Mapping the distribution of an invasive marine alga (Codium fragile spp. tomentosoides) in optically shallow coastal waters using the compact airborne spectrographic imager (CASI)

Abstract

We collected 19 bands of ocean colour data (spanning 391–904 nm) at 1 m2 spatial resolution from Mahone Bay on the south coast of Nova Scotia during July 2001 with an airborne hyperspectral sensor (CASI). The data were classified using 16 different protocols in an effort to accurately map benthic communities over 7 km2 of seabed to 6 m water depth. The primary objective was to depict the spatial pattern of invasion of the rocky subtidal zone by an introduced macroalga: Codium fragile. The best classification results were obtained using the first three axes of a principal component analysis of all 19 spectral channels and the maximum likelihood classification of four classes of benthic community (i.e., those dominated by Codium meadow, kelp bed, Codium–kelp mix, and sand) at three depth strata using an added bathymetry channel and no correction for water-column attenuation. The overall accuracy obtained for the entire visible seabed of the bay was 83.30% (Kappa statistic = 0.82). User's and producer's accuracies of the four classes ranged from 35.00% to 97.00% and 65.00% to 100.00%, respectively, depending primarily on depth. The addition of a bathymetry channel typically increased the overall accuracy by 20.00%, and a correction for water-column attenuation had little effect at these depths. Patches of the invasive alga, kelp, and sand were clearly distinguishable at spatial scales of 1–1000 m2, but there was also patchiness at subpixel scales (i.e., <1 m2), such that Codium was frequently confused with mixed communities. We interpret this multiscale patchiness as indicative of ongoing invasion dynamics and conclude that airborne hyperspectral technology is suitable for portraying the time-dependent outcomes of these dynamics at ecologically meaningful spatial scales.