SHALLOW WATER BATHYMETRY DERIVED BY MACHINE LEARNING AND MULTI-TEMPORAL SATELLITE IMAGES

Shallow water bathymetry is essential information for coastal science and nautical navigation. In this study, a satellite derived bathymetry (SDB) map, considered suitable for shallow water, was created using random forests (RF) and multi-temporal satellite images from Google Earth Engine. RF performance was assessed with a training dataset varying in size. The root mean square error (RMSE) of the SDB map created by the RF model decreased with an increase in training dataset size. Tide-level correction methods are proposed and satellite-image based correction methods improved accuracy significantly. For our Hateruma case study, the average RMSE for the SDB map created using 25 satellite images was 1.79 m.