Nearshore Bathymetry of Atolls in the South China Sea Estimated by Satellite Multispectral Data

In the field of satellite remote sensing, the exponential attenuation of visible light in clear and shallow water has been demonstrated to estimate water depth. Green and blue bands in satellite multispectral data which have better water penetration are especially appropriate to calculate water depth beyond 10 m. The method based on Beer's Law was introduced to deal with the mixing substrate issue, such as the ratio model. In this study, we examine the performance of bathymetry retrieval by using Sentinel-2A/B, Landsat-8 and WorldView-2 with different ground resolution. The study area is at Pratas (Dongsha) islands in the South China Sea, which is composed of various types of coral atolls and reef flats. Our result shows that in the depth range between 0 and -25 m, the relative error (root-mean-square error divided by the depth) is 7-12% with >5 Sentinel-2 imageries, and 10-20% with >40 Landsat-8 imageries as compared against airborne LiDAR survey. The accuracy can be further improved to 7.2% while taking an average of Sentinel-2 estimates, or 13% for that of Landsat-8 estimates.