Adaptive semianalytical inversion of ocean color radiometry in optically complex waters

Abstract

To address the challenges of the parameterization of ocean color inversion algorithms in optically complex waters, we present an adaptive implementation of the linear matrix inversion method (LMI) [J. Geophys. Res.101, 16631 (1996)], which iterates over a limited number of model parameter sets to account for naturally occurring spatial or temporal variability in inherent optical properties (IOPs) and concentration specific IOPs (SIOPs). LMI was applied to a simulated reflectance dataset for spectral bands representing measured water properties of a macrotidal embayment characterized by a large variability in the shape and amplitude factors controlling the IOP spectra. We compare the inversion results for the single-model parameter implementation to the adaptive parameterization of LMI for the retrieval of bulk IOPs, the IOPs apportioned to the optically active constituents, and the concentrations of the optically active constituents. We found that ocean color inversion with LMI is significantly sensitive to the a priori selection of the empirical parameters g0 and g1 of the equations relating the above-surface remote-sensing reflectance to the IOPs in the water column [J. Geophys. Res.93, 10909 (1988)]. When assuming the values proposed for open-ocean applications for g0 and g1 [J. Geophys. Res.93, 10909 (1988)], the accuracy of the retrieved IOPs, and concentrations was substantially lower than that retrieved with the parameterization developed for coastal waters [Appl. Opt.38, 3831 (1999)] because the optically complex waters analyzed in this study were dominated by particulate and dissolved matter. The adaptive parameterization of LMI yielded consistently more accurate inversion results than the single fixed SIOP model parameterizations of LMI. The adaptive implementation of LMI led to an improvement in the accuracy of apportioned IOPs and concentrations, particularly for the phytoplankton-related quantities. The adaptive parameterization encompassing wider IOP ranges were more accurate for the retrieval of bulk IOPs, apportioned IOPs, and concentration of optically active constituents.

Generalized ocean color inversion model for retrieving marine inherent optical properties

“…With the exception ofâ ϕ λ , median percent differences (MPD; the caption for Table 2 provides its calculation) for NOMAD showed greater overall variability than for the IOCCG data set, with values ranging largely from 20%-40% compared to 10%-30%. Both data sets showed less variability forb bp λ andâ λ than for the component absorption products,â dg λ andâ ϕ λ [38,39]. As demonstrated by their median ratios,b bp λ andâ λ for NOMAD showed high and low biases (over and underestimates), respectively, while both approached unity for the IOCCG data set.…”

“…Choices made for the inversion itself also impacted the retrieved eigenvalues, a discussion of which infrequently appears in previous literature (Table 5) [9,39]. Using a linear matrix inversion method in lieu of the nonlinear LM optimization resulted in departures from GIOP-DC of only 2%-7% (MPD), with comparable ΔIOPs.…”

“…produced C a with little bias (panel C), but a ϕ 443 with a definitive slope greater than unity (panel D). Such variability in derived optical and biogeochemical products calls into question the ability of existing approaches to universally estimate G λ and relate a ϕ λ to C a within the SAA paradigm [39,44]. With regard to G λ , the authors acknowledge the limitations in their approaches.…”

“…Ensemble (or, bootstrapping) approaches provide another vehicle to allow for spatial and temporal variations in optical parameters [9,39]. The approach of Wang et al [9], for example, successively iterates on a single satellite pixel using ranges of eigenvectors (e.g., S dg from 0.01 to 0.02 nm −1 , and S bp from −2 to 0 at assigned intervals) and outputs the median eigenvalues (and their standard deviations) for all valid retrievals.…”

A semianalytical ocean color inversion algorithm with explicit water column depth and substrate reflectance parameterization

“…Regarding optically shallow waters, these can be characterized as zones in which light reflected from the seafloor influences the waterleaving radiance signal [Lee et al, 1998] thereby confounding contemporary ocean color algorithms developed for optically deep waters [Cannizzaro and Carder, 2006;Qin et al, 2007;Zhao et al, 2013] (see Appendix A for further discussion). Whilst a range of ocean color algorithms have been developed and proven effective within optically complex waters [Doerffer and Schiller, 2007;Smyth et al, 2006;Werdell et al, 2013a], only a few approaches for optically shallow waters have been published [Barnes et al, 2014Brando et al, 2012] with none in operation that explicitly use pre-existing water column depth and benthic albedo data sets to improve IOP retrievals.…”