Learning Enhanced Structural Representations with Block-Based Uncertainties for Ocean Floor Mapping

High-resolution bathymetric data is crucial for accurate ocean modeling and coastal hazard prediction, yet current global datasets remain too coarse for precise numerical simulations. While recent deep learning advances have improved earth observation data resolution, existing methods struggle with the unique challenges of generating detailed ocean floor maps, particularly in maintaining physical structure consistency and quantifying uncertainties. This paper introduces a novel uncertainty-aware mechanism with block-based conformal prediction that effectively captures local bathymetric complexities through spatial blocks. The integration of this uncertainty quantification framework with a Vector Quantized Variational Autoencoder (VQ-VAE) enables spatially adaptive confidence estimates while preserving topographical features through discrete latent representations. The block-based design adapts uncertainty estimates to local bathymetric complexity, with smaller uncertainty widths in well-characterized regions and appropriately larger bounds in areas of complex seafloor structures. Experimental results across diverse ocean regions demonstrate significant improvements in both reconstruction quality and uncertainty estimation reliability compared to traditional methods. By maintaining structural integrity while providing spatially adaptive uncertainty estimates, this framework enhances the reliability of bathymetric reconstructions — paving the way for more robust climate modeling and coastal hazard assessment.