Using satellite data to map coral reefs in the South China Sea
The Dong-Sha Islands remote sensing survey project, part of a broader Chinese offshore investigation and assessment that began in 2004, aims to evaluate the capability of satellite imagery for assessing coral reef distribution. Images from three different satellites (QuickBird, ETM aboard Landsat-7, and SPOT5) have been used to categorize the current distribution of coral reefs at the Dong-Sha Atoll.
More than 300km2 large, with an average lagoon depth of 10m, the Dong-Sha Atoll is located at the northern continental margin of the South China Sea. It comprises several small banks and a 1.74km2 small island which rises 2m above sea level on the west side of the atoll. The surface water temperature reaches ~30° C in summer and falls to 21° C in winter. According to estimates from previous research,1 the number of recorded coral species decreased by 55.5%, from 137 species in 1994 to 61 species in 1998. Up to 90% of the corals in the reef communities were found to be dead in 1998, reportedly due to extensive use of poisons and explosives in illegal fishing practices.
Maps of this reef habitat could be a valuable planning tool. Ideally, mapping could be used to identify key representative reef sites for repeated study, and to manage the area. However, the vertical variation in the community structures at this site complicates surveying: at low tide, the water is too shallow for boat-based surveys; but at high tide, the water is too deep for foot-based surveys. Therefore, traditional field surveys are too time-consuming and expensive to conduct over a continuum of scales. Indeed, the most cost-effective approach would be to use satellite imagery for large-scale reef surveys, as previously demonstrated in Kaneohe Bay, Hawaii.2
Both multiband and higher-resolution pan-color images were collected from all three satellites for the whole area of the atoll. In preprocessing, the only known global positioning system (GPS) control point in the Dong-Sha Islands was used to geometrically calibrate the QuickBird images, and in turn the related low-resolution images from SPOT5 and ETM were calibrated using the QuickBird data. For information extraction, unsupervised classification functions were applied to the ETM and SPOT5 data. The QuickBird images were unsupervisedly classified using the ISODATA clustering image classification technique and maximum likelihood classification method in succession. The results of classification, which were (or not) merged into the coral reef, were analyzed by simplification and topological analysis. The resulting vector data was imported into a geographic information system.
The total number of coral patches at Dong-Sha Atoll is 1,995, according to results from classification of the QuickBird multi-band data. One thousand three hundred and thirty-one coral reefs are larger than 100m2, with a detection limitation of 3×3 pixels. QuickBird images show finer details of the island and the atoll due to a channel setting that enables greater penetration into the water and finer spatial resolution.
The results extracted from SPOT5 and ETM images failed to reveal many smaller reefs that the results from the QuickBird images did reveal, though they showed a similar number of reefs larger than 1,000m2. Furthermore, the results from SPOT5 data are superior to those of ETM data for silty lagoon conditions owing to its higher resolution, but inferior for the reef bodies, where the blue waveband is missing. The channel setting of SPOT5 is not specialized for undersea coral reef monitoring, and its XS band 2 failed to distinguish deep substrate because of its poor penetration capability. SPOT5's XS band 1 cannot differentiate coral reef from sand bottom, while ETM can distinguish these types. The pixel resolution of ETM multiband data is 30×30m. Therefore, pixel sizes (20–30m) are of a similar magnitude to the size of habitat patches. In the SPOT5 pan-color images, 4×4 pixels offer enhanced spatial resolution to deliver greater thematic accuracy, so the resolution of SPOT5 may offer better results. However, in the case of both SPOT5 and ETM, sensor limitations in spectral and spatial resolution led to ambiguous classes: distinct substances like coral and water were ultimately not discernible due to the spectral and spatial coarseness in the imagery.
QuickBird images show finer details of the atoll. Results extracted from SPOT5 and ETM images are fewer in number and area than those from QuickBird images. Merging SPOT5 multiband data with pan-color data and referring to the results from ETM data are expected to provide a cost-effective satellite-based approach to coral reef mapping.
The study was supported by the National Natural Science Foundation of China (grant no. 40606040) and the Chinese Offshore Investigation and Assessment (the Dong-Sha Islands remote sensing survey experimental project).
