Mapping mangrove communities in coastal wetlands using airborne hyperspectral data

Abstract

Mapping and monitoring coastal wetlands and mangrove distributions as well as changes in cover help us better manage wetlands. The purpose of this study is to study the efficacy of airborne hyperspectral remote sensing to map and detect black mangroves (Avicennia germinans) in coastal wetlands in Galveston, TX. To overcome the scarcity of labeled mangrove data, superpixel segmentation is used to expand the limited training set for subsequent classification and detection. The spatial distributions of black mangrove are then predicted with a support vector machine (SVM) classifier. The presence of black mangrove is also tested with two standard target detection approaches, including modified generalized likelihood ratio test (GLRT), and constrained energy minimization (CEM). The experimental results indicate that the black mangrove species can be effectively distinguished using hyperspectral images, from other wetland vegetation and background classes while requiring very limited labeling effort.