Deriving optical metrics of coastal phytoplankton biomass from ocean colour

An approach to develop accurate local models for the estimation of chlorophyll a concentration (Chl a) and spectral phytoplankton absorption (a(ph)(lambda)) from hyperspectral in situ measurements of remote sensing reflectance (R-rs(lambda)) in an optically complex water body is presented. The models are based on empirical orthogonal function (EOF) analysis of integral-normalised R-rs(lambda) spectra, and spectral normalisation was found to be key to the models' success. Accurate model estimates of both Chl a and a(ph)(lambda) were obtained, with le values in logic, space (N= 42) of 0.839 found for Chl a, and for a(ph),(lambda). R-2 values ranging from 0.771 (547 nm) to 0.910 (655 nm). A statistical resampling exercise to create training and test data sets showed that stable models could be built with similar to 15 training spectra and corresponding measurements of Chl a and a(ph)(lambda), providing important guidance for the implementation of this approach at other locations. The applicability of the models to a reduced-wavelength resolution (8 wavebands) dataset was tested, and showed that reduction in wavelength resolution had little impact on the models' skill, with R-2 values obtained within similar to 1% of the hyperspectral (101 wavelengths) R-2 values for both Chl a and a(ph)(lambda). That the reduced-wavelength resolution models performed as well as the hyperspectral models points to their potential utility for satellite sensors. (C) 2011 Elsevier Inc. All rights reserved.