ABSTRACT        

Owing to the high spatial heterogeneity of substrate types and terrain, the present satellite-derived bathymetry (SDB) methods have low accuracy in deriving large-scale bathymetry in coral reef habitats. Taking 11 coral reefs of Xisha Islands (ocean area of 607 km2) in the South China Sea as the study area, a parametric multimodel combination approach based on geomorphic segmentation (PMCGS) for obtaining bathymetry was constructed by combining the Ice, Cloud andLandElevation Satellite-2 (ICESat-2) datawith Gaofen-1 (GF-1) medium- andWorldview-2/3 (WV-2/3) high-resolution multispectral images. In this approach, five parametric SDB models were trained in each geomorphic zone by combining ICESat-2 and multispectral satellite images. Then, the optimal SDB models of each geomorphic zone were combined and extrapolated to other coral reefs in the same geomorphic zone. Results showed that the multiple ratios model was optimal for the reef flat, shallow lagoon, and patch reef zones. The binomial model was optimal for the reef slope and deep lagoon zones. Validated by the in situ bathymetric data and ICESat-2 data, the bathymetry inverted using the PMCGS had an RMSE of 0.91 m in GF-1 image and 0.70–0.88minWV-2/3 imageswhen extrapolated to other reefs, which is significantly more accurate than active–passive one entire model methods with the same resolution. Our method performed better at 0–10mand 15–25mdepth than the results obtained from previous studies, especially in the shallowwater areas of the reef flat and shallow lagoon. The proposed PMCGS can efficiently improve the bathymetry inversion accuracy of medium- and high-resolution satellite images and it has great potential applications in deriving large-scale bathymetry, especially in Indo-Pacific coral reef habitats.