Denoising of ICESat-2 Photon-Counting Data Based on the Improved DENCLUE Algorithm

The Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) light detection and ranging data prove significant potential for oceanographic applications including shallow water bathymetry and coral reef monitoring. However, their utilization in precise oceanographic investigations is constrained by substantial noise contamination. Traditional spatial clustering denoising algorithms exhibit high parameter sensitivity, where a suboptimal parameter selection frequently leads to inadequate noise suppression. To overcome these limitations, this study develops an improved DENCLUE (density-based clustering) algorithm with improved noise mitigation capabilities. The methodology initially determines the sea surface elevation through characteristic analysis of surface point clouds, followed by systematic data classification. Subsequent processing involves the segmentation of residual noise points using along-track distance and elevation coordinates. The improved algorithm implements a density-optimized DENCLUE clustering through regional density partitioning while incorporating a simulated annealing mechanism to refine the density core search strategy, thereby optimizing denoising performance. Finally, to further refine the data after initial denoising, a refined denoising approach based on elevation distribution pattern analysis is conducted to eliminate residual discrete noise. Post-processing validation involves the refraction correction of seafloor point cloud data and comparative analysis with in situ bathymetric measurements. Experimental results demonstrate the exceptional accuracy of the denoised ICESat-2 data against field measurements: R2=0.91-0.99, MAE=0.22-0.51m, RMSE=0.26-0.73m, and MSE=0.07-0.53m2. These findings confirm the improved DENCLUE algorithm's effectiveness in ICESat-2 data denoising and establish a methodological framework for shallow water bathymetric exploration using remote sensing technologies.