Abstract: Satellite-Derived Bathymetry (SDB) is often affected by noise from cloud cover, surface waves, and nearshore mixed pixels, which limits the ability of single-date imagery to achieve high-precision depth retrieval. To address this issue, this study proposes an SDB framework based on the Standard Deviation Weighted Laplacian Pyramid Transform (SDW-LPT), integrating ICESat-2 photon data with multi-temporal Sentinel-2 imagery. Taking Yongle Atoll in the South China Sea as a case study, three image fusion strategies—SDW-LPT, median-based compositing, and extreme outlier removal—were employed to generate enhanced reflectance products. Subsequently, three representative SDB models (the multi-band model, logarithmic ratio model, and quadratic polynomial ratio model) were applied to assess retrieval accuracy within the 0-20 m depth range. Results indicate that the SDW-LPT fusion imagery achieved RMSEs of 0.45 m, 0.68 m, and 0.52 m across the respective models, consistently outperforming the baseline approaches. Moreover, as the number of fused images increased, inversion accuracy improved steadily and stabilized once more than 20 images were integrated. Even under limited-temporal conditions, SDW-LPT sustained a high level of accuracy, demonstrating strong robustness and noise resistance. Overall, this study validates the effectiveness of SDW-LPT for SDB applications and offers a viable framework for multi-source data-driven shallow water bathymetry retrieval.

Key words: multispectral remote sensing, bathymetry inversion, multi-temporal imagery, standard deviation weighting, Laplacian pyramid transform