The Application Of Multivariate Analysis And Excitation-Emission Matrix Fluorescence Spectroscopy Of Dissolved Organic Matter In The Identification Of Red Tide Algae

Harmful red tides occur frequently in the China Sea, which have caused severe damage to the ocean ecology, resources and environment. The key to predicting the red tides and reducing the influence is to identify the red tide algae real time and precisely. The fluorescence methods can satisfy the requirements of identifying the red tide algae rapidly and real time due to the high sensitivity and simple equipments, which have attracted more and more attention.Some identification methods based on the Excitation-Emission matrix spectroscopy (EEMs) of phytoplankton have been reported while the methods based on the EEMs of the filtrate of phytoplankton culture have never been seen. In this paper, ten algae species which belong to Dinophyta and Bacillariophyta were chosen to culture under laboratory conditions. The Excitation-Emission matrix spectroscopy (EEMs) of the filtrate of each alga culture during its growth process was obtained. Parallel factor analysis (PARAFAC), principal component analysis (PCA) and wavelet methods were chosen to extract the features of the Excitation-Emission matrix spectroscopy. Then multivariate linear regression was used to identify the algae species.(1) The EEMs of each alga are combined to form three-dimensional data matrix and analyzed using PARAFAC. The characteristic spectra of each alga are obtained. The characteristic spectra of each alga have dissimilarities. The cluster analysis results show that the characteristic spectra of algae belonging to the same genus are similar. The relationships between the fluorescence intensities and algae growth phases were investigated. In the exponential phase, the protein-like and humic-like substances accumulate in the filtrate of algae culture. In the stationary phase, the protein-like and humic-like fluorescence intensities increased quickly, which suggest that the two kinds of fluorescent substances were produced by the broken algal cells and the degradation by marine bacteria. These results are consistent with our previous research, which imply that the PARAFAC model can be successfully used to extract the features of EEMs and investigate the relationship between fluorescence intensities and algae growth phase.(2) The application of PCA in the identification of algae species. The PCA is used to extract features of EEMs. The first principal component loading of EEMs is chosen as the identification characteristic spectra according to the results of Bayes discriminate analysis. The identification feature spectra are established using the cluster analysis. The ten algae are tested using multivariate linear regression (using non-negative least square method) based on the feature spectra. Among the five Dinophyta species, the correct identification rates(CIR) of Prococentrum marinum and Prorocentrum donghaiense are relatively low compared to the other 3 Dinophyta species(≧ 95%). Among the five Bacillariophyta species, CIR of Skeletonema costatum and Chaetoceros debilis are low compared to the other 3 Bacillariophyta species(≧ 83%). In the level of genus, CIR of Chaetoceros can reach 96%.(3) The application of wavelet analysis in the identification of algae species. Three wavelet functions coiflet2 (coif2), daubechies-3(db-3) and daubechies-7(db-7) are used to extract the features of the EEMs respectively. The third scale vectors are selected respectively as the identification characteristic spectra for the three wavelet functions according to the results of Bayes discriminate analysis. The identification feature spectra of three wavelet functions are obtained respectively using cluster analysis. The ten algae are tested using multivariate linear regression(using non-negative least square method) based on the three types of identification feature spectra. The results show that the CIR of Prorocentrum donghaiense, Skeletonema costatum and Chaetoceros debilis are relatively low compared the the other 7 species(≧ 80%). In the level of genus, CIR of Chaetoceros are 96% for all the three wavelet function.(4) The CIR of Skeletonema costatum and Chaetoceros debilis are relatively low for all 4 types of identification feature spectra. Among the 4 types of identification feature spectra, db-3 provides the best CIR. In order to improve CIR, more work needs to be done such as feature extraction of EEMs of the red tide algae cultured under different temperature & illumination conditions and the research on the EEMs of mixtures of red tide algae.