Hyperspectral identification of coral reef features

Remote sensing is regarded as an efficient and accurate tool for mapping and monitoring changes in coral reef extent and well being consistently over large geographic areas. However, several operational restrictions limit the accuracy with which coral reefs can be monitored remotely. A primary restriction is that the spectral responses of numerous features in the coral reef environment are optically similar, which has the potential of contributing to misclassification errors. In 1996, spectral reflectance data were collected in Fiji using an in situ underwater cosine receptor and a 10m underwater fibre optic cable, which permitted sampling at depth while scuba diving. In 1997, spectral reflectance in situ measurements of exposed coral reef features with little or no water cover were collected in Indonesia using the same radiometer, but a nonwaterproof remote cosine receptor. These spectral datasets were compared and analysed to test the following hypotheses. First, geographic location does not affect the spectral reflectance characteristics; second, the morphology of reef features does affect the spectral reflectance characteristics; third, bleached coral and healthy coral have distinct spectral reflectance characteristics; and finally, a spectral reflectance index will aid image classification. Results indicate that the spectra measured in Fiji and those measured in Indonesia are statistically similar, so all spectra were merged into one large spectral dataset and principal components analysis was used to determine the most representative spectra. Derivative spectroscopy was then used to conclude that spectral discrimination is indeed possible between 654 and 674nm, between 582 and 686nm and between 506 and 566nm. The proportion of correctly identified spectra using the three-step procedure of first derivatives is 75% with the main source of error resulting from spectral variability of algae reflectance. The results of this feasibility study indicate that hyperspectral remote sensing of a coral reef environment will lead to accurate identification and subsequent monitoring of changes in coral health and overall well being.