This research compared the ability of Landsat ETM+, Quickbird and three image classification methods for discriminating amongst coral reefs and associated habitats in Pacific Panama. Landsat ETM+ and Quickbird were able to discriminate coarse and intermediate habitat classes, but this was sensitive to classification method. Quickbird was significantly more accurate than Landsat (14% to 17%). Contextual editing was found to improve the user's accuracy of important habitats. The integration of object-oriented classification with non-spectral information in eCognition produced the most accurate results. This method allowed sufficiently accurate maps to be produced from Landsat, which was not possible using the maximum likelihood classifier. Object-oriented classification was up to 24% more accurate than the maximum likelihood classifier for Landsat and up to 17% more accurate for Quickbird. The research indicates that classification methodology should be an important consideration in coral reef remote sensing. An object-oriented approach to image classification shows potential for improving coral reef resource inventory.
No Reference information available - sign in for access.
No Citation information available - sign in for access.
No Supplementary Data.
Document Type: Research Article
Marine & Coastal Development Unit, Argyll & Bute Council, Oban, Argyll, UK
Smithsonian Tropical Research Institute, Panama
Heriot-Watt University, School of Life Sciences, Edinburgh, UK
Heriot-Watt University, School of the Built Environment, Edinburgh, UK
Publication date: 2007-01-01
More about this publication?