Subsurface unmixing with application to underwater classification

Hyperspectral remote sensing is an increasingly important tool for evaluating the complex spatial dynamics associated with estuarine and nearshore benthic habitats. Hyperspectral remote sensing is being utilized to retrieve information about coastal environments, such as coastal optical water properties and constituents, benthic habitat composition, and bathymetry. Essentially, the spectral detail offered by hyperspectral instruments facilitates significant improvements in the capacity to differentiate and classify benthic habitats. A design tradeoff in the design of existing and proposed hyperspectral spaceborne platforms is that high spectral resolution comes with a price of low spatial resolution when compared to existing multispectral spaceborne sensors. The expectation is that the high spectral resolution will compensate for the reduction in spatial resolution by providing information to retrieve some of the lost spatial detail as well as other pieces of information not possible to retrieve using multispectral sensors. This paper reviews different approaches to unmixing of hyperspectral imagery over benthic habitats. Two specific methods that combine water optical properties retrieval with linear unmixing are then described and compared with a standard approach to linear unmixing over land as applied to benthic habitat unmixing. Results show that water column correction is necessary for accurate mapping and that, by removing the water column, we obtain significant improvement in retrieval of bottom fractional coverage for algae, sand and reef endmembers.