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

Area of Science:

• Geophysics
• Marine Geology
• Artificial Intelligence

Background:

• Seafloor topography inversion is crucial for marine geological surveys and resource exploration.
• Traditional methods for depth inversion often face limitations in accuracy and data integration.
• Backpropagation (BP) neural networks offer potential for advanced data analysis in geophysics.

Purpose of the Study:

• To introduce and evaluate a multivariate data fusion Three-Channel BP neural network (MFT-BP) for seafloor topography depth inversion.
• To assess the performance of the MFT-BP network by integrating gravity anomalies, vertical gravity gradient anomalies, and vertical deflection data.
• To compare the accuracy of the MFT-BP network with traditional gravity-geology models (GGM) and other alternative models.

Main Methods:

• Development of a multivariate data fusion Three-Channel BP neural network (MFT-BP).
• Integration of gravity anomalies, vertical gravity gradient anomalies, and vertical deflection as input features.
• Utilizing sounding data for validation of the depth prediction model.
• Comparative analysis against gravity-geology models (GGM) and alternative methods.

Main Results:

• The MFT-BP network demonstrated superior accuracy in depth inversion compared to GGM and alternative models.
• Validation against sounding data showed that 89.72% of depth estimates had errors less than 100 m, and 97.06% had errors less than 200 m.
• The MFT-BP network achieved an average relative error of 5.474%, a 1.4% improvement over GGM depth estimations.

Conclusions:

• The integration of BP networks with multi-source geophysical data is effective for accurate underwater topography inversion.
• The MFT-BP network provides a reliable and accurate method for seafloor depth prediction.
• This approach advances the application of artificial intelligence in marine geophysical exploration.