MEPS 370:69-85 (2008)  -  DOI: https://doi.org/10.3354/meps07647

ABSTRACT: Multivariate calibration methods were compared for the retrieval of phytoplankton biomass in different taxonomic groups from the spectral fluorescence signal of a living phytoplankton community. During a mesocosm experiment in the northern Gulf of Finland (Baltic Sea), the natural phytoplankton community was manipulated by additions of nutrients and fish. The changes in phytoplankton biomass and species structure were followed using traditional microscopic methods and examination of the spectral fluorescence of living cells. Based on the pigmentation of photosystem II, phytoplankton were divided into 4 groups: (1) cyanobacteria with phycobilins, (2) cryptophytes with phycobilins, chlorophyll a (chl a), chlorophyll c (chl c) and carotenoids, (3) chromophytes with chl a, chl c and carotenoids, and (4) chlorophytes with chl a and chlorophyll b (chl b) and a small amount of carotenoids. The phytoplankton biomass in these groups was predicted from the spectral fluorescence signal using classical least squares, principal component regression, and partial least squares (PLS) regression. The prediction ability of the models was compared using the root mean square error of prediction during full cross validation, partial cross validation and external validation. Regarding relevancy for the operational monitoring of phytoplankton community dynamics using spectral data, the PLS model gave the closest predictions for all taxonomic groups and with the accuracy needed for phytoplankton bloom detection.

KEY WORDS: Multivariate calibration · Partial least squares · Spectral fluorescence signal · Phytoplankton pigments · Baltic Sea