Multi-sensor image analysis in relation to coastal wetland mapping of southern India

摘要

The coastal zone is a complex ecosystem under the influence of various physical, chemical and biological processes. Physiographic and ecological processes in this zone are interlocked in a narrow expanse of land and water, to result in many biogeomorphic units such as rivers, wetlands, coastal lagoons, beaches, bays and estuaries. These units forming habitats for many biological communities are highly dynamic in nature and often affected by various natural and anthropogenic factors, leading to the degradation of these habitats. Accurate mapping and monitoring of these ecologically important units are, therefore, very essential in management of the coastal zones. This can only be possible with the use of currently existing remote sensing technology. The present study attempts to explore the potential use of multi sensor image data from Indian Remote Sensing Satellites (IRS), Landsat and European Remote Sensing Satellites (ERS) in order to map and monitor valuable coastal wetland features in southern India. Image fusion has been identified as potential tool to enhance different wetland features and discriminate different mangroves communities present in the study sites. Image classification methods such as ISODATA clustering and maximum likelihood classifier (MLC) were attempted on IRS-1C/1D and Landsat-TM image data to produce maps of wetland classes. ISODATA method produced maps of broad wetland cover classes with unsatisfactory classification accuracy, while MLC resulted in maps of different wetland classes with improved accuracy. However, the overall accuracy decreased when more number of classes was subject to be derived. For instance, classification accuracy was 89.3% for 5 classes, 80% for 10 classes, 76% for 15 classes, and 74.5% for 20 classes in Pitchavaram site. The deceased classification accuracy for increased number classes is thought to result from the spectral overlapping/confusion and sub-pixel mixture problem. Thus, the present study attempted to use of linear spectral unmixing model in order to produce more realistic results about various wetland features in the study sites. This method considers number of reflectance end-members that form the scene spectra, followed by the determination of their nature and finally the decomposition of the spectra into their end-members. The LSMM in general would seem to be well suited to locating small wetland habitats (cover types) which occurred as sub-pixel inclusions, and to representing continuous gradations between different habitat types.