CORETEX · Mapping and monitoring coral Reefs using satellite image data and texture analysis methods

Final Activity Report Summary - CORETEX (Mapping and Monitoring Coral Reefs Using Satellite Image Data and Texture Analysis Methods)

The aim of this project was to investigate how texture analysis methods could be used to improve the mapping of coral reefs from high spatial resolution satellite images. Coral reefs are extremely heterogeneous in nature and as such exhibit complex spatial patterns. Identifying and understanding these patterns involves looking at the composition and the configuration of habitat elements in the reefscape. From a remote sensing perspective, this means moving beyond traditional pixel-based spectral processing techniques to incorporate textural information from the scene. A major outcome of this project was the identification of a suitable texture analysis method for the segmentation of high spatial resolution satellite images of coral reefs. The Gray level co-occurrence matrix (GLCM) method was selected because of the success of its application to the spatial analysis of remote sensing data, as well as because of its suitability for use within highly heterogeneous environments. Using QuickBird satellite imagery acquired over Heron reef in Australia, extensive analysis was carried out to answer the question of whether the extraction of textural information from the satellite image led to improved capabilities for identification of different zones on the reef. The analysis clearly demonstrated that the GLCM method outperformed spectral-based methods on this task. These results highlighted the promising role of texture analysis in the production of more detailed and accurate coral reef thematic maps. In order to associate an ecological meaning to the GLCM textural information, it was first necessary to collect field and image data. Field surveys were carried out at two major sites, namely a number of reefs in Palau in Micronesia and Heron reef in Australia. Ground truth information in the form of Global positioning system (GPS) tracked underwater benthic photographs and spectral readings were collected at a large number of locations across each reef. For each survey carried out, a QuickBird satellite image was captured around the same time as the field data was collected. Placement of the field and image data sets within a Geographic information system (GIS) allowed for their simultaneous analysis in an attempt to discover the relationship between them. A hierarchical classification scheme was developed for the analysis of underwater benthic photographs. This scheme moved beyond basic substrate categories to include aspects of the benthic cover, such as magnitude and morphology. The hierarchical nature of the scheme produced a fully adaptable classification which could be used both in detail to answer new interesting questions about the field data or in a collapsed form to allow for comparisons with previous survey results. This hierarchical classification scheme was used to analyse the thousands of benthic photographs that were taken during the field trips. The analysis resulted in a summary of benthic cover for the surveyed areas. Initial efforts to understand the relationship between the field and image data sets uncovered the actual complexity of the task. A decision was made to reduce the problem to a controlled study using simulated habitat maps. The benefit of this approach was the ability to control the compositional and configurational characteristics of the habitat maps and test how specific changes influenced the GLCM textural information. A collection of simulated habitat maps were generated using three control variables, i.e. the number of habitat types, the amount of each habitat type and the spatial configuration of each habitat type as controlled by an aggregation factor. Each map in the collection was processed using the GLCM method. Preliminary results demonstrated that the results of the GLCM method responded to changes in the composition and configuration of habitat types in a consistent manner. This allowed for the construction of a predictive model within this simplified simulation framework which acted as a starting point for understanding the relationship in the real data sets.

Текст от CORDIS, на английски · Данни: CORDIS, © Европейски съюз