Inter-Research
Marine Ecology Progress Series

ABSTRACT: In principle, a priori knowledge of organism-scale spectral signatures for key ecological end-members is a basic requirement for identifying coral reef benthic communities using hyperspectral remotely-sensed imagery. Spectral signatures of end-members are now relatively well known for predominant reef taxa (coral, algae) and for the background of the living communities (e.g. sediments). What remains unclear is whether the criteria for spectral discrimination between end-members at the millimeter or centimeter scale remain valid when attempting to process images at several meters resolution. In other words, is it possible to scale-up spectral criteria of identification from species/organisms to communities? We address this issue with in situ and ‘compact airborne spectrometer imager’ (CASI) hyperspectral measurements of the tropical marine flora of 2 South Pacific Ocean coral reefs. Targets were the dominant algal species and communities encountered in the shallow (0 to 3 m) barrier and fringing reefs of Moorea Island and the outer margin of the rim of Rangiroa Atoll (French Polynesia). Stepwise wavelength selection and linear discriminant analysis highlighted the key non-redundant wavelengths necessary to achieve good separation between the predefined ecological groups. Comparison of the wavelengths identified from in situ and airborne measurements allowed definition of a subset of common wavelengths that were robust to changes in spatial scale and still provided excellent discrimination and classification accuracy between the ecological groups. These results suggest that continuous spectral signatures acquired in situ at the centimeter scale can be used to select key discrete wavelengths for remote-sensing observations of communities at the meter scale despite the spatial heterogeneity in benthic cover and the resulting spectral mixing.

KEY WORDS: CASI · Remote sensing · Hyperspectral · Atmospheric correction · Macroalgae · Coral reef · Atoll · Moorea · Rangiroa

Full text in pdf format

Published in MEPS Vol. 283 (2004) on November 30
Print ISSN: 0171-8630; Online ISSN: 1616-1599. Copyright © Inter-Research, Oldendorf/Luhe, 2004