SCBMLD: a photon signal extraction method for ICESat-2 satellite-derived bathymetry

The integration of satellite remote sensing imagery with ICESat-2 photons offers an effective, economical, and large-scale approach for mapping shallow seabed topography. However, ICESat-2 photon data are vulnerable to noise from sea surface fluctuations, water quality, and instrument errors, which can compromise the accuracy of bathymetric products. To address this challenge, we propose a scale compression-based machine learning denoising (SCBMLD) method for accurate seafloor photon signal extraction from ICESat-2 data, enhancing bathymetry inversion in multispectral imagery. Results show that SCBMLD achieves an average accuracy improvement of 21%, and provides high-quality bathymetric data across diverse conditions while reducing the need for complex parameter tuning and lowering computational costs. Using this approach, reliable bathymetric products were generated for two study regions, with RMSE of 0.51 m and 1.09 m, meeting C-level zone of confidence requirements.