Synergistic Fusion of ICESat-2 Lidar and Sentinel-2 Data to Leverage Potential Mapping of Bathymetry in Remote Islands Using SVR

In the study presented in this paper, bathymetric maps in shallow waters of Neil and Havelock Islands of Andaman were produced based on Support Vector Regression (SVR) machine learning (ML) technique with two popular empirical approaches (log-linear and log-ratio models) with only satellite remotely sensed data. A new adaptive Density-based Spatial Clustering of Applications with Noise (DBSCAN) algorithm has been proposed to automatically extract the seafloor photons from the Ice, Cloud and Land Elevation Satellite (ICESat-2) Level 2A data after applying tide and refraction corrections. Later, these depth points were used in place of the in situ bathymetric points to train (with 80% of data) the SVR model. With trained models and Sentinel-2 multispectral images of 2019, the bathymetric maps were produced. The performance of the models was evaluated using remaining independent (20%) data points from ICESat-2. The results indicate that the overall errors during the test phase for the range of depth (0–25 m) are 0.60 (0.47 m) and 0.41 m (0.32 m) based on Log-Ratio Model—LRM—(Log Linear Model—LLM) methods in Neil and Havelock Islands, respectively. The LLM-based SVR ML produced relatively accurate bathymetry as compared to the LRM approach over the entire depth range. Therefore, this approach can be extended to produce large-scale bathymetry maps in shallow waters of other islands and reefs along Indian coastal waters.