A statistical model of marine reflectance

摘要

Based on the analysis of an extensive bio-optical data set, i.e., the NOMAD dataset, the simultaneous stochastic behavior of the marine reflectance and chlorophyll concentration is characterized using nonparametric techniques. A statistical model of the conditional distribution of the marine reflectance given the chlorophyll concentration is proposed, that takes into account the natural correlations between the various optical variables. The model can be used to simulate realistic marine reflectance spectra for a given chlorophyll content, and to define prior distributions for atmospheric correction of satellite ocean-color imagery. It may also help to define bio-optical algorithms for chlorophyll concentration that minimize the influence of phytoplankton type. Conversely, considering a nonparametric regression model to retrieve chlorophyll concentration from marine reflectance leads to an improvement of about 10% on the average relative error over the polynomial OC4v4 algorithm. The prediction error of the nonparametric model provides a lower bound on the possible accuracy of chlorophyll concentration retrievals from in-situ marine reflectance, i.e., 49.2%.