Enhancing the accuracy of spaceborne GNSS-R sea surface altimetry using a feature-enhanced multiple base learners ensemble retracking model

Sea surface altimetry is critical for ocean dynamics research, including global marine gravity field inversion, sea level change monitoring, and seafloor topography detection, all of which are vital for defense applications such as submersible navigation and underwater strategic operations. Global Navigation Satellite System Reflectometry (GNSS-R) enables monitoring sea surface height (SSH), but the current accuracy of spaceborne GNSS-R altimetry remains low. One of the key factors affecting its accuracy is the imprecise acquisition of waveform retracking points. To address this issue, this study proposes a Feature-enhanced Multiple base learners Ensemble Retracking Model (FMERM) that innovatively combines feature enhancement and ensemble learning. It accurately estimates the retracking points of reflected waveforms based on geometric information. The sea surface height retrieved using FMERM aligns well with the Technical University of Denmark (DTU21) model after tidal correction. The root mean square error (RMSE), mean absolute error (MAE), and Pearson correlation coefficient (PCC) reach 4.281 m, 2.480 m, and 0.987, respectively, representing improvements of 59%, 66%, and 76% over traditional retracking methods. Furthermore, FMERM maintains good accuracy under different incoherent integration times (500 ms and 2000 ms), achieving average RMSE, MAE, and PCC of 4.927 m, 2.572 m, and 0.984, respectively, demonstrating the FMERM's robustness. The proposed FMERM provides an effective solution to waveform retracking inaccuracy and offers strong support for high-precision spaceborne GNSS-R sea surface altimetry.