**Multiple linear regression analysis**

The principal motivation behind the use of multiple linear regression studies of phytoplankton pigments was to establish a relationship between diagnostic pigments & the ubiquitous light harvesting pigment, chlorophyll-*a*, a universal proxy of phytoplankton biomass. One of the aims of the approach was to account for the fraction of phytoplankton cell biomass that may not be detected by conventional microscopy counts. Using multiple linear regression of diagnostic pigments against chlorophyll-a concentration, Gieskes & Kraay (1983) revealed that cryptophytes accounted for about half of the chlorophyll-a standing stock during a spring bloom study of the central North Sea, although these cells were not detected by microscope counts. The evidence was compelling, given that cryptophytes have a unique 'pigment

Ocean Color Remote Sensing of Phytoplankton Functional Types 113

the analyses, the CHEMTAX pigment matrixes make use of a wider range of accessory pigments. However, some criticisms within the community have occurred regarding its

Given that the growth conditions of microalgae are known to influence the pigment composition of phytoplankton. The principal factors that led to chromatic adaptation of microalgae include the quality and spectral quantity of irradiance and nutrient status. In addition to the effect of taxa, growth conditions of microalgae are known to influence the pigment composition of phytoplankton. The principal factors that lead to chromatic adaptation of microalgae include the quality and spectral quantity of irradiance and nutrient status. Variability in pigment composition caused by these factors is seen both in the change in the intracellular concentration of the cells as well as the ratios of accessory pigments (such as the relative concentrations of photosynthetic or photoprotective pigments). Therefore, similar to inversion methods, it is recommended that sample be divided according to light

Other inverse methods focus on the absorption term related to phytoplankton as has been estimated using a wide range of algorithms. An alternative approach proposed by Bidigare et al. (1989 & 1990) utilizes the absorption signatures of the various pigments to reconstruct the unpackaged absorption spectra of marine phytoplankton. Using the pigment outputs for various algal groups from CHEMTAX, an unpackaged absorption spectra for a particular phytoplankton group can be constructed when regarding the absorption spectra solely as a function of the chlorophyll-a concentration and the chlorophyll-a specific absorption spectra. Several inverse modeling capabilities are available that can make use of these absorption formulations for other pigments. Moisan et al. (2011ab) have developed a matrix inverse modeling technique that produces photoprotective and photosynthetic pigments with

Fortuitously, the major taxa of marine phytoplankton tend to fall within the three size classes originally proposed by Sieburth (1979), the micro- (>20m), nano- (2-20m) and picophytoplankton (<2m). Thus pigment markers of particular taxonomic groups can provide insight into the relative contribution of a particular size class to the pigment biomass. The use of pigment markers to derive a size index of phytoplankton populations was proposed by Claustre (1994). The approach used seven diagnostic pigments to obtain an index of the relative contribution of microphytoplankton to pigment biomass integrated over the watercolumn. The two marker pigments ascribed to the microphytoplankton size class are fucoxanthin, associated with the diatom fraction, and peridinin, representing the dinoflagellate fraction. The index also used markers pigments that tend be dominant accessory pigments in natural populations: 19'-hex, 19'-but, alloxanthin, zeaxanthin and chlorophyll-b. Thus by summing the watercolumn-integrated concentration of diagnostic pigments associated with the microphytoplankton fraction, and dividing by the sum of all water-column integrated accessory pigments, an index of the relative contribution of microphytoplankton to total integrated pigment biomass can be calculated according to the

( )

 .

*fuco perid hex but allox zeax chlb*

relatively accurate results (r2 >0.80) using HPLC pigments as validation (Figure 7).

application because it assumes constant pigment ratios (Latasa 2007).

regime before running the matrix factorization.

**Modeling pigments in relation to cell size fractions** 

19' 19' *<sup>p</sup> fuco perid <sup>F</sup>*

equation:

fingerprint' (alloxanthin), and this carotenoid was the most abundant accessory pigment in the field samples. The explanation for the discrepancy between the pigment data and the cell counts was that small flagellates and photosynthetic ciliates that harbour alloxanthin are poorly preserved in Lugol's solution, resulting in underestimated cell abundances whereas the larger microphytoplankton (diatoms and dinoflagellates) are better preserved and more easily to identified. The authors also commented on the difficulty in assigning other pigments, such as fucoxanthin, to a particular class, given it is also found in other algal classes, such as prymnesiophytes and chrysophytes.

There are caveats to using the multiple linear regression approach to examine shifts in phytoplankton community structure (Gieskes et al. 1988). In the study by Gieskes et al. (1988), a clear influence of light adaptation on pigment composition was apparent in surface and deep populations, which were characterized by different ratios and intracellular concentrations of diagnostic pigments. As a result, poor taxon specificity of pigment markers such as fucoxanthin may mislead authors to conclude that microscopic counts provide superior quantitative estimates of taxon-specific biomass compared to pigment data.

#### **Inverse methods**

In order to estimate the contribution of various algal taxa to chlorophyll-a, inverse methods have been adopted to studies of phytoplankton pigments. The inversion method uses ratios of chlorophyll-a to accessory pigments based on literature values for representative taxa, (Everitt et al. 1990, Letelier et al. 1993, Bidigare et al. 1996). From these data a series of simultaneous equations for each algal group is derived to determine their contribution to total chlorophyll-a biomass. The method also addresses the problem of shared pigments between classes by subtracting out the contribution by other taxonomic groups for a particular pigment marker. The culture-based pigment: chlorophyll-a seed values are then modified by matrix inversion to the field data to find the least squares best solution. Constraints are also applied to the pigment ratios to avoid negative contributions by taxa.

The inverse method was used by Everitt et al. (1990) to demonstrate that nanophytoplankton species dominated the waters of the western Equatorial Pacific, and showed significant variation in community structure. Seasonal changes in the composition of phytoplankton groups (*Prochlorococcus*, other cyanobacteria, prymnesiophytes and chrysophytes) that comprise the deep chlorophyll maximum were examined for the Station ALOHA time series (Letelier et al. 1993). This approach has also been adopted by Vidussi et al. (2001) to assess the shift in community structure during a spring bloom in the northwestern Mediterranean Sea.

#### **Matrix factorization of HPLC pigments (CHEMTAX)**

CHEMTAX (CHEMical TAXonomy) is a MATLAB program that estimates the relative contributions of different phytoplankton taxa to the bulk chlorophyll-a concentration of a given sample (Mackey et al. 1996, Wright et al. 2000). Over the years this method has been used to separate out the phytoplankton community into taxonomic groups to at least the class, and in some cases, the genus or species level. To implement this algorithm an input pigment matrix is constructed based on knowledge of the kinds of taxa likely to be present in the study area and information on the cell-specific pigment composition and concentrations of these groups from culture studies. Unlike the multiple linear regression or inversion methods, which use only one or two diagnostic pigment markers for each taxa in

fingerprint' (alloxanthin), and this carotenoid was the most abundant accessory pigment in the field samples. The explanation for the discrepancy between the pigment data and the cell counts was that small flagellates and photosynthetic ciliates that harbour alloxanthin are poorly preserved in Lugol's solution, resulting in underestimated cell abundances whereas the larger microphytoplankton (diatoms and dinoflagellates) are better preserved and more easily to identified. The authors also commented on the difficulty in assigning other pigments, such as fucoxanthin, to a particular class, given it is also found in other algal classes, such as

There are caveats to using the multiple linear regression approach to examine shifts in phytoplankton community structure (Gieskes et al. 1988). In the study by Gieskes et al. (1988), a clear influence of light adaptation on pigment composition was apparent in surface and deep populations, which were characterized by different ratios and intracellular concentrations of diagnostic pigments. As a result, poor taxon specificity of pigment markers such as fucoxanthin may mislead authors to conclude that microscopic counts provide superior quantitative estimates of taxon-specific biomass compared to pigment

In order to estimate the contribution of various algal taxa to chlorophyll-a, inverse methods have been adopted to studies of phytoplankton pigments. The inversion method uses ratios of chlorophyll-a to accessory pigments based on literature values for representative taxa, (Everitt et al. 1990, Letelier et al. 1993, Bidigare et al. 1996). From these data a series of simultaneous equations for each algal group is derived to determine their contribution to total chlorophyll-a biomass. The method also addresses the problem of shared pigments between classes by subtracting out the contribution by other taxonomic groups for a particular pigment marker. The culture-based pigment: chlorophyll-a seed values are then modified by matrix inversion to the field data to find the least squares best solution. Constraints are also applied to the pigment ratios to avoid negative contributions by taxa. The inverse method was used by Everitt et al. (1990) to demonstrate that nanophytoplankton species dominated the waters of the western Equatorial Pacific, and showed significant variation in community structure. Seasonal changes in the composition of phytoplankton groups (*Prochlorococcus*, other cyanobacteria, prymnesiophytes and chrysophytes) that comprise the deep chlorophyll maximum were examined for the Station ALOHA time series (Letelier et al. 1993). This approach has also been adopted by Vidussi et al. (2001) to assess the shift in community structure during a spring bloom in the

CHEMTAX (CHEMical TAXonomy) is a MATLAB program that estimates the relative contributions of different phytoplankton taxa to the bulk chlorophyll-a concentration of a given sample (Mackey et al. 1996, Wright et al. 2000). Over the years this method has been used to separate out the phytoplankton community into taxonomic groups to at least the class, and in some cases, the genus or species level. To implement this algorithm an input pigment matrix is constructed based on knowledge of the kinds of taxa likely to be present in the study area and information on the cell-specific pigment composition and concentrations of these groups from culture studies. Unlike the multiple linear regression or inversion methods, which use only one or two diagnostic pigment markers for each taxa in

prymnesiophytes and chrysophytes.

northwestern Mediterranean Sea.

**Matrix factorization of HPLC pigments (CHEMTAX)** 

data.

**Inverse methods** 

the analyses, the CHEMTAX pigment matrixes make use of a wider range of accessory pigments. However, some criticisms within the community have occurred regarding its application because it assumes constant pigment ratios (Latasa 2007).

Given that the growth conditions of microalgae are known to influence the pigment composition of phytoplankton. The principal factors that led to chromatic adaptation of microalgae include the quality and spectral quantity of irradiance and nutrient status. In addition to the effect of taxa, growth conditions of microalgae are known to influence the pigment composition of phytoplankton. The principal factors that lead to chromatic adaptation of microalgae include the quality and spectral quantity of irradiance and nutrient status. Variability in pigment composition caused by these factors is seen both in the change in the intracellular concentration of the cells as well as the ratios of accessory pigments (such as the relative concentrations of photosynthetic or photoprotective pigments). Therefore, similar to inversion methods, it is recommended that sample be divided according to light regime before running the matrix factorization.

Other inverse methods focus on the absorption term related to phytoplankton as has been estimated using a wide range of algorithms. An alternative approach proposed by Bidigare et al. (1989 & 1990) utilizes the absorption signatures of the various pigments to reconstruct the unpackaged absorption spectra of marine phytoplankton. Using the pigment outputs for various algal groups from CHEMTAX, an unpackaged absorption spectra for a particular phytoplankton group can be constructed when regarding the absorption spectra solely as a function of the chlorophyll-a concentration and the chlorophyll-a specific absorption spectra. Several inverse modeling capabilities are available that can make use of these absorption formulations for other pigments. Moisan et al. (2011ab) have developed a matrix inverse modeling technique that produces photoprotective and photosynthetic pigments with relatively accurate results (r2 >0.80) using HPLC pigments as validation (Figure 7).

#### **Modeling pigments in relation to cell size fractions**

Fortuitously, the major taxa of marine phytoplankton tend to fall within the three size classes originally proposed by Sieburth (1979), the micro- (>20m), nano- (2-20m) and picophytoplankton (<2m). Thus pigment markers of particular taxonomic groups can provide insight into the relative contribution of a particular size class to the pigment biomass. The use of pigment markers to derive a size index of phytoplankton populations was proposed by Claustre (1994). The approach used seven diagnostic pigments to obtain an index of the relative contribution of microphytoplankton to pigment biomass integrated over the watercolumn. The two marker pigments ascribed to the microphytoplankton size class are fucoxanthin, associated with the diatom fraction, and peridinin, representing the dinoflagellate fraction. The index also used markers pigments that tend be dominant accessory pigments in natural populations: 19'-hex, 19'-but, alloxanthin, zeaxanthin and chlorophyll-b. Thus by summing the watercolumn-integrated concentration of diagnostic pigments associated with the microphytoplankton fraction, and dividing by the sum of all water-column integrated accessory pigments, an index of the relative contribution of microphytoplankton to total integrated pigment biomass can be calculated according to the equation:

$$F\_p = \frac{(\sum fuco + \sum perid)}{\sum fuco + \sum perid + \sum 19\text{'} hex + \sum 19\text{'} but + \sum 2 llox + \sum zeax + \sum dhlb} \cdot 1000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000$$

Ocean Color Remote Sensing of Phytoplankton Functional Types 115

A *Fp* value equal to one means that fucoxanthin and peridinin made up the entire integrated diagnostic pigment concentration, whereas a value of zero means that neither indicator

The size-based approach was refined by Vidussi et al. (2001) to obtain a pigment-derived indices of the contribution of the three size classes of phytoplankton (micro- (>20mm), nano- (2-20mm) and picophytoplankton (<2mm) to integrate pigment biomass. To determine the relative contribution of the three size classes to chlorophyll-a biomass, Uitz et al. (2006) combined the multiple linear regression analysis of Gieskes et al. (1983, 1988) with the sizespecific index of Vidussi et al. (2001) to derive estimates of all three size fractions and used not only to derive an index of the presence of the three size classes, but also to estimate their relative contribution to total chlorophyll concentration, similar to the inversion methods and CHEMTAX. It is important to note that in this approach integrated rather than discrete pigment concentrations are used. The approach to using size as a description of the phytoplankton community is becoming a well-accepted PFT, which provides an important

**6. Modeling trophic structure through correspondence between taxa, size** 

The relationship between phytoplankton size and biogeochemical function has been well established. Diatoms (silicifiers) tend to be large, and their negative buoyancy leads to their significant contribution to export production (Sarmiento & Gruber 2004). Picophytoplankton, on the other hand, are more important in microbial food webs were rapid recycling of organic matter in the surface ocean leads to a reduction in biogenic export to the deep ocean. Recent molecular studies on the picocyanobacterium *Prochlorococcus*, which dominates the chlorophyll-a biomass in the oligotrophic gyres, have revealed that many strains are unable to utilize nitrate as a nutrient source, and consequently can be

The use of pigment markers to provide insight on the trophic structure of marine ecosystems was examined by Claustre (1994). He proposes that the contribution of microphytoplankton (in particular diatoms) to the integrated concentration of diagnostic pigments, which can be considered to be a indicator of the *f*-ratio and hence new production. In addition, information on the contribution of various phytoplankton taxa to chlorophyll-a standing stock has led to the generation of taxon-specific maps of primary production using global relationships on the relationship between surface chlorophyll-a concentration and the ratio of integrated diagnostic pigments to integrated chlorophyll-a concentration (Uitz et al. 2006). Although the simple pigment indices invoked by Claustre (1994) and Uitz et al. (2006) cannot resolve instances where other taxa contribute to the diagnostic pigment markers used in their analysis, it provides a useful first attempt at obtaining information on the global distribution of phytoplankton size classes (Uitz et al. 2006), the environmental factors that control their biogeography (Bouman et al. 2003) and their relative contribution to marine productivity (Bouman et al. 2005, Uitz et al. 2006).

**7. Societal benefits of ocean color approaches to algorithm development for** 

A fundamental goal of phytoplankton biogeography is to describe how PFTs are distributed spatially and temporally and how these patterns relate to processes that control primary

**key phytoplankton functional types: Fisheries and habs applications** 

pigment of microphytoplankton were found in the water column.

link between the fields of remote sensing and marine ecology.

considered obligate participants in regenerative production.

**and function** 

Fig. 7. (top) Predicted absorption versus measured absorption (Algorithm II) for values every 10 nm ranging from 300 nm to 700 nm using matrix inversion methods using a single value decomposition. (bottom) Predicted pigments versus measured HPLC pigments for use in photosynthesis and photoprotection pigments. See Moisan et al. (2011 ab for more details).

Fig. 7. (top) Predicted absorption versus measured absorption (Algorithm II) for values every 10 nm ranging from 300 nm to 700 nm using matrix inversion methods using a single value decomposition. (bottom) Predicted pigments versus measured HPLC pigments for use in photosynthesis and photoprotection pigments. See Moisan et al. (2011 ab for more details).

A *Fp* value equal to one means that fucoxanthin and peridinin made up the entire integrated diagnostic pigment concentration, whereas a value of zero means that neither indicator pigment of microphytoplankton were found in the water column.

The size-based approach was refined by Vidussi et al. (2001) to obtain a pigment-derived indices of the contribution of the three size classes of phytoplankton (micro- (>20mm), nano- (2-20mm) and picophytoplankton (<2mm) to integrate pigment biomass. To determine the relative contribution of the three size classes to chlorophyll-a biomass, Uitz et al. (2006) combined the multiple linear regression analysis of Gieskes et al. (1983, 1988) with the sizespecific index of Vidussi et al. (2001) to derive estimates of all three size fractions and used not only to derive an index of the presence of the three size classes, but also to estimate their relative contribution to total chlorophyll concentration, similar to the inversion methods and CHEMTAX. It is important to note that in this approach integrated rather than discrete pigment concentrations are used. The approach to using size as a description of the phytoplankton community is becoming a well-accepted PFT, which provides an important link between the fields of remote sensing and marine ecology.
