Multispectral derivation of bathymetry in Singapore's shallow, turbid waters

In very shallow waters, active sensing determinations of bathymetry are often expensive and unwieldy. Sea depth estimation using passive remote-sensing methods is an attractive alternative, especially using cheap multispectral imagery with high spatial resolution. Three models for the determination of bathymetry from multispectral imagery were utilized with new eight-band images from DigitalGlobe's Worldview-2 satellite platform. All three were trained with electronic navigational chart data and evaluated for accuracy in Singapore's turbid shallow coastal waters. These waters are characterized by high turbidity, suspended sediment, and vehicle traffic. Of the three models, a linear band algorithm performed best, with a root-mean-square error (RMSE) of 0.48 m. A look-up table classification provided a precision of 0.64 m, but was limited by a training set that did not fully represent variance in water column and benthic properties. Possibly owing to the domination of particle backscatter over pigment absorption in these turbid waters, a linear ratio algorithm did not perform as well as the linear band algorithm, achieving an RMSE of only 0.56 m. Analysis found that the usual relationship between ratios of low-absorption to high-absorption bands and depth does not hold as well for these waters, likely due to backscatter dominating leaving-water signals, masking relative absorption effects. High turbidity, with a Secchi disk depth of 1.9 m, limited analysis to shallow reefs and coastline and likely impacted the sensitivity of the bathymetric algorithms. A larger validation data set containing water quality and benthic data is required for further investigation to determine specific sources of error.