Deriving Highly Accurate Shallow Water Bathymetry from Sentinel-2 and ICESat-2 Datasets by a Multitemporal Stacking Method

Abstract

Empirical models have been widely used to retrieve shallow water bathymetry from multispectral/hyperspectral satellite imagery. In traditional studies on deriving the topography and monitoring its temporal changes, a single date satellite image without clouds corresponded to a bathymetric map and multidate images corresponded to multiple bathymetric maps. The satellite image noise caused by various environmental conditions and satellite sensors can inevitably introduce errors or gaps in deriving bathymetric maps. Also, empirical models are limited in some remote areas due to the lack of prior bathymetric points. In this article, using only satellite data, including multitemporal Sentinel-2 images and Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) data, a multitemporal stacking method was developed to derive highly accurate and cloud free shallow water bathymetry with accuracy of approximately 1 m and the depth range exceeding 22 m. The proposed method was tested and validated by an airborne bathymetric lidar. To be specific, our method using multitemporal Sentinel-2 images can achieve a mean root mean square error (RMSE) of 1.08 m (R2 = 0.94) by comparing with in-situ airborne lidar data around Ganquan Island, which is better than the result (R2 = 0.92, RMSE = 1.46 m) derived from single date image based methods.Also, the gaps in a bathymetric map due to clouds or other noise can be avoidable benefitting from the stacking of multiple date satellite images. In the future, this satellite data driven method can be further extended to the globe to produce highly accurate and cloud free bathymetry around clear shallow water benefited from prior ICESat-2 bathymetric data.