Statistical Combination of Spatial Interpolation and Multispectral Remote Sensing for Shallow Water Bathymetry

There is often a need for making a high-resolution or a complete bathymetric map based on sparse point measurements of water depth. Well-known feasible methods for this problem include spatial interpolation and passive remote sensing using readily available multispectral imagery, whose accuracies depend strongly on geometric and optical conditions, respectively. For a more accurate and robust water-depth mapping, in this letter, the two methods are combined into a new method in a statistically reasonable and beneficial manner. The new method is based on a semiparametric regression model that consists of a parametric imagery-based term and a nonparametric spatial interpolation term that complement one another. An accuracy comparison in a test site shows that the new method is more accurate than either of the existing methods when sufficient training data are available and far more accurate than the spatial interpolation method when the training data are scarce.