Improved Spectral Angle Mapper applications for mangrove classification using SPOT5 imagery

Abstract. The traditional Spectral Angle Mapper (SAM) is an image classification method that uses image endmember spectra. Image spatial structure information may be neglected, especially in mangrove classification research where there is greater spectral similarity between species. This study combined object-oriented classification to improve the accuracy of the method in mangrove ecosystems. A mangrove area in Guangxi's coastal zone was chosen as the study site, and spectral feature analysis and ground investigations were carried out, combining pixel purification, training sample set optimization, and watershed image segmentation algorithm to improve the SAM. The improved SAM was used to classify SPOT5 remote sensing image data for a mangrove ecosystem and then classification accuracy was assessed. The results showed that the improved SAM had better classification accuracy for SPOT5 imagery. Accuracy for each mangrove species was greater than 80 % and overall accuracy was greater than 90 %, which showed that SAM was applicable for mangrove remote sensing. This application potential for classification and information extraction lays the foundation for commercialized remote sensing monitoring of mangrove ecosystems.