Spatial Characterization Of Marine Vegetation Using Semisupervised Hyperspectral Unmixing

Marine vegetation is considered as a bio-indicator of the quality of coastal environments. Specifically, seagrass constitutes a valuable indicator for ecosystem biodiversity. However, the estimation of the spatial extent of this vulnerable ecosystem is a challenging task. In the present research, hyperspectral airborne imagery, acquired using the AISA Eagle system, was used in combination with a spectral unmixing approach as a tool to discriminate and map costal marine vegetation, with special focus on seagrass meadows along intertidal and subtidal areas. The proposed approach is based on three main steps: extraction of endmembers, automatic classification of endmembers using a spectral library and computation of their abundances. For the endmembers extraction step, multiple runs of the VCA algorithm are followed by a clustering step to obtain the final endmembers. The intraclass variability of the data is accounted for by using the scaled version of partially constrained least squares (SCLSU). The accuracy assessment carried out shows the potential of using hyperspectral data since comparing the obtained abundance maps with the existing marine vegetation maps shows great consistency in the identification and distribution, including the two seagrass species.