Abstract:

One Landsat-7 image of 1999 and two Landsat-5 images of 2001, 2009 were chosen to map and detect changes of coral reef community at Dongsha atoll. As the support vector machine (SVM) is very attractive for the classification of remotely sensed data, it was introduced into the extraction and classification of the coral reefs. Considering the training time and classification accuracy, ‘one against one’ and radial basis kernel function (RBF) were determined. The results show that the total area of the coral reefs at Dongsha atoll is about 140.93 km² in 2009. From 1999 to 2009, the coral reefs were in the intermediate stage of degradation, because their areas decreased 17.54 km² and they behaved evident fragmentation. High spatial resolution images are helpful to extract coral reefs properly and accurately, especially to minor area of coral reefs.

Key words: Dongsha atoll, remote sensing, coral reef, support vector machines, geographic information systems