Spectral absorption and fluorescence characteristics of the Baltic Sea phytoplankton. ICES CM 2003/L:01

The fundamental differences in spectral properties between algal groups are well established and can be used as a starting point in the identification of algal groups. Here, I will review the seasonal variability in spectral absorption and fluorescence characteristics of living phytoplankton in the northern Baltic Sea, and study their relation to phytoplankton community structure. Especially the phycoerythrin and phycocyanin fluorescence seem to be relevant indicators for picocyanobacteria and filamentous cyanobacteria, respectively. Due to variable fluorescence and acclimation of pigmentation, deriving the biomass of spectral groups using the spectral fluorescence of multicomponent natural samples is not trivial and calls for advanced algorithms. For this purpose, multivariate methods (e.g. partial least squares, PLS) seem superior to univariate methods (e.g. classical least squares). PLS is especially applicable when signals from different constituents are overlapping, the background noise is high and variable, and not all of the optically active compounds are known. With experimental data, PLS -model was noted to give the best predictions for all spectral taxonomic groups and with the accuracy needed for algal bloom detection. However, the success of PLS -model with other data-sets is not self-evident as these models are extremely sensitive to the calibration data-sets. Possibilities for online detection of spectral phytoplankton groups by absorption and fluorescence methods will be critically examined. Introduction Recent developments in the instrumentation for algal detection are numerous, but common to the most methods is that they rely on the optical properties of the algal cells . A major drawback for optical methods is that they generally provide a bulk composite signal for sample taxonomically related signatures have to be extracted by models with variable complexity and success. Differences in the pigmentation and consequently in the spectral properties between different algal groups are well established (Fig 1, Table 1), and can be used as a starting point in the chemotaxonomical identification of the phytoplankton. Briefly, all algae contain Chla, Chlb is found in division Chlorophyta, Chlc is found in Chromophyta (except in Eustigmatophyceae). More than 400 carotenoids are known, some of them are class-specific. Phycobilins are found in classes Cyanophyta, Rhodophyta, and Cryptophyta but also some Dinoflagellates and ciliates may contain phycobilins. Instruments measuring spectral absorbance or reflectance of water sample are available, but the spectral signal is contaminated by signals from the other optically active compounds, like humic substances. This represents an additional obstacle for deriving taxonomic information from the spectra; especially in the turbid waters like in the Baltic Sea and in Finnish lakes. On the contrary, fluorescence of phytoplankton pigments takes place at the wavelengths not disturbed by other compounds. Furthermore, as the phytoplankton fluorescence is a response of photosynthetic pigments only, the signal from non-photosynthetic pigments with low taxonomic specificity is not disturbing. Spectral characteristics of the Baltic Sea phytoplankton Seasonal variability in phytoplankton absorption and fluorescence was measured in the northern Gulf of Finland for the period 11.4.-15.11.2000. Spatial variability, from the Gulf of Finland to the Bothnian Bay, was measured during a late summer cruise in 12.-18.8.2000. The seasonal variability in the light absorption (Fig. 2) was affected by the light availability, phytoplankton biomass, taxonomy, and size-structure. The low chlaspecific absorption values at red peak, indicating high package effect, were observed during times with deep mixing, high phytoplankton biomass, and during domination of large cells. The ratio of blue-to-red absorption peak was high during high light conditions, i.e. during summer and shallow mixed layer, and obviously reflecting the high amount of photoprotective carotenoids. The green peak or shoulder for phycoerythrin at 570 nm was significantly related to the biomass of picoplankton (mainly picocyanobacteria). During a summer cruise, at the phase of declining cyanobacterial bloom, it was observed that most of the phycocyanin and phycoerythrocyanin fluorescence signals originate from the large filamentous cyanobacteria (Fig. 3). The phycoerythrin signal, in contrary originates mainly from the picocyanobacteria. Multivariate calibration of phytoplankton biomass from spectral fluorescence signal Fluorescence characteristics of phytoplankton cells are highly variable due to their physiological conditions. Thus the basic assumption for analytical spectrofluorometric methods fluorescence intensity of single component is linearly related to its concentration is not valid for living phytoplankton communities. Thus, deriving the biomass of spectral groups using the spectral fluorescence of multicomponent natural samples is not trivial. For this purpose, multivariate methods (e.g. partial least squares, PLS) seem superior to univariate methods (e.g. classical least squares). PLS is especially applicable when signals from different constituents are overlapping, the background noise is high and variable, and not all of the optically active compounds are known. To construct a calibration model, spectral fluorescence must be measured from samples with known taxonomic composition. Then, phytoplankton biomass data must be organised into larger chemotaxonomic groups. The obtained model should be validated using external data-set or by cross-validation. It seems obvious that the calibrations are rather event specific, thus valid for e.g. for single bloom event. As an example for multivariate calibration, during a mesocosm experiment in SW coast of Finland, a bloom of Eutreptiella gymnastica was followed. It was obvious that PLS method was more accurate than CLS or PCR for predicting the phytoplankton biomass in the different chemotaxonomic groups (Figure 4). More important, it seems that the biomass levels as well as the chemotaxonomic composition of the phytoplankton were predicted at the level needed for bloom detection.