Mapping marshland vegetation of San Francisco Bay, California, using hyperspectral data

Abstract

Sustainable management of wetland ecosystems requires monitoring of vegetation dynamics, which can be achieved through remote sensing. This paper assesses the use of hyperspectral imagery to study the structure of wetlands of San Francisco Bay, California, USA. Spectral mixture analysis (SMA) and multiple endmember spectral mixture analysis (MESMA) were applied on an AVIRIS (Airborne Visible and Infrared Imaging Spectrometer) image to investigate their appropriateness to characterize marshes, with emphasis on the Spartina species complex. The role of rms. error as a measure of model adequacy and different methods for image endmember extraction were also evaluated. Results indicate that both SMA and MESMA are suitable for mapping the main components of the marsh, although MESMA seems more appropriate since it can incorporate more than one endmember per class. rms. error was shown not to be a measure of SMA model adequacy, but it can be used to help to assess model adequacy within groups of related models.