Assessing Phytoplankton Using a Two-Rank Database Based on Excitation-Emission Fluorescence Spectra

The feasibility of using a two-rank database of reference spectra based on in vivo fluorescence excitation-emission matrix (EEMs) spectra to assess dominant groups of phytoplankton was explored. Twenty-six species belonging to 20 genera of seven divisions were studied. First, fluorescent characteristics of these EEMs were extracted using Daubechies-7 wavelet analysis. Second, the optimal characteristic spectra of scale vectors (SOCS) and time-series vectors (TOCS) were selected; phytoplankton of different genera were classified using Fisher linear discriminant analysis. Third, SOCS and TOCS reference spectra databases were obtained by hierarchical cluster analysis. Using non-negative least squares as the method to assess the phytoplankton, a two-rank reference spectra database was established according to their respective ability to identify the 2818 single-species samples. Using this fluorimetric technique, the correct identification rates (CIRs) for single-species samples were 88.8–100% at the genus level and 98.8–100% at the division level. Dominant species in the 465 laboratory mixtures had corresponding CIRs of 85.6% and 96.1%. Moreover, 15 of the 19 species used as dominants were correctly identified at the genus level. In 27 natural seawater samples, Prorocentrum donghaiense, Thalassiosira nordenskioldi, and Chaetoceros socialis (bloom-forming species with a density of about 10⁷ cell L⁻¹), and Alexandrium sp. (dominant species with abundance of about 10⁶ cell L⁻¹) were qualitatively identified at the genus level; other dominant species, with densities of 10⁵ to 10⁶ cell L⁻¹, were identified at the division level. The quantitative identification was relatively poor in the natural water samples, and several potential resolutions, including trying both new measuring methods and calculating methods, for future study are given.