Remote estimation of phytoplankton size fractions using the spectral shape of light absorption

Abstract

Phytoplankton size structure plays an important role in ocean biogeochemical processes. The light absorption spectra of phytoplankton provide a great potential for retrieving phytoplankton size structure because of the strong dependence on the packaging effect caused by phytoplankton cell size and on different pigment compositions related to phytoplankton taxonomy. In this study, we investigated the variability in light absorption spectra of phytoplankton in relation to the size structure. Based on this, a new approach was proposed for estimating phytoplankton size fractions. Our approach use the spectral shape of the normalized phytoplankton absorption coefficient (a(ph)(lambda)) through principal component analysis (PCA). Values of a(ph)(lambda) were normalized to remove biomass effects, and PCA was conducted to separate the spectral variance of normalized a(ph)(lambda) into uncorrelated principal components (PCs). Spectral variations captured by the first four PC modes were used to build relationships with phytoplankton size fractions. The results showed that PCA had powerful ability to capture spectral variations in normalized a(ph)(lambda), which were significantly related to phytoplankton size fractions. For both hyperspectral a(ph)(lambda) and multiband a(ph)(lambda), our approach is applicable. We evaluated our approach using wide in situ data collected from coastal waters and the global ocean, and the results demonstrated a good and robust performance in estimating phytoplankton size fractions in various regions. The model performance was further evaluated by a(ph)(lambda) derived from in situ remote sensing reflectance (R-rs(lambda)) with a quasi-analytical algorithm. Using R-rs(lambda) only at six bands, accurate estimations of phytoplankton size fractions were obtained, with R-2 values of 0.85, 0.61, and 0.76, and root mean-square errors of 0.130, 0.126, and 0.112 for micro-, nano-, and picophytoplankton, respectively. Our approach provides practical basis for remote estimation of phytoplankton size structure using a(ph)(lambda) derived from satellite observations or rapid field instrument measurements in the future. (C)2015 Optical Society of America