Item

Abstract

The acquisition of accurate bathymetry is a challenging task in Marine Geodesy, especially for the area without sonar sounding data, the ocean gravity field is needed to realize the seabed topography inversion process. The gravity-geological method (GGM) is one of the most classic methods for seabed topography inversion. According to the approximate linear relationship between the seabed topography and the short-wave gravity anomaly, the GGM method constructs the regional bathymetry model. However, the correlation between seabed topography and gravity anomaly are non-lineardue to factors such as the geology of the seafloor. To address this issue, based on the short-wave gravity anomaly obtained by the GGM method, residuals neural network (ResNet) is used by introducing a variety of prior geophysical attribute data information, such as vertical gradient, magnetic anomaly, and sediment thickness. The non-linear relationship between gravity anomaly and seabed topography is then obtained. In the Puerto Rico test area, the accuracy of the seabed topography over the inspection points is improved by ~10m compared with results using the GGM method. The seabed topography inversion combined with GGM and neural network will provide a new idea for bathymetric survey based on satellite altimetry, which has high feasibility and important application value.