Evaluation of large-scale unsupervised classification of New Caledonia reef ecosystems using Landsat 7 ETM+ imagery

Abstract

The capacity of the Landsat 7 Enhanced Thematic Mapper Plus sensor to classify the shallow benthic ecosytems of New Caledonia (South Pacific) is tested using a novel unsupervised classification method. The classes are defined by using a set of multiple spectral decision rules based on the image spectral bands. A general model is applied to the entire Southwest lagoon (5500 km²) and tested on three representative sites: a section of the barrier reef, a cay reef flat rich in corals, and a cay reef flat rich in algae and seagrass beds. In the latter one, the classification results are compared with a locally optimized model, with aerial color photographs and extensive ground-truthed observations. Results show that a reconnaissance of the main benthic habitats in shallow areas (<5 m depth) is possible, at a geomorphological scale for coral reef structure and at a habitat scale for seagrass beds. However, results directly issued from the model must be cautiously interpreted according to empirical spatial rules, especially to avoid confusion between coral slopes and shallow dense seagrass.