Performance of two semi-analytical algorithms in deriving water inherent optical properties in the Southern Ocean.

Remotely sensed inherent optical properties (IOPs) are key proxies for synoptic mapping of primary production and carbon export in the global ocean. However, the IOPs inversion algorithms are scarcely evaluated in the Southern Ocean (SO) because of limited field observations. In this study, the performance of two widely used semi-analytical algorithms (SAAs), i.e., the quasi-analytical algorithm (QAA) and the generalized IOP model (GIOP), were evaluated using a compiled in situ bio-optical dataset in SO, as well as measurements from the Visible Infrared Imaging Radiometer Suite (VIIRS). Evaluations with in situ data show that QAA and GIOP have comparable performance in retrieving the total absorption coefficient (a(λ)), absorption coefficients of phytoplankton (aph(λ)), and that of detritus and colored dissolved organic matter (adg(λ)). Overall, it was found that remotely sensed a(λ) and aph(λ) by both SAAs agreed well with field measurements, with the mean absolute percentage difference (MAPD) of derived a(λ) and aph(λ) in the blue-green bands being ∼20% and ∼40%, respectively. However, derived adg(λ) by both SAAs were higher than the measured values at the lower end (adg(443) < ∼0.01 m-1), but lower at the higher end (adg(443) > ∼0.02 m-1), with MAPD of ∼60%. Results of this effort suggest confident products of a(λ) and aph(λ) from VIIRS in SO, but more dedicated efforts on the measurements and evaluation of adg(λ) in SO would be desired.

Evaluation of forward reflectance models and empirical algorithms for chlorophyll concentration of stratified waters.

For waters with stratified chlorophyll concentration (Chl), numerical simulations were carried out to gain insight into the forward models of subsurface reflectance and empirical algorithms for Chl from the ocean color. It is found that the Gordon and Clark (1980) forward model for reflectance using an equivalent homogeneous water with a weighted average Chl (⟨Chl⟩) as the input works well, but depending on the contribution of gelbstoff, the difference in reflectance between stratified and the equivalent homogeneous water can be more than 10%. Further, the attenuation of upward light is better approximated as ∼1.5times that of the diffuse attenuation coefficient of downwelling irradiance. On the other hand, although the forward model for reflectance developed in Zaneveld et al. [Opt. Express13, 9052 (2005)] using equivalent homogeneous water with a weighted average of the backscattering to absorption ratio as the input also works well, this model cannot be used to obtain equivalent ⟨Chl⟩ for reflectance. Further, for empirical Chl algorithms designed for "Case 1" waters, it has been discovered that, for surface Chl in a range of ∼0.06-22.0mg/m3, the predictability of surface Chl is basically the same as that of ⟨Chl⟩ from the blue-green band ratio or the band difference of reflectance. Because ⟨Chl⟩ is wavelength and weighting-formula dependent, and it is required to have profiles of both Chl and the optical properties, these results emphasize that for empirical Chl algorithms, it is easier, less ambiguous, and certainly more straightforward and simple to use surface Chl for algorithm development and then its evaluation, rather than to use ⟨Chl⟩, regardless of whether or not the water is stratified.

Impacts of pure seawater absorption coefficient on remotely sensed inherent optical properties in oligotrophic waters.

The spectral absorption coefficient of pure seawater (aw(λ)) in published studies differ significantly in the blue domain, yet the impacts of such discrepancies on the inherent optical properties (IOPs) derived from ocean color have been scarcely documented. In this study, we confirm that changes in aw(λ) may have significant impacts on retrieved IOPs in oligotrophic waters, especially for the phytoplankton absorption coefficient (aph(λ)). Two sets of aw(λ) data, aw_PF97 (Appl. Opt. 36, 8710, 1997) and aw_Lee15 (Appl. Opt. 54, 546, 2015), were selected for optical inversion analysis. It is found that aph(λ) retrieved with aw_Lee15 agree better with the in-situ measurements in oligotrophic waters. Further applications to satellite images show that the derived aph(λ) using aw_Lee15 can be up to 238% higher than the retrievals using aw_PF97 in the core zone of the subtropical ocean gyres. Given that aw_PF97 is commonly accepted as the "standard" aw(λ) by the ocean color community in the past decades, this study highlights the need and importance to update aw(λ) with aw_Lee15 for IOPs retrievals in oligotrophic waters.

Deriving inherent optical properties from classical water color measurements: Forel-Ule index and Secchi disk depth.

Secchi disk depth (ZSD) and Forel-Ule index (FUI) are the two oldest and easiest measurements of water optical properties based on visual determination. With an overarching objective to obtain water inherent optical properties (IOPs) using these historical measurements, this study presents a model for associating remote-sensing reflectance (Rrs) with FUI and ZSD. Based upon this, a scheme (FZ2ab) for converting FUI and ZSD to absorption (a) and backscattering coefficients (bb) is developed and evaluated. For a data set from HydroLight simulations, the difference is <11% between FZ2ab-derived a and known a, and <28% between FZ2ab-derived bb and known bb. Further, for a data set from field measurements, the difference is < 30% between FZ2ab-derived a and measured a. These results indicate that FZ2ab can bridge the gap between historical measurements and the focus of IOP measurements in modern marine optics, and potentially extend our knowledge on the bio-optical properties of global seas to the past century through the historical measurements of FUI and ZSD.

Progressive scheme for blending empirical ocean color retrievals of absorption coefficient and chlorophyll concentration from open oceans to highly turbid waters.

To achieve a smooth transition between algorithms for "clear" water and "turbid" water, we propose a single formula to calculate the input parameter (ip) used for empirical retrieval of absorption coefficients (a) or chlorophyll concentration ([Chl]) from remote-sensing reflectance (Rrs). This formula for ip takes the ratio of the maximum Rrs in the blue-green bands to the sum of Rrs(green) and the scaled Rrs in the red and infrared bands (termed as ipMax-Sum). We found that, compared to the widely used OC4-type formula for ip, ipMax-Sum can improve the coefficient of determination from ∼0.88 to 0.99 for absorption coefficient at 440 nm [a(440)] in ∼0.01-20.0 m-1 ([Chl] ∼0.01-500 mg m-3). Especially, the sensitivity of ipMax-Sum to the change in a(440) is about five times greater than that of OC4-type for a(440)>∼1.0 m-1 ([Chl]>∼10 mg m-3). These results indicate an advantage of ipMax-Sum for generating robust and seamless a(440) or [Chl] from clear to highly turbid waters. The inclusion of such a scheme in a quasi-analytical algorithm is also presented.

Enhance field water-color measurements with a Secchi disk and its implication for fusion of active and passive ocean-color remote sensing.

Inversion of the total absorption (a) and backscattering coefficients of bulk water through a fusion of remote sensing reflectance (Rrs) and Secchi disk depth (ZSD) is developed. An application of such a system to a synthesized wide-range dataset shows a reduction of ∼3 folds in the uncertainties of inverted a(λ) (in a range of ∼0.01-6.8 m-1) from Rrs(λ) for the 350-560 nm range. Such a fusion is further proposed to process concurrent active (ocean LiDAR) and passive (ocean-color) measurements, which can lead to nearly "exact" analytical inversion of an Rrs spectrum. With such a fusion, it is found that the uncertainty in the inverted total a in the 350-560 nm range could be reduced to ∼2% for the synthesized data, which can thus significantly improve the derivation of a coefficients of other varying components. Although the inclusion of ZSD places an extra constraint in the inversion of Rrs, no apparent improvement over the quasi-analytical algorithm (QAA) was found when the fusion of ZSD and Rrs was applied to a field dataset, which calls for more accurate determination of the absorption coefficients from water samples.

Secchi disk observation with spectral-selective glasses in blue and green waters.

Radiative transfer modeling of Secchi disk observations has historically been based on conjugated signals of eye response and radiance, where water's attenuation in the entire visible band is included in the attenuation when deciding the Secchi disk depth in water. Aas et al. [Ocean Sci.10(2), 177 (2014)Remote Sens. Environ.169, 139 (2015)] hypothesized that it is actually the attenuation in water's transparent window that matters to the observation of a Secchi disk in water. To test this hypothesis, observations of Secchi disks in blue and green waters were conducted via naked eyes, blue-pass glasses, and green-pass glasses. Measurement results indicate that for blue waters, the observed Secchi depths via naked eyes match the depths obtained with blue-pass glasses and much deeper than the depths with green-pass glasses, although the green-pass glasses match the highest response of human eyes. These observations experimentally support the hypothesis that our eye-brain system uses the contrast information in the transparent window to make a judgement decision regarding sighting a Secchi disk in water.

On the modeling of hyperspectral remote-sensing reflectance of high-sediment-load waters in the visible to shortwave-infrared domain.

We evaluated three key components in modeling hyperspectral remote-sensing reflectance in the visible to shortwave-infrared (Vis-SWIR) domain of high-sediment-load (HSL) waters, which are the relationship between remote-sensing reflectance (R(rs)) and inherent optical properties (IOPs), the absorption coefficient spectrum of pure water (a(w)) in the IR-SWIR region, and the spectral variation of sediment absorption coefficient (a(sed)). Results from this study indicate that it is necessary to use a more generalized R(rs)-IOP model to describe the spectral variation of R(rs) of HSL waters from Vis to SWIR; otherwise it may result in a spectrally distorted R(rs) spectrum if a constant model parameter is used. For hyperspectral a(w) in the IR-SWIR domain, the values reported in Kou et al. (1993) provided a much better match with the spectral variation of R(rs) in this spectral range compared to that of Segelstein (1981). For a(sed) spectrum, an empirical a(sed) spectral shape derived from sample measurements is found working much better than the traditional exponential-decay function of wavelength in modeling the spectral variation of R(rs) in the visible domain. These results would improve our understanding of the spectral signatures of R(rs) of HSL waters in the Vis-SWIR domain and subsequently improve the retrieval of IOPs from ocean color remote sensing, which could further help the estimation of sediment loading of such waters. Limitations in estimating chlorophyll concentration in such waters are also discussed.

Spectral interdependence of remote-sensing reflectance and its implications on the design of ocean color satellite sensors.

Using 901 remote-sensing reflectance spectra (R(rs)(λ), sr⁻¹, λ from 400 to 700 nm with a 5 nm resolution), we evaluated the correlations of R(rs)(λ) between neighboring spectral bands in order to characterize (1) the spectral interdependence of R(rs)(λ) at different bands and (2) to what extent hyperspectral R(rs)(λ) can be reconstructed from multiband measurements. The 901 R(rs) spectra were measured over a wide variety of aquatic environments in which water color varied from oceanic blue to coastal green or brown, with chlorophyll-a concentrations ranging from ~0.02 to >100 mg m⁻³, bottom depths from ~1 m to >1000 m, and bottom substrates including sand, coral reef, and seagrass. The correlation coefficient of R(rs)(λ) between neighboring bands at center wavelengths λ(k) and λ(l), r(Δλ)(λ(k), λ(l)), was evaluated systematically, with the spectral gap (Δλ=λ(l)-λ(k)) changing between 5, 10, 15, 20, 25, and 30 nm, respectively. It was found that r(Δλ) decreased with increasing Δλ, but remained >0.97 for Δλ≤20 nm for all spectral bands. Further, using 15 spectral bands between 400 and 710 nm, we reconstructed, via multivariant linear regression, hyperspectral R(rs)(λ) (from 400 to 700 nm with a 5 nm resolution). The percentage difference between measured and reconstructed R(rs) for each band in the 400-700 nm range was generally less than 1%, with a correlation coefficient close to 1.0. The mean absolute error between measured and reconstructed R(rs) was about 0.00002 sr⁻¹ for each band, which is significantly smaller than the R(rs) uncertainties from all past and current ocean color satellite radiometric products. These results echo findings of earlier studies that R(rs) measurements at ~15 spectral bands in the visible domain can provide nearly identical spectral information as with hyperspectral (contiguous bands at 5 nm spectral resolution) measurements. Such results provide insights for data storage and handling of large volume hyperspectral data as well as for the design of future ocean color satellite sensors.

Impact of computational methods and spectral models on the retrieval of optical properties via spectral optimization.

Spectral optimization algorithm (SOA) is a well-accepted scheme for the retrieval of water constituents from the measurement of ocean color radiometry. It defines an error function between the input and output remote sensing reflectance spectrum, with the latter modeled with a few variables that represent the optically active properties, while the variables are solved numerically by minimizing the error function. In this paper, with data from numerical simulations and field measurements as input, we evaluate four computational methods for minimization (optimization) for their efficiency and accuracy on solutions, and illustrate impact of bio-optical models on the retrievals. The four optimization routines are the Levenberg-Marquardt (LM), the Generalized Reduced Gradient (GRG), the Downhill Simplex Method (Amoeba), and the Simulated Annealing-Downhill Simplex (i.e. SA + Amoeba, hereafter abbreviated as SAA). The Garver-Siegel-Maritorena SOA model is used as a base to test these computational methods. It is observed that 1) LM is the fastest method, but SAA has the largest number of valid retrievals; 2) the quality of final solutions are strongly influenced by the forms of spectral models (or eigen functions); and 3) dynamically-varying eigen functions are necessary to obtain smaller errors for both reflectance spectrum and retrievals. Results of this study provide helpful guidance for the selection of a computational method and spectral models if an SOA scheme is to be used to process ocean color images.

Long-term trend of Ulva prolifera blooms in the western Yellow Sea.

Blooms of the green macroalga Ulva prolifera in the western Yellow Sea occurred every year since 2008, and they have been reported and studied extensively using a variety of means including remote sensing. However, to date, long-term bloom patterns have not been reported except for a few case studies showing examples in different years. Here, using MODIS observations and an objective method to perform statistical analysis, mean Ulva coverage in the western Yellow Sea has been derived and analyzed between 2007 and 2015 at both monthly and annual scales. On annual scale, mean Ulva coverage decreased after 2008, but increased rapidly after 2012 from 8km2 in 2012 to 116km2 in 2015 (the largest ever reported in history for this region). In the month of June the mean coverage increased from 18km2 in 2012 to 363km2 in 2015. Other than 2009 and 2010, the month of June showed maximum Ulva coverage in every year. These coverage estimates are significantly lower than previously reported values as they represent "pure" algae coverage after taking into account of partial pixel coverage. Several environmental factors were examined in an attempt to determine the reasons behind such long-term changes, yet the results are inconclusive, suggesting a strong necessity of further coordinated and multi-disciplinary researches.