New Reference Bathymetric Point Cloud Datasets Derived From ICESat-2 Observations: A Case in the Caribbean Sea
Congshuang Xie, Peng Chen, Cédric Jamet, Delu Pan
Abstract
Satellite-derived bathymetry (SDB) methods have been traditionally hindered by the need for in situ reference bathymetric points. However, the light detection and ranging (LiDAR) instruments on the new ICESat-2 satellite have revolutionized SDB by providing high-precision reference bathymetric point cloud datasets (RBPCDs) in shallow water. While the density-based spatial clustering of applications with noise (DBSCAN) has been effective in photon cloud processing, it has been challenging to determine key parameters due to the complexity of terrain changes. Furthermore, ICESat-2 is unable to measure deep water depths greater than 50 m, which would be less efficient if it has to process the entire track data. To overcome these challenges, we have developed an adaptive ellipse denoising algorithm with adjustable key parameters and a shallow-water feature photon (SWFP) extraction method. These innovative techniques were applied to the Caribbean Sea and the South China Sea, resulting in impressive datasets consisting of 848 395 and 438 643 RBPCDs, respectively. The mean absolute error (MAE) of RBPCDs was found to be within 0.6 m, and the RBPCDs were consistent with in situ data. By combining RBPCDs with Sentinel-2 data using a neural network (NN)-based SDB method, we have created detailed bathymetry maps over 15 islands in the Caribbean Sea. Our adaptive method has great potential for large-scale nearshore RBPCD construction, and these RBPCDs will undoubtedly enhance SDB implementations in the future.