Examining CDOM fluorescence variability using principal component analysis: seasonal and regional modeling of three-dimensional fluorescence in the Gulf of Mexico

Abstract

Alterations in colored dissolved organic matter (CDOM) characteristics were examined for an extensive fluorescence data set generated by excitation emission matrix spectroscopy (EEMS). Principal component analysis (PCA) was used to assess changes in fluorescence bandwidth and wavelength shifts of entire peak regions, enhancing previous CDOM characterization that was based on changes observed at a few excitation and emission wavelength pairs. Variability within the data set is expressed as a series of linear terms, which can be used to quantitatively evaluate changes in CDOM fluorescence within the perspective of a single cruise or a large geographic region. In this paper we will present PCA results from a comprehensive fluorescence data set examining seasonal and regional CDOM variability within the Gulf of Mexico.

Introduction

Fluorescence methods such as excitation emission matrix spectroscopy (EEMS) are well suited to analyze a mixture of fluorophores of unknown spectral properties, utilizing multiple excitation (λ_i) and emission wavelengths (λ_j) with intensity to examine total luminescence characteristics. Application of EEMS to colored dissolved organic matter (CDOM) in seawater has resulted in the distinguishing of humic and protein-like fractions of CDOM and descriptions of CDOM fluorescence in terms of excitation and emission maxima Coble et al., 1990, Coble et al., 1998, Mopper and Schultz, 1993, De Souza Sierra et al., 1994, Coble, 1996. Using variations in the natural fluorescence properties of CDOM, differing water masses in the marine environment have been identified Cabaniss and Schuman, 1987, Chen and Bada, 1992, Mopper and Schultz, 1993, Zika et al., 1993, Coble, 1996, and CDOM in both the water column and sediment porewaters has been characterized Coble et al., 1990, Coble et al., 1998, Benamou et al., 1994, Chen and Bada, 1994, De Souza Sierra et al., 1994, Coble, 1996.

Changes in CDOM fluorescence reflect effects of physical and chemical processes that occur in the aquatic environment as well as variations in CDOM chemical composition and concentration from different sources. Fluorescence characteristics are selective for changing chemical components based on the closely spaced molecular energy levels unique to each component's chemical structure and its concentration in solution. For an excitation emission matrix (EEM), this can be expressed in an equation similar to Beer's Law (Malinowsky, 1991)

$$\text{M}{}_{\mathrm{ij}}\text{=}\text{∑}\text{k=1}\text{n}\text{α}{}_{\mathrm{k}}\text{X}{}_{\mathrm{ik}}\text{Y}{}_{\mathrm{kj}}$$

where α is the concentration of component k, X_ik is the number of photons absorbed by fluorophore k at λ_i, and Y_kj is the fraction of emission produced by fluorophore k at λ_j. Discrimination and identification of individual fluorophores in a mixture is based on wavelength and intensity differences between matrices.

A large amount of information regarding CDOM composition and alteration is presented within each matrix. Traditionally, variability between the three-dimensional fingerprints produced by EEMS has been assessed by examining changes in intensity and wavelength position at peaks in fluorescence, or by observing intensity changes at discrete wavelength pairs. In the case of CDOM, this generally results in the reporting of four or perhaps five wavelength pairs to characterize CDOM fluorescence. This type of “peak picking” for data analysis is cumbersome, time-intensive and disregards most of the collected data. Given that CDOM is a mixture of unknown components with potentially numerous fluorophores De Souza Sierra et al., 1994, Boehme and Coble, 2000, it is desirable to include as many wavelengths as possible when evaluating fluorescence changes between matrices. Confidence in visual comparison of fluorescence differences decreases within a large data set as the comparison becomes prone to bias. Since the amount of EEMS data accumulated for different geographic regions and different seasons continue to expand, there is an increased requirement for a quantitative method of characterizing individual samples within a larger data set.

Principal component analyses (PCA and related spectral decomposition methods) have been applied successfully to 2d and 3d fluorescence data sets Warner et al., 1977, Machado and Esteves da Silva, 1993, Henrion et al., 1997, Persson and Wedborg, 2001 and were selected to statistically differentiate changes in CDOM fluorescence. Uncorrected matrix correlation (Mobed et al., 1996) has been used to compare two fluorescence matrices at a time for significant differences, but is inappropriate for large data sets. Principal component analysis (PCA) allows for comparison between many matrices, including large numbers of excitation and emission wavelength pairs in determining sample similarity rather than focusing on wavelength and intensity changes at absolute and local fluorescence maxima. This approach represents an improvement over previous CDOM characterization by including variability easily overlooked on the excitation and emission shoulders of peaks caused by shifts in peak wavelengths and changes in spectral bandwidth, and results in a more comprehensive assessment of CDOM fluorescence differences. Variability within the data set is expressed in each of the components generated by PCA. By examining the scores and loadings associated with sequential principal components, it is possible to group seawater samples based on changes in three-dimensional fluorescence and examine the variability contributed by different spectral regions. This technique was applied to a fluorescence data set collected from the Gulf of Mexico over several years to examine the humic fluorescence variability observed in that region. However, by expanding data sets to include other regions PCA could be applied to improve our understanding of CDOM dynamics in the ocean on a global scale.