Coral reef habitat discrimination using multivariate spectral analysis and satellite remote sensing

Abstract

Coral reefs are important both biologically and economically, but are under increasing pressure from pollution and human-induced disturbance. Coral reefs are often remote, relatively large and shallow. Thus, optical remote sensing techniques provide an efficient and cost-effective approach to mapping and monitoring the condition of reefs. We tested the hypothesis that in-situ spectral signatures, together with water column attenuation measurements, satellite imagery and Geographical Information System, can effectively differentiate reef substrates. In situ upwelling radiances were collected for coarse descriptive habitat types near Lee Stocking Island, Bahamas between 17 July and 27 July, 1998. Down-looking spectral scans from 300-750 nm were taken 1 m above the substrate and 1 m below the water surface, using a LiCor-1800 underwater spectroradiometer. Principal component analysis of upwelling radiances indicated that wavelengths from 515-580 nm are most useful in distinguishing between substrates. Landsat Thematic Mapper (TM) multi-spectral imagery was geo-registered and an ISODATA unsupervised classification was used to delineate reef features and associated communities. Using this method, we achieved an overall accuracy of 74% for coarse descriptive resolution mapping of seven substrate types (i.e. ocean, deep sand, shallow sand, deep coral, shallow coral, seagrass, and mixed sand and seagrass) within an area of 590 km2 near Lee Stocking Island.