Seagrass Mapping in the Northern Gulf of Mexico using Airborne Hyperspectral Imagery: A Comparison of Classification Methods

Abstract Mapping areas of seagrass is important, in part because the extent of seagrass habitat can serve as a general indicator of coastal ecosystem health. While aerial photography and multispectral imagery have commonly been used to map seagrass, digital hyperspectral imagery is at the forefront of current mapping technology in many natural resource applications. In this study, HyMap hyperspectral imagery at 2.9-m resolution was used to map seagrass distribution off Horn Island, Mississippi, and estimate its areal coverage. Seagrass beds and sand-bottom classes were defined based on visual interpretation of the imagery coupled with field observations. Image spectra were sampled for each class in three water-depth zones determined by distance from shore. Supervised image classifications were performed using maximum likelihood, minimum distance to means, and spectral angle mapper methods to compare relative accuracies in mapping seagrass coverage. The maximum likelihood classification produced the highest overall accuracy of 83%. The spectral angle mapper yielded the lowest accuracy due to the predominant influence of water-column optical properties on the apparent spectral characteristics of seagrass and sand bottom. The ML classification indicated total seagrass coverage of 107 ha. This compared favorably with the results of a separate, independent study based on aerial photography acquired 1 day after the HyMap flyover. For tracking sea-grass coverage in the northern Gulf of Mexico, the mapping of individual seagrass patches at a spatial resolution of at least 3 m is recommended.