Automatic detection of an invasive plant species on a barrier island in the Virginia

Invasive plant species such as Phragmites australis pose a threat to coastal habitats. This study compares a number of methods for automatically detecting Phragmites using a HYMAP scene of Smith Island, Virginia, acquired on May 8, 2000, and an IKONOS scene of the same region acquired on June 6, 2000. The best model for the phragmites distributions used both spectral and spatial-spectral input windows from HYMAP and combined Projection Pursuit (PP) for feature extraction and dimensionality reduction with the traditional ISODATA clustering technique. Although not perfect, this hybrid, unsupervised approach produced the lowest false alarm rate when compared with supervised learning models. Supervised algorithms found phragmites in the open and along the swale edges, but had inordinately high false alarm rates when compared with the PP-ISODATA hybrid.