Objectives Bathymetric datasets are often constrained by limited observation coverage and high acquisition costs, resulting in insufficient spatial resolution, loss of fine-scale terrain features, and discontinuities in large-area reconstructions. To address these issues, a high-precision super-resolution reconstruction framework is developed to enhance both geometric fidelity and spatial consistency of seafloor topography.

Methods A single-step diffusion-based super-resolution model incorporating terrain-aware constraints is constructed. The model is derived through distillation from an accelerated diffusion framework, enabling efficient one-step inference while preserving high reconstruction quality. To better capture complex geomorphological structures, slope and curvature information are introduced as geometric priors and embedded into the loss function design. In addition, a frequency-domain constraint based on discrete cosine transform is incorporated to enhance high-frequency detail recovery. To mitigate discontinuities caused by patch-wise reconstruction in large-area applications, a distance transform-based weighted fusion method is proposed. By assigning spatially adaptive weights according to the distance to patch boundaries, smooth transitions and consistency across overlapping regions are achieved.

Results Experiments are conducted using bathymetric datasets, with the 3″ global digital elevation model as the baseline and multibeam sounding data as ground truth. Quantitative evaluations over multiple representative regions demonstrate that the proposed method significantly improves reconstruction accuracy and structural consistency. Specifically, the average root mean square error and mean absolute error are reduced from 37.23 m to 24.05 m and from 22.55 m to 14.80 m,corresponding to reductions of 35.40% and 34.37%,respectively. Meanwhile, the ave‑rage peak signal-to-noise ratio increases by 4.25 dB. Visual comparisons further indicate that the proposed method effectively preserves steep slopes and fine-scale terrain features while suppressing artifacts and over-smoothing effects. The proposed fusion strategy also substantially alleviates stitching artifacts and improves spatial continuity across reconstructed regions.

Conclusions By integrating terrain geometry and frequency-domain constraints into a single-step diffusion framework, the proposed method achieves an effective balance between reconstruction accuracy and spatial consistency. The method provides an effective and scalable solution for bathymetric super-resolution reconstruction, with strong potential for applications in marine science, resource exploration, and ocean engineering.