Use of remote sensing for discrimination of mangrove vegetation patterns in Papua New Guinea

Exploitation of mangrove vegetation for commercial and subsistence purposes has reached critical proportions in neighbouring South East Asian countries. Initial investigations elucidated existing and potential uses of mangrove forests. Current trends indicate the largely pristine mangrove forests of Papua New Guinea will be gradually developed for commercial purposes. The objective of this study is to assess the potential of remote sensing for mapping and assessment of mangrove vegetation in Papua New Guinea. A preliminary study of oblique aerial photographs using colour and false colour infrared emulsions indicated that separate spectral classes within the broad grouping of mangrove vegetation could be identified. LANDSAT multi spectral scanner data was selected for the main project due to its superior spectral, radiometric and temporal resolution, and wide-area coverage compared to other remote sensor data available in A study area comprising 14 000 hectares of mangrove forest was selected to assess the interpretation of the data. Papua New Guinea. Analysis by a parallelepiped classification algorithm indicated that operational mapping of mangrove vegetation using ecologically-based vegetation classes was not feasible using spectral attributes alone, and a basic parallelepiped classification algorithm. The main drawback of the paralellepiped algorithm was its inability to optimise band to band correlation. Integration of terrain data with remote sensor data was used to improve the classification accuracy. Statistical analysis of the data indicated that ecologically-based vegetation classes were not valid, and modified classes based upon spectral characteristics were identified. Discriminant analysis was employed to optimize the band to band correlation in the data. A minimum distance to the mean classification algorithm using revised vegetation classes produced significantly higher overall mapping accuracies. Comparison of results from LANDSAT multi spectral scanner data classified using a parallelepiped algorithm, and manual interpretation of a colour composite image, indicated vegetation boundaries could be reliably determined using standard photo-interpretation principles on a colour composite LANDSAT image. The major conclusion of this study is that with further development of the image processing system and analysis techniques, mapping of mangrove vegetation using LANDSAT multi spectral scanner data is feasible at the Papua New Guinea University of Technology