Studies On Fluorescence Features Of The Excitation-emission Matrix (EEM) Of Phytoplankton

Monitoring phytoplankton classes and their abundance is a rountine task in marine scientific research. With the frequent occurances of red-tide, there is an urgent need for rapid analysis methods that can provide qualitative and quantitative information. Three dimensional fluorescence spectrum, also termed as excitation-emission matrix (EEM) spectrum, gives total fingerprints information in the whole range of wavelenghths. By using chemomatrics on extracting their spectral features, EEM spectrum is often used for multicomponents analysis. However, there is no report on analysis of phytoplankton classes using EEM fluorescence spectra in the world.This paper aims to study the characteristics of EEM spectra of phytoplankton. Eleven phytoplankton species, that is Alexandrium tamarense, Prorocentrum donghaiense, Pseudo-nitzschia pungens, Skeletonema costatuma, Nitzschia closterium, Chaetoceros curvisetus , Chaetoceros debilis, Chaetoceros didymus, Isochrysis galbana, Platymonas helgolanidica and Synechococcus sp., belong to five divisions of the typical red-tide algae species and dominant species in the East China Sea, were chosen and grown in lab cultures under different temperatures and different illumination intensities. The fingerprint features are extracted to identify these different phytoplankton species .All these work will lay the foundation for developing sea algae analyzer in the future. The main research work is as follows:1. The measuring precision of fluorescence spectra of the "suspension" living phtoplankton was studied. The definition of relative error on two EEM spectra was given. The results show that the difference of the relative errors between parallel measurements of two times and that of eight times is less than 1%. Thus, in the formal experiments, every sample was measured twice. For all the eleven species of phytoplankton, the average relative errors of parallels are less than 10%, except for some single species, such as Pseudo-nitzschiapungens, the relative errors of its parallels reach 15%. For EEM spectra removed of the Rayleigh scatterings, most of the relative errors of the parallels are below 5%. 2. Spectral feature extraction methods were studied. By comparison, the 95-99 rows (Emission wavelenghs: 670-690nm) of the EEM removed Rayleigh scattering were processed by singular value decomposition (SVD), the first principal component of the obtaining excitation spectrum, symbolized as CharDVl, has distinguishing ability, and it was taken as the characteristic spectrum. Observing these characteristic spectra, it is clear to see that Isochrysis galbana, Platymonas helgolanidica and Skeletonema costatuma have high degrees of similarity to their own species samples, while the spectra similarities of Alexandrium tamarense, Prorocentrum donghaiense, Pseudo-nitzschia pungens, Chaetoceros curvisetus, Chaetoceros debilis, Chaetoceros didymus, and Synechococcus sp are not as significant as the other three species. At the same temperature, the similarities to one species samples increase. Bayes discriminant analysis was used to test the validity and repeatability of the characteristic spectra. The results show that the recogintion error rates are less than 9% neglecting the discriminant error of the six diatom species in various combinations of training sets and test sets at the temperatures of 15 °C and 20 °C. Correspondingly, the recogintion error rates are below 16% at circumstances of 25 °C. The correlation coefficients, the ratio value of S2 and Si are both used to compare the similarities of spectra. The results are the same as the visual results. All the characteristic spectra of a phytoplankton at one certain condition were processed by SVD to form the standard fluoroscence spectrum, which reveal some important information. First, the spectrum of Synechococcus sp. (a species of Cyanophyta) is totally different from the other spectra. Second, although all the other 10 spectra have a peak around 450nm, the spectrum of 7s (a species of Chrysophyta) and that of PI (a species of Chlorophyta) have no obvious peak at 275nm. This makes them quite different from the rest. The minor differences between them can be distinguished clearly. Third, even though the other eight spectra have peaks around 450nm and 275nm, the essential difference is that for Al and Pr (species of Dinophyta), the highest peak in their spectra appears at a wavelength more than 450nm, while that in the other 6 spectra, belonging to Bacillariophyta, appears at a wavelength less than 450nm. Fourth, all 6 species belonging to Bacillariophyta, except Sk, have similar characteristicspectra. The spectra of Cu, De and Di are identical as they all belong to the same genus, Chaetoceros.3. Fine spectral features were extracted from the characteristic spectra {CharDVl and CharCUl), hierarchical clustering was used to calssify the characteristic spectra of each phytoplankton and the standard fluorescence spectra database for the typical phytoplankton of the East China Sea were formed. The discriminant analysis results show that CharDVl still have better discriminant ability than the extracted spectral features and the combination features from CharDVl and CharCUl. Hierarchical clustering was preceeded to classify the CharDVl of each phytoplankton. The correlation coefficient equals to 0.7, Isochrysis galbana, Platymonas helgolanidica, Prorocentrum donghaiense, Chaetoceros didymus and Synechococcus sp. form one standard fluorescence spectrum of each species; Alexandrium tamarense, Pseudo-nitzschia pungens, Skeletonema costatuma .Chaetoceros curvisetus and Chaetoceros debilis, each species have two standard fluorescence spectra. Nitzschia closterium have three standard fluorescence spectra. In the standard fluorescence database, each phytoplankton has one to three spectra.4. Qualitative and quantitative analysis of 210 mixtures of phytoplankton were carried out using the Non-negative Least Square (NNLS) method. For NNLS, methods for estimating number of components in the mixture and the regulation of correct recognition were discussed. The threshold value of 0.25 was chosen to assure the most abundant phytoplankton classes present can be identified and their relative abundance estimated, but the less abundant classes cannot. In this experiment, the dinoflagellates can be distinguished from diatoms at a rocoginition rate of 76%. A red-tide seawater sample was also analysized. By comparison with cell counting result, the identified principal component is correct, and its abundance is at the same quantitative level with the cell-counting number.In summary, this paper takes the lead in studying the EEM spectrum of living phytoplankton. Using the obtained characteristic spectra and NNLS method, dinoflagellates and diatoms can be distinguished from each other in laboratory cultures and seawater sample.