Abstract:Rapid degradation and large-scale coral bleaching of coral reef ecosystems around the world are important ecological and environmental problems. Knowledge of community structure is of great importance in determining the world's coral reef resources. Remote sensing has great potential for assessing the composition of coral reefs and the extent of reef change. Coral spectral response characteristics are affected by coral ecological habits, which may become confused by optical similarities. High spectral resolution sensors are required to perceive these subtle differences. To map the proportion of live and dead coral, the remote sensing sensor must be able to distinguish between their reflectance spectra. In this paper, we collected samples of various typical coral species with different growth characteristics from reefs near the Xisha Islands (also known as the Paracel Islands). Spectral reflectance of these samples was measured using an AvaField portable spectrometer. Spectral character analysis was carried out based on reflectance spectra, principle components analysis and derivative spectroscopy. Hyperspectral remote sensing criteria were established to distinguish corals with different growth habits. For coral, reflectance is a complex function of pigmentation, material composition structure and morphology. Coral spectral characteristics are highly variable, and are controlled by coral species and the growth environment. The coloration of corals is mainly due to symbiotic photosynthetic dinoflagellates commonly referred to as zooxanthellae. Despite variations in absolute magnitude between species, many healthy species remain a similar shape, and exhibit relatively depressed reflectance in the visible band because of the absorption of symbiotic algal photosynthesis and very rapidly increasing reflectance at wavelengths greater than 675 nm. There are either peaks or shoulders near 575, 600 and 650 nm. Coral bleaching is the result of symbiotic algae loss, which exposes the underlying white skeleton. Reductions in photosynthetic absorption and coral tissue scattering cause a rapid increase in reflectivity in visible bands, especially from 500 to 650 nm. The spectral reflectance of bleached corals has a higher amplitude and flatter shape. Algae-covered dead corals give rise to pigmentation that may be similar to that of healthy coral. Peridinin may be used as an indicator of algae-cover because it is a diagnostic pigment for dinoflagellates. Principal components analysis was employed to determine whether there are statistical differences between healthy, bleached and algae-covered corals. The first principal component explains 90%, 94%and 97%of the variance, respectively. Intra-species variability of the spectral features was not significantly different from interspecies variability. Spectral discrimination of these three states of coral is indeed possible with relative spectral response functions. Derivative spectra showed particular differences between species at specific wavelengths where several chlorophylls and other accessory pigments absorb. The results demonstrated that healthy coral, bleached coral and algae-covered dead coral can be distinguished according to derivative spectra at 522-530 nm, 564-574 nm and 600-605 nm. The overall accuracy is above 80% with the main source of error resulting from intra-species spectral variability. These results show that the application of hyperspectral remote sensing to quantitatively assess the extent of coral bleaching is feasible. The spectral libraries are an important resource, which can be used not only in remote sensing, but also in photosynthetic studies.