Application of clustering techniques to multispectral optical data over the ocean

MERIS, on Envisat, provides high-resolution radiometric data at nine discrete channels in the visible band. This paper looks at the potential of an unsupervised classification technique for utilizing these multi-spectral data to provide better discrimination between water masses according to their optical properties, and in particular whether phytoplankton groups can be distinguished. Although the majority of data do show a spectral peak associated with chlorophyll's red fluorescence line, clustering using only the red bands was found to separate out coastal waters according to their sediment content. Red-end classification also appeared to identify sub-pixel cloud, and demonstrate that the smile correction had not removed all the striping from the data. Classification using bands from the blue-green end showed a response to changes in chlorophyll concentration, but also indicated other variations. However, without in situ data no firm conclusions can be drawn on which phytoplankton groupings are present.