A Bayesian based quasianalytical algorithm for retrieval of inherent optical properties from ocean color

Abstract

The recently developed quasi-analytical algorithm (QAA) is a promising algorithm for deriving inherent optical properties from ocean color. Unlike the conventional semi-analytical algorithm, QAA does not need a priori knowledge of the spectral shape of chlorophyll absorption. However, several empirical relations, which may not be universally applicable and can result in low noise tolerance, are involved in QAA. In this study, the Bayesian inversion theory is introduced to improve the performance of QAA. In the estimation of total absorption coefficient at the reference wavelength, instead of empirical algorithms used in the QAA, the Bayesian approach is employed in combination with an optical model that uses separate parameters to account explicitly for the contribution of molecular and particle scatterings to remote sensing reflectance, a priori knowledge produced by the QAA, the Akaike's Bayesian information criterion (ABIC) for choosing, the optimal regularization parameter, and genetic algorithms for global optimization. Coefficients at other wavelengths are then derived using an empirical estimate of particle backscattering spectral shape. When applied to a simulated dataset synthesized by IOCCC the Bayesian algorithm outperforms QAA algorithm, especially in higher chlorophyll concentration waters. The root mean square errors (RMSEs) between the true and the derived a(440) and b(b)(440) are reduced from 0.918 and 0.039 m(-1) for QAA-555 to 0.367 and 0.023 m(-1) for Bayes-555, 0.205 and 0.007 m(-1) for QAA-640 to 0.092 and 0.005 m(-1) for Bayes-640, and 0.207 and 0.007 m(-1) for QAA-blending to 0.096 and 0.005 m(-1) for Bayes-blending. Results of noise sensitivity analysis show that the Bayesian algorithm is more robust than QAA.