Coastal water bathymetry for critical zone management using regression tree models from Gaofen-6 imagery

Coastal water depth information is fundamental to coast management and coastal critical zone development. The traditional bathymetric sounding method depends on radar or surveying vessels, which are expensive and timeconsuming. Thus, it is of high importance to develop a rapidly updating water depth detection method. In order to bridge this gap, we explored the Chinese Gaofen-6 wide field of view (GF-6 WFV) visible-near infrared satellite imagery for large-scale accurate coastal bathymetric mapping and to understand the vertical water column environment with regression tree models. The predictors, including the Blue band/Violet band (BV), Green band/Violet band (GV), Yellow band/Violet band (YV), Green band (G), Yellow band (Y), and red-edge1 band (Re1), were derived via statistical analysis, as well as spectroscopy knowledge and experience. Compared with the conventional bathymetric method, such as single band algorithm (SBA), band ratio algorithm and support vector regression (SVR), the two regression trees used in this study yielded better accuracy. The R2, MAE and RMSE of the classification and regression tree (CART), the Cubist tree model were 0.74, 3.78 m and 5.35 m vs 0.78, 3.56 m and 4.88 m. Besides the improved accuracies, these tree-based models can effectively reveal water depth-associated environment types with the hierarchical relationships present in the visible spectral characteristics, which supports spatial zoning strategies for coastal critical zone management based on the Gaofen-6 imagery.