In this section, we briefly review the SeaWiFS atmospheric correction and aerosol retrieval algorithms, the SeaWiFS data processing procedure, as well as the SeaWiFS aerosol and ocean color products that are used in this study.
2.1. Theoretical Bases
 At the satellite altitude, the sensor-measured radiance at a given wavelength for the ocean-atmosphere system can be written as a linear sum from various contributions:
where Lr(λ), LA(λ), Lwc(λ), Lg(λ), and Lw(λ), are the radiance contributions at the top of the atmosphere (TOA) from air molecules, aerosols and Rayleigh-aerosol interactions (i.e., La(λ) + Lra(λ)) [Gordon and Wang, 1994a], whitecaps [Gordon and Wang, 1994b; Frouin et al., 1996; Moore et al., 2000], Sun glint [Wang and Bailey, 2001], and ocean waters, respectively. T(λ) and t(λ) are the atmospheric direct and diffuse transmittance [Yang and Gordon, 1997] at the sensor viewing direction, respectively. The water-leaving radiance Lw(λ), which can be related to the ocean near-surface physical and bio-optical properties, is the desired quantity in the atmospheric correction for the ocean color remote sensing [Gordon and Wang, 1994a; Fukushima et al., 1998; Antoine and Morel, 1999]. To remove the atmospheric effects in the derived Lw(λ), SeaWiFS produces the normalized water-leaving radiance, [Lw(λ)]N, which is defined as
where θ0 is the solar zenith angle and t0(λ) is the atmospheric diffuse transmittance at the solar direction. The two-band ratio values of [Lw(λ)]N can then be used to derive the ocean chlorophyll-a concentration [Gordon et al., 1988; Morel, 1988; O'Reilly et al., 1998].
 In equation (1), the Rayleigh scattering radiance Lr(λ) is computed using the vector radiative transfer theory (accounting for polarization) for a Rayleigh-scattering atmosphere overlying a rough (wind speed dependent) Fresnel-reflecting ocean surface [Gordon and Wang, 1992; Wang, 2002]. The whitecap radiance Lwc(λ) is modeled with input of the sea surface wind speed [Gordon and Wang, 1994b; Wang, 2000]. The Sun glint Lg(λ) is mostly avoided by titling sensor 20° away from the nadir at subsolar point and residual contamination is corrected [Wang and Bailey, 2001]. Thus, to derive the ocean contribution Lw(λ) in equation (1) (ocean color products), the aerosol effects LA(λ) at the visible wavelengths need to be estimated. However, from aerosol contribution LA(λ), the aerosol optical properties can be retrieved. Therefore aerosol products are by-products from the atmospheric correction of the ocean color remote sensors [Gordon and Wang, 1994a].
 By using a set of candidate aerosol models, effects of the spectral variation of LA(λ) at the SeaWiFS two NIR bands centered at 765 and 865 nm can be evaluated from equation (1) because Lw(λ) at the NIR bands are usually negligible for the open ocean waters due to strong water absorption [Hale and Querry, 1973; Smith and Baker, 1981]. For productive ocean waters (with high chlorophyll concentration), however, the NIR Lw(λ) values are no longer negligible and can be estimated using a bio-optical model [Siegel et al., 2000; Stumpf et al., 2003]. Thus SeaWiFS has accounted for the ocean contributions at the NIR bands for the productive ocean waters. This is particularly important for the coastal regions where the NIR Lw(λ) values are usually significant [Siegel et al., 2000; Stumpf et al., 2003]. The LA(λ) values that are derived in the SeaWiFS NIR bands, i.e.,
are then used to select two most appropriate aerosol models from a suite candidate aerosol models. A weight that is best fit to the measured NIR radiances from the radiances computed using the two selected aerosol models is estimated. Using the two aerosol models with a weight and the SeaWiFS-measured radiance, the AOT at 865 nm τa(865) (or τa(λ) for all the SeaWiFS wavelengths) and Ångström exponent from wavelengths 510 and 865 nm α(510) can then be retrieved [Gordon and Wang, 1994a; Wang, 2000]. The Ångström exponent α(λ) is defined as
where τa(λ) is the AOT at the wavelength λ. Therefore SeaWiFS algorithm uses aerosol spectral information from two-band measurements to derive appropriate aerosol models and aerosol optical properties [Gordon and Wang, 1994a]. Two-band approach in deriving aerosol optical properties is also discussed by other investigators [e.g., Nakajima and Higurashi, 1998; Mishchenko et al., 1999].
 For the atmospheric correction, the LA(λ) radiances derived in the SeaWiFS NIR bands are extrapolated into the visible wavelengths. This can be achieved using the derived aerosol models and the AOT τa(865) through the radiative transfer computations. The normalized water-leaving radiances [Lw(λ)]N at the visible wavelengths are then retrieved through equations (1) and (2), and the chlorophyll-a concentration Chl-a is obtained using the two-band ratio value of the derived [Lw(λ)]N [O'Reilly et al., 1998]. This is a typical two-step atmospheric correction procedure in which the atmosphere and ocean are assumed to be de-coupled [Gordon and Wang, 1994a; Gordon, 1997; Fukushima et al., 1998; Antoine and Morel, 1999].
2.2. Aerosol Models for the SeaWiFS Lookup Tables
 A set of realistic aerosol models is needed for the atmospheric correction and aerosol retrievals. The current SeaWiFS algorithm uses 12 aerosol models for generating the aerosol lookup tables [Wang, 2000]. They are the oceanic model with the relative humidity (RH) of 99% (denoted as O99), the maritime model with RH of 50%, 70%, 90%, and 99% (denoted as M50, M70, M90, and M99), the coastal model with RH of 50%, 70%, 90%, and 99% (denoted as C50, C70, C90, and C99), and the tropospheric model with RH of 50%, 90%, and 99% (denoted as T50, T90, and T99). The oceanic, maritime, and tropospheric models are from Shettle and Fenn  and also provided in the work of d'Almeida et al. , whereas the coastal model was introduced in the work of Gordon and Wang [1994a]. In these 12 aerosol models, the single-scattering albedo at 865 nm varies from 0.930 for the T50 model to 1.0 for the O99 model, while the Ångström exponent α(510) changes from −0.087 for the O99 model to 1.53 for the T50 model. Figure 1 is a re-plot of results from Knobelspiesse et al.  comparing the Ångström exponents from the SeaWiFS 12 aerosol models with those of the ground-based measurements. The Ångström exponent values for the SeaWiFS 12 aerosol models are represented as vertical dashed lines that are compared with the histograms from the ground-based data collected by the SIMBIOS project (see discussions also in section 3). Clearly, the in situ Ångström exponents obtained from maritime environment can be well represented with the SeaWiFS 12 aerosol models most of the time. For aerosols with very large Ångström exponents, e.g., >1.5, the current SeaWiFS models may not be representative. Therefore the SeaWiFS aerosol models represent mostly the nonabsorbing and weakly absorbing aerosols that are usually present in the ocean maritime environment.
2.3. SeaWiFS Data Processing and Products
 Since September of 1997, SeaWiFS routinely provides ocean color and atmospheric products, e.g., the normalized water-leaving radiance [Lw(λ)]N for six visible wavelengths (412, 443, 490, 510, 555, and 670 nm), chlorophyll-a concentration Chl-a, AOT at 865 nm τa(865), and the aerosol Ångström exponent derived from the wavelength 510 and 865 nm α(510). SeaWiFS has gone through four major reprocessings of the entire data set. Each reprocessing has addressed the data quality issues that are related to the sensor calibration, instrument navigation, data masks and flags, and retrieval algorithms. The SeaWiFS data used in here are from the fourth data reprocessing which was carried out in July 2002 [Patt et al., 2003]. Currently, the SeaWiFS data processing is optimized for the ocean color measurements (e.g., chlorophyll-a concentration). Very high AOT data such as the dust and smoke plumes are usually masked out due to large uncertainties in the ocean color products with these cases [Gordon, 1997]. The SeaWiFS has a reflectance threshold at 865 nm corresponding to the AOT of ∼0.3. Thus the current SeaWiFS aerosol products are mostly applicable and valid in the open ocean regions where the marine aerosols are often the dominant sources.
 It is noted that, however, the SeaWiFS measurements can be used to derive the aerosol optical properties for very thick aerosol layers, e.g., for the dust study [Husar et al., 2001; Moulin et al., 2001a], and possibly for the ocean color data in the regional case study [Moulin et al., 2001b]. Implementation is planned for an aerosol retrieval scheme in which very high AOT (e.g., dust and smoke) can still be retrieved even though the ocean color data may have large uncertainties. In these cases, flags can be applied to the ocean color products.