Retrieval of pigment concentrations from the absorption spectra of phytoplankton by multilayered perceptrons
用多层感知器模型由吸收光谱反演浮游植物色素

The absorption spectra of phytoplankton have gradually become an accessible parameter from hyperspectral ocean color. The retrievals of main pigment concentrations from absorption spectra of phytoplankton by multilayered perceptrons were performed. The input parameters for multilayered perceptions were phytoplankton absorption spectra, and the retrieved parameters were Chlorophyll a, b, c, and photosynthetic and non-photosynthetic carotenoids, respectively. The results indicated that Chlorophyll a and c were the most accurate retrieval pigments with average relative deviation of 19.06% and 15.90%, respectively, for the test data subsets, while the poorest performances were found for photosynthetic and non-photosynthetic carotenoids, with average relative deviation of 37.62% and 36.96%, respectively, for the test subsets. Chlorophyll b was retrieved with average relative deviation of 27.47% for the test subsets. The comparable performances of multilayer perceptrons for the training subset and the test subsets suggest that they could be successfully operated on retrieval of main pigments for the waters around the Dangan Islands in the Pearl River Estuary. The multilayer perceptrons of retrieving main pigment from phytoplankton absorption spectra also provide an important way to detect phytoplankton community from hyperspectral ocean color.