Gravity Spectrum: A Multi-Band Spectral for High-Resolution Seafloor Topography Recovery

Development of satellite altimetry to map high-resolution and high-precision seafloor topography is of great significance for the rapid and low-cost acquisition. Existing methods of satellite altimetry gravity inversion of seafloor topography have the disadvantages of inability to recover high-frequency seafloor topography information, insufficient feature calculations and so on. To overcome these constraints, gravity spectrum (GS) is proposed in the study based on the wavelength-dependent response relationship between gravity and seafloor topography. Gravity spectrum is a three-dimensional spectral cubic established by multiple discrete wavelength bands through spectral decomposition, where each band encapsulates wavelength-specific gravity energy that corresponds to distinct topography-gravity response mechanisms. The GS multi-band architecture enhances neural network-based seafloor topography inversion with enriched gravity features while preserving geophysical principles of topography-gravity interactions at discrete wavelengths. We applied neural network to invert seafloor topography in the Gulf of Guinea using Scripps Institution of Oceanography (SIO) datasets, including gravity anomaly (GA) from conventional nadir altimeters and the wide-swath SWOT altimeter. The experimental results indicated that for the same gravity, the accuracy of the seafloor topography is improved by about 54% using gravity spectrum with RMSE decreased 54m from 153.54m to 69.83m, which compares with the S&S method and the Topo27.1 seafloor topography model released in 2025 getting a 20% reduction in the RMSE by around 20m. Gravity spectrum has improved accuracy and resolution in areas without or with limited ship-survey depth data and areas with complex topography, effectively reconstructing small-scale seabed features. With the increase of the number of gravity spectrum bands, the seafloor topography recovered by gravity spectrum is clearer. Gravity spectrum, leveraging its advantages in enhanced data features, exploiting the non-linear performance of neural networks, and enhancement of short-wave information, holds significant potential to become a novel paradigm for high-resolution seafloor topography inversion driven by satellites like SWOT.