Making sense of hyperspectral, remotely-sensed data for habitat mapping in Ningaloo Marine Park, Western Australia
Kobryn, H.T., Wouters, K., Pinnel, N., Beckley, L.E., Harvey, M.J. and Heege, T. (2009) Making sense of hyperspectral, remotely-sensed data for habitat mapping in Ningaloo Marine Park, Western Australia. In: AMSA2009 46th Annual Conference for the Australian Marine Sciences Conference, 5 - 9 July, Adelaide, Australia.
Globally, hyperspectral surveys are emerging as a particularly useful technology for mapping connected benthic habitats over large areas of optically clear coastal water. Airborne hyperspectral data covering Ningaloo Marine Park are currently being used to map benthic habitats and develop a reliable and repeatable procedure for this operation. The HyMap data, at wavelengths from visible to near infrared in 126 spectral bands, covered 3 400 km2 at 3.5 m spatial resolution over the terrestrial coastal strip and out to 20m depth over lagoon and reef areas. The data were corrected for atmospheric, air-water interface and water column effects using the physics-based Modular Inversion and Processing System. This approach allowed for quantitative and automated steps, as well as the removal of subjectivity from the classification process. Fieldwork was carried out to support the interpretation, classification and validation of the bottom reflectance data. Spectral reflectance of corals, macro-algae and sediment from several habitats were measured underwater with an Ocean Optics 2000 spectroradiometer and used for development of algorithms for automated image classification. Based on linear discriminant analysis, the in situ spectra of six benthic groups (branching Acropora, digitate Acropora, tabulate Acropora, massive corals (e.g. Porites), submassive corals (e.g. Pocillopora) and macro-algae) could be classified to 90 % accuracy using as few as six optimally-positioned bands in the visible wavelengths. Hyperspectral image analysis of the Ningaloo Marine Park has confirmed that at least 16 major, cover-forming, benthic habitat categories are spectrally separable. Outputs of image analysis include percent cover and probability of these various habitat components to a depth of 15 m. Our results show that hyperspectral remote sensing techniques offer an efficient and cost-effective approach to mapping and monitoring coastal habitats over large, remote and inaccessible areas which are typical of Australia’s vast marine domain.
|Publication Type:||Conference Item|
|Murdoch Affiliation:||School of Environmental Science|
|Item Control Page|