Holden, Heather M.2006-07-282006-07-2819991999http://hdl.handle.net/10012/381Increased awareness of the vulnerability of coral reef ecosystems to the synergistic effects of natural and anthropogenic environmental changes has lead to subjective reporting of observed changes, but there has been a significant lack of objective monitoring of coral reef ecosystems on a repetitive basis. Only consistent, repetitive monitoring over time will increase our understanding of the dynamic and complex nature of coral reef ecosystems and the ways in which changes in the ecosystem are related to environmental changes. One limiting factor to remote detection of coral reef well-being is the lack of a quantitative means of identifying optically similar features such as healthy coral and macroalgae. In this study, a field program was designed to explore the differences in spectral reflectance characteristics of various coral reef features. High spectral resolution in situ data were collected with a hand-held hyperspectral radiometer. In 1996, in situ spectral reflectance data of submerged coral reefs were collected in Beqa Lagoon, Fiji; in 1997, in situ spectral reflectance measurements of exposed coral reef features were collected in Manado, Indonesia. Finally, in situ data of submerged coral reefs were collected in 1998 in Savusavu Bay, Fiji. The spectra collected were divided into populations of healthy coral, unhealthy coral, algae-covered surfaces and rubble surfaces based on feature type according to field notes and photographic records. These data sets were compared and analyzed to test the following hypotheses. First, the within-population variability is low such that spectra of similar coral reef features display similar spectral reflectance characteristics, and conversely, there are discernable spectral reflectance differences between populations. Secondly, the geographic location of measurement does not affect the spectral reflectance characteristics. The final hypothesis tested is that the slopes and changes in slopes of the spectral reflectance curves will allow differentiation of populations and subsequent classification. Cluster and correlation analyses indicate that both the within- and between-population variability is low. Therefore, while spectra of similar features are comparable, there are predictable inaccuracies in classification due to spectral similarities between populations. Nevertheless, principal components analysis was used successfully as a data reduction tool to reduce the large data set of 334 spectra to 6 spectra representative of the pre-defined populations. A classification scheme was devised based on these representative spectra such that the slope, change in slope and magnitude of reflectance of the spectral curves enabled identification. This classification scheme was devised based on these representative spectra such that the slope, change in slope, and magnitude of reflectance of the spectral curves enabled identification. This classification procedure was applied to the remainder of the data set and an error analysis was performed to investigate accuracy of identification. The overall accuracy was 80.1% and an investigation of the errors of omission and commission indicate that the majority of the errors are a result of an inability to characterize bleached coral correctly. The results of this study indicate that hyperspectral remote sensing may be a feasible means of accurate identification and subsequent monitoring of changes in coral health and overall well-being.application/pdf10088728 bytesapplication/pdfenCopyright: 1999, Holden, Heather M.. All rights reserved.Harvested from Collections CanadaDoctoral Thesis