Enhancing underwater topography estimation by integrating backpropagation networks with multivariate geophysical data

This study introduces Backpropagation networks to seafloor topography inversion and proposes an approach for depth inversion based on the integration of gravity anomalies, vertical gravity gradient anomalies, and vertical deflection, using the multivariate data fusion Three-Channel BP neural network (MFT-BP). This work uses feature data from gravity anomalies, vertical gravity gradients, and vertical deflections to construct a reliable depth prediction model. Additionally, sounding data serves as validation data. Experimental findings show that the MFT-BP network performs better than traditional methods, such as gravity-geology models (GGM) and alternative models, regarding accuracy. Validation against sounding data reveals compelling results, 89.72 % of depth estimates differ by less than 100 m, and 97.06 % differ by less than 200 m. The average relative error between the multibeam data and the BP network's depth predictions is 5.474 %, showing a 1.4 % improvement over GGM depth estimations. These outcomes highlight the effectiveness of integrating BP networks with multi-source geophysical data for accurate underwater topography inversion.