the Cross-Calibrated Multi-Platform (CCMP) product for CYGNSS, FY-3, and Tianmu-1, respectively, and 1.15, 1.38, and 1.18 m/s against the fifth-generation European Centre for Medium-Range Weather Forecasts Atmospheric Reanalysis (ERA5). Independent external validation using National Data Buoy Center (NDBC) observations further supports the reliability of the single-system retrievals. For multi-system fusion, this study develops a reliability-adjusted Best Linear Unbiased Estimator (BLUE) weighting method. Unlike equal-weight fusion, the proposed method estimates source-dependent errors from local residual statistics and adjusts the weights using effective sample support. This design reduces the influence of unreliable local error estimates in sparsely sampled regions. Compared with equal weighting, the reliability-adjusted BLUE fusion reduces gridded RMSE over broad oceanic regions, with local improvements approaching 20%, and improves the spatial continuity of the combined GNSS-R wind field. These results demonstrate the potential of the proposed framework for consistent wind retrieval and robust fusion from heterogeneous multi-system GNSS-R observations.