A Progressive Denoising Methodology for ICESat-2 Bathymetric Photons Based on Underwater Topographic Trend Extraction

The Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) employs 532 nm laser active detection for surface elevation, providing reliable data support for nearshore bathymetry. Owing to its high sensitivity, the raw data contain substantial noise, and effective denoising is essential for extracting underwater signal photons. To acquire high-precision underwater signal photons, this paper introduces a progressive denoising-based underwater topographic trend extraction (PUTTE) method. First, regional photon separation is performed with ICESat-2 data to extract water-surface and underwater photons. Afterward, underwater noise is classified into three categories, namely, isolated noise, low-density noise, and proximity signal noise photons, and progressive processing is applied to these three types of noise to extract the underwater topographic trend. Finally, refraction correction is carried out for the underwater topographic signal photons. The results show that the coefficient of determination (R2) is as high as 0.99, the root mean square error (RMSE) ranges from 0.07 to 0.61 m, and the mean absolute error (MAE) ranges from 0.06 to 0.45 m. The precision exceeds 85%, the recall is above 80%, the F1 score exceeds 88%, and the accuracy at least 83%. Compared with the density-based spatial clustering of applications with noise (DBSCAN), ordering points to identify the clustering structure (OPTICS), adaptive elevation difference thresholding algorithm (AEDTA) and Hui's Progressive Noise Photon Removal Method, the method proposed in this paper performs better in terms of accuracy and completeness, especially in areas with abrupt topographic changes and regions with low-density signal photons. This paper introduces a reliable approach for high-precision water depth measurement.