A novel bathymetry signal photon extraction algorithm for photon-counting LiDAR based on adaptive elliptical neighborhood

Ice Cloud and Land Elevation Satellite-2 (ICESat-2) uses a 532-nm laser with a strong seawater penetration ability, and can provide reliable water depth information for sea areas that in situ bathymetric data are lacking. However, because of the complexity of underwater terrain, background noise, optical signal attenuation in water, and other factors, there are still some obstacles to accurately extract bathymetry photons using ICESat-2 data This paper proposes an novel bathymetry signal extraction algorithm based on an adaptive elliptical neighborhood window. The direction-adaptive elliptical neighborhood strategy ensures the directionality and continuity of signal extraction. The size-adaptive neighborhood strategy satisfies the demand for extraction of both shallow, dense signals and deep, sparse signals. The kernel density estimation (KDE) based strategy for sea surface elevation extraction and above-water signal separation makes it possible to determine the sea surface elevation accurately while correctly separating the above-water photons. Four study areas with different environmental conditions in the Xisha Islands, the Bahamas, and the Gulf of Aden were selected, and the key results are as follows. (1) For a water-land interface area with dense noise, the proposed algorithm can effectively reduce the impact of land and accurately eliminate noise photons in shallow water. (2) For an extremely sparse signal in a deep-water area, which is difficult to process using classical algorithms, the proposed algorithm can accurately extract signal photons in water approximately 50 m deep while effectively extracting the dense signal photons in shallow water. (3) In the extraction results obtained using this algorithm, the signal strip has better continuity and smoothness than the classical algorithm and therefore can portray the seafloor terrain more completely. As the in situ water depth confirms, the overall mean absolute error (MAE) and mean relative error (MRE) of the water depth information extracted by the proposed algorithm are 0.79 m and 7.54%, respectively, which indicate high reliability.