dc.description.abstract | Bathymetry estimation plays an essential role in navigation, area management, and marine ecological research. With the advancement of satellite technology, passive satellite derived bathymetry (SDB) gradually replaced active bathymetry methods such as shipborne echo sounders or airborne LiDAR. Large-scale surveys by satellite image can effectively improve the disadvantages of active telemetry including high cost, time-consuming, and susceptible to climate. Especially for traditional shipborne echo sounders, because of navigational safety, shallow water areas are often not accessible. Most of the bathymetry estimation performed by optical satellites were based on two kinds of methods, physical method and empirical method, based on Beer-Lambert Law, such as ratio algorithm, look-up table (LUT) and experience learning method. However, these methods do not consider the mixture of the bottom material on the sea bottom, but treat each pixel as a pure pixel containing only a single endmember.
Since the spatial resolutions of satellite images are usually several meters, most pixels contain more than one materials and can be considered as mixed pixels. This study combines the technology of spectral unmixing and Beer-Lambert Law to explore the relationship of radiance spectra and water depth from geometry point of view. The linear mixture model has been widely used for remote sensing and it assumes that the spectrum of each pixel is a linear combination of pure endmembers, so it is located inside a convex polyhedron with vertices of endmembers in multi-dimension. Based on Beer-Lambert Law, spectral extinction is exponentially related to the water depth, and the extinction coefficient of the spectral bands are different and independent, the polyhedron will rotate and shrink with the increasing of water depth. This research proposes a new bathymetry estimation algorithm by combining spectral unmixing and Beer-Lambert law to not only improve the accuracy of coastal bathymetric mapping, but also estimate the spectral signatures of endmembers and their abundances on the seabed. In addition, the spectral unmixing technology enables this algorithm to estimate the depth of water in satellite images with lower spatial resolution. This study observes the feasibility of this theory under different conditions through four simulations. The experimental results showed that this algorithm requires enough pixels with purity level greater than 0.9 and evenly distributed close to water surface to reach good estimation in water depth. | en_US |