Abstract
Timely mapping of underwater topography over turbid coastal waters is very important to navigation. Such a task is ideally accomplished through moderate- resolution imaging spectroradiometer (MODIS) satellite data that have a temporal resolution measured in hours. In this article we propose a simple regression method for retrieving bathymetry from MODIS bands. It only requires concurrently collected total suspended solids and water depth samples at limited spots, without considering the downward attenuation coefficient, surface reflectance or bottom reflectance. Regression analysis of the observed spot depth (D) against individual bands and their transformation enables an empirical model to be established. The model in which band 3 (M 3) is the exploratory variable is the most accurate, with an r value (correlation coefficient) of only 0.654. Correction of this model by the concentration level of suspended solids using bands 2 (M 2) and 5 (M 5) improves the prediction accuracy from 3.26 to 1.52 m, or from 39% to 24%. The best model takes the form of D = –7.833 + 0.0326M 3 / (M 2 M 5) (r = 0.815, n = 3318). Application of this model to the MODIS imagery led to the generation of a bathymetric map over the 15 000 km2 study area. Assessed against four profiles, the retrieved bathymetry has a mean absolute accuracy of around 2 m or a relative inaccuracy of 10% to 18%. The remotely sensed bathymetry contains many minor relief features absent from its in situ-surveyed counterpart. It is concluded that this proposed simple method can produce reasonably accurate results without the need to consider atmospheric effects or bottom reflectance over the range of 5–20 m. However, it may not work so well in clear oceanic Case I waters.
Acknowledgements
Funding for this research was provided by the National Science Foundation of China (Grant no. 40606044). Two anonymous reviewers provided helpful comments that further improved the quality of the manuscript.