Xiaoli Deng (School of Engineering, University of Newcastle, Newcastle, Australia., Australia); Ole Baltazar Andersen (DTU Space, Technical University of Denmark, Lyngby, Denmark., Denmark); Richard Coleman (Institute for Marine and Antarctic Studies (IMAS), University of Tasmania, Tasmania, Australia, Australia)

Accurate bathymetric mapping in coastal zones is essential for marine navigation, environmental monitoring, and geophysical modeling. However, the complexity of shallow-water environments—combined with sparse in situ measurements and the limitations of traditional inversion methods—poses a significant challenge, particularly along Australia’s extensive and geologically diverse coastlines.

This study presents a deep learning approach tailored to improve coastal bathymetry estimation in Australian waters using geophysical data derived from the Surface Water and Ocean Topography (SWOT) mission. The model integrates shipborne bathymetric measurements with a suite of SWOT-based geophysical features, including marine gravity anomalies, vertical gravity gradients (VGG), and geoid slopes. These features are processed and fused within a deep neural network (DNN) to capture both large-scale and fine-scale variations in seafloor topography.

By focusing on the Australian coastal zone, the model addresses regional complexities such as continental shelf breaks, reef structures, and sediment-heavy margins. Validation against independent shipborne datasets reveals substantial improvements in prediction accuracy, especially in shallow waters, where global models often underperform. The results demonstrate the importance of integrating multiple geophysical features and tailoring machine learning frameworks to specific regional settings.

This work highlights the potential of SWOT altimetry, when coupled with advanced deep learning techniques, to significantly enhance coastal bathymetric mapping in data-sparse regions. The approach provides a scalable framework that can be extended to other coastal zones globally.