There are numerous locations around the world for which nearshore, shallow bathymetry is unavailable. Due to the expense and logistical challenges associated with collecting sonar or other forms of in situ bathymetric data in remote coastal locations, remote sensing techniques for bathymetric mapping are of increasing interest. Previous work has shown the potential for mapping nearshore bathymetry from NASA's ICESat-2 Advanced Topographic Laser Altimeter System (ATLAS), a green (532 nm), single photon, satellite-based lidar. The aim of this study is to develop and test an efficient, semi-automated method for bathymetric mapping of remote sites through fusion of ICESat-2 ATLAS and Landsat 8 Operational Land Imager (OLI) data. In our approach, refraction-corrected ICESat-2 bathymetry serves to provide reference depths for a satellite-derived bathymetry (SDB) approach using multiple bands of Landsat 8 OLI imagery. The SDB approach is based on the well-known Stumpf algorithm and involves first creating a relative bathymetry raster, which is computed as the ratio of the logarithms of two image bands, typically blue and green. The refraction-corrected ICESat-2 depth data is then linearly regressed on the relative bathymetry, providing coefficients that can be used to convert the relative bathymetry raster to absolute bathymetry, with depths referenced to the same vertical datum as the ICESat-2 bathymetry. Key steps in the process, which have been automated or semi-automated to create an efficient workflow, include: 1) data extraction for the area of interest, 2) coefficient determination for the SDB algorithm, and 3) creation of the final georeferenced depth raster. An additional area of ongoing refinement in our workflow involves automating the classification of the ICESat-2 lidar points into water surface and bathymetric bottom classes. Possible approaches include using a modified random sample consensus (RANSAC) algorithm and a statistical approach based on elevation histograms. We present the results of testing these methods in three locations: Lake Tahoe, California (~11 km2), Western Australia (~1,900 km2), and Tarawa in the Micronesian country of Kiribati (~430 km2). These sites were chosen for their geographic diversity and range of coastal water types. The island of Tarawa provides an example of how the workflow can produce a bathymetric DEM for an extremely remote location. Additionally, Lake Tahoe was chosen in order to compare against recent airborne topobathymetric lidar data. The final bathymetric data products are being disseminated via a web GIS. We conclude with recommendations for ongoing work, including total propagated uncertainty (TPU) modeling for the bathymetric data products, analysis of benthic terrain models from the bathymetric data products, selection of optimal image bands for SDB, and further automation of the end-to-end workflow.
