Radiative forcing by aerosol is one of the greatest sources of uncertainty in climate change modeling. One of the main objectives of the Asian Pacific Regional Aerosol Characterization Experiment (ACE-Asia) was to assess the spatial and temporal variability of aerosol properties in northeast Asia. Aerosol optical parameters are retrieved from the Sea-Viewing Wide Field-of-View Sensor (SeaWiFS) satellite data to study aerosol characteristics over northeast Asia during the ACE-Asia intensive observation period 2001. The aerosol optical thickness (AOT) and Ångström exponent are retrieved over land and ocean at a spatial scale of 1 × 1 km2 by using the Bremen aerosol retrieval (BAER) technique. SeaWiFS-derived AOT data were compared with those ground-measured AERONET AOT. The validation between SeaWiFS and Sun photometer data showed good correlation coefficients (r > 0.89) during the ACE-Asia IOP. The SeaWiFS-retrieved AOT showed a high AOT value of around 0.8 and an Ångström exponent of 0.51 during an Asian dust case of 13 April 2001 over Gosan. Aerosol parameters retrieved from SeaWiFS data by using the BAER technique showed promising results to monitor atmospheric aerosol loading and estimate its optical properties.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Since the lifetime of atmospheric aerosols is from a few days to a few months, their spatiotemporal distribution is highly variable. The satellite remote-sensing technique has the advantage of being capable of measuring optical and physical aerosol parameters over a large area. Satellite remote sensing can also provide the spatial and spectral resolution necessary to monitor the highly variable aerosol pattern. However, aerosol properties can be retrieved from satellite data only for cloud-free scenes. In that case, the radiance received by a satellite at the top of the atmosphere (TOA) is composed of contributions due to scattering by gases and aerosols and reflection at the surface. Absorption by molecular species and aerosols further modifies the TOA radiance. Hence the signal received by an electro-optical sensor on a satellite in principle contains information about the surface properties and atmospheric constituents. Various mathematical retrieval algorithms have been used to separate them out [King et al., 1999]. Until recently, retrieval of aerosol properties from satellite data was only possible over dark surfaces, such as oceans with very low reflectivity. Over the oceans, aerosol optical thickness (AOT) has been retrieved for the past two decades, thus providing a long time record of observation data. The AOT over the ocean with measurements from the advanced very high resolution radiometer (AVHRR) [Rao et al., 1989; Ignatov et al., 1995] is a representative product for aerosol retrieval. However, new retrieval techniques utilizing the features of modern satellite sensors now allow for the accurate retrieval of aerosol properties over brighter surfaces, such as the aerosol index (AI) provided by Total Ozone Mapping Spectrometer (TOMS) data [Herman et al., 1997], Moderate-Resolution Imaging Spectroradiometer (MODIS)–retrieved aerosol products [Kaufman et al., 1997a, 1997b], and the Sea-Viewing Wide Field-of-View Sensor (SeaWiFS)–retrieved AOT [von Hoyningen-Huene et al., 2003].
 In this study, AOT retrieval using a separation technique called the Bremen aerosol retrieval (BAER) [von Hoyningen-Huene et al., 2003] has been applied to SeaWiFS data obtained during the Asian Pacific Regional Aerosol Characterization Experiment (ACE-Asia) intensive observation period (IOP), ∼1–30 April 2001. The ACE-Asia program focuses on aerosol outflow from Asia to the Pacific basin since both anthropogenic aerosols and mineral dust from the Asian continent greatly affect the atmospheric environment and radiation balance in the downwind regions [Huebert et al., 2003]. One of the main objectives of ACE-Asia was to assess the spatial and temporal variability of aerosol properties by continuous measurements at a number of ground stations and intensive in situ measurements on mobile platforms such as aircrafts and ships [Huebert et al., 2003].
 During ACE-Asia IOP, several severe Asian dust storms were observed over northeast Asia, affecting local air quality and radiation budget [Kim et al., 2004]. Therefore satellite-retrieved AOT can be a useful parameter to estimate optical and physical characteristics of aerosol in the region of interest. In order to estimate the AOT, SeaWiFS data were analyzed in this study using the BAER method. A retrieval strategy based on look-up tables (LUTs) determined using Mie theory calculations and radiative transfer code [Borhen and Hoffman, 1983] was adapted to retrieve aerosol optical parameters over both ocean and land. The results from SeaWiFS-retrieved aerosol data were validated and intercompared with those derived from the Aerosol Robotic Network (AERONET) [Holben et al., 1998] ground-based Sun photometer data.
2. Data Sets
 The SeaWiFS instrument on board the SeaStar satellite was launched on 1 August 1998. In this study, SeaWiFS level 1A (L1A) local area coverage (LAC) products, which are the radiance counts measured by a sensor, from high resolution picture transmission (HRPT) stations were used for aerosol retrieval over northeast Asia for the ACE-Asia IOP, ∼1–30 April 2001. SeaWiFS has eight bands, six in the visible (412, 443, 490, 510, 555, and 670 nm) and two in the near-infrared wavelengths (765 and 865 nm), and a tilt mechanism to avoid Sun glitter [Hooker et al., 1992]. The spatial resolution of SeaWiFS LAC data is approximately 1 × 1 km2 at nadir with a swath width of 2800 km.
 The study area of this paper, including the Korea peninsula and part of northeast China, is shown in Figure 1. The ACE-Asia Gosan supersite (126.10°E, 33.23°N, 50 m above sea level) is located 45-km southeast of Jeju city on the western tip of Jeju Island, Korea. A number of domestic and international research teams had participated in the surface aerosol measurements at Gosan during the ACE-Asia IOP [Huebert et al., 2003]. Sun-photometer-measured aerosol optical and physical parameter data from three AERONET sites (Gosan, 126.10°E, 33.23°N; Anmyun, 126.19°E, 36.31°N; and Beijing, 116.38°E, 39.98°N) during the ACE-Asia IOP are available (http://aeronet.gsfc.nasa.gov). Postcalibrated and cloud-screened level 2 AERONET data are used in this study to compare with SeaWiFS-retrieved AOT. Among the common wavelengths between AERONET instruments and SeaWiFS, aerosol optical thicknesses at 440 nm and 670 nm were selected to compare with each other.
 The TOMS AI [Herman et al., 1997] has been used to detect UV absorbing aerosol including dust and biomass-burning aerosols. The TOMS AI is defined as the difference between the observations and model calculations from a pure molecular atmosphere with the same surface reflectivity and measurement conditions [Hsu et al., 1996; Herman et al., 1997]. Generally, positive AI values reflect the presence of absorbing aerosols; the presence of nonabsorbing aerosols will suppress the AI value. However, the TOMS AI has a strong dependence on aerosol altitude, with higher values when the same amount of aerosol is at higher altitudes. Also, the TOMS AI does not detect aerosols in the lowest part (lower than ∼1.5 km) of the atmosphere since particle scattering of absorbing aerosol can dominate over the absorption of Rayleigh scattering [Hsu et al., 1999]. TOMS AI data from TOMS web site (http://toms.gsfc.nasa.gov) are used in this study to establish relationships with SeaWiFS aerosol map over ocean and land since the surface reflectance is low and nearly constant over both water and land in the UV band [Herman et al., 1997].
 The daily aerosol product [Tanré et al., 1997; Kaufman et al., 1997a, 1997b] of MODIS level 2 aerosol data sets (MOD04 L2: MODIS aerosol product, version 4.1.3) was also collected from National Aeronautics and Space Administration (NASA) Distributed Active Archive Center (DAAC) to compare with SeaWiFS-retrieved AOT obtained in this study. The MOD04 data have various aerosol physical and optical parameters with 10 × 10 km2 spatial resolution. The MOD04 contains information about atmospheric aerosols, including aerosol optical thickness at 550 nm (over oceans and land), aerosol size distribution (over ocean), and aerosol type (over land) globally. The MODIS AOT has been validated with ground-based Sun photometer AOT by a spatiotemporal approach [Ichoku et al., 2002]. It has been shown that the MODIS aerosol retrievals over land surface, except in coastal zones, are found within retrieval errors Δτa = ±0.05 ± 0.2τa [Chu et al., 2002].
 Additionally, the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT-4) [Draxler, 1992] model was used to characterize the transport pattern of Asian dust plumes over the study area. The description of the HYSPLIT-4 model can be found at the National Oceanic and Atmospheric Administration (NOAA) Air Research Laboratory (ARL) web page (http://www.arl.noaa.gov/ss/models/hysplit.html). Air mass backward trajectories calculations were made using the global gridded final run (FNL) meteorological data set produced by the U.S. National Centers for Environmental Prediction (NCEP). Vertical motions were calculated using the isentropic method with 96-hour intervals.
3. Retrieval of the Aerosol Optical Parameters From SeaWiFS Data
 The Bremen aerosol retrieval (BAER) method was used previously for aerosol retrieval from SeaWiFS data during the Lindberg Aerosol Characterization Experiment (LACE-98) [Ansmann et al., 2002]. A detailed description of the BAER technique was given by von Hoyningen-Huene et al. . The BAER algorithm was designed to be applied over both ocean and land. Six visible channels (∼412–670 nm) of SeaWiFS were used for aerosol retrieval over land, and all eight channels were used over ocean. Figure 2 shows a logical flowchart explaining the AOT retrieval method based on SeaWiFS data.
 Since the upwelling radiance at the top of the atmosphere (TOA) over dark surfaces increases with increasing AOT, there exists a relationship between TOA radiance and AOT [Durkee et al., 1986]. In our algorithm, unitless reflectance instead of radiance was used. TOA reflectance ρTOA is defined as
where L is the measured TOA radiance of the satellite, E0 is the extraterrestrial solar irradiance, and θ0 is the solar zenith angle.
 The TOA reflectance can be affected by many factors, including solar and observation geometry, Rayleigh and aerosol scattering, atmospheric transmittance, surface reflectance, etc. Therefore, in order to separate aerosol reflectance from TOA reflectance, other contributors such as Rayleigh scattering and surface reflectance should be removed from it. Aerosol reflectance ρa can be expressed as [von Hoyningen-Huene et al., 2003]
where ρRay (λ, θ, p, M0, MS) is the normalized Rayleigh path reflectance inclusive multiple scattering for scattering angle θ, pressure p, and air mass factors for illumination M0 and satellite MS and ω0 (λ) is the aerosol single-scattering albedo, τ is the total transmission for zenith distance of satellite zS from Sun z0, and ρSurf (λ, z0, zS) is the surface reflectance for Sun and satellite geometry given by ρSurf (λ, z0, zS) = ρSurf(λ) · cos(z0) · cos (zS). To calculate height-dependent Rayleigh scattering, the pressure p(z) at elevation z (km) was calculated by using the parameterized barometric equation defined as
where g is the gravity acceleration (9.807 m/s2), z is the altitude above sea level in km, and TSurf is the surface temperature, which was assumed as 298 K.
 To separate Rayleigh path reflectance from TOA reflectance over the surface, the digital elevation model (DEM) was used in each pixel. The DEM used in this study is the global 30-arc sec elevation data (GTOPO30) from the U.S. Geological Survey (USGS). The height, z km, of each pixel from DEM was used to calculate the pressure determined by the parameterized barometric equation. Then the Rayleigh path reflectance can be determined with this pressure at each wavelength in each pixel.
 In case of separation of the surface reflectance from TOA reflectance over land, the BAER algorithm contains the separation of the restriction on “dark targets” [Kaufman et al., 1997a, 1997b]. The land surface reflectance was determined by a linear mixing model of the spectral reflection of “green vegetation” and “bare soil.” The Normalized Differential Vegetation Index (NDVI) was used to estimate green vegetation cover from the aerosol-corrected reflectance of channels 6 (670 nm) and 8 (865 nm). von Hoyningen-Huene et al.  showed that determination of surface reflectance using the NDVI is a useful tool for aerosol retrieval over land surface.
 However, the NDVI derived from 670- and 865-nm reflectance is affected by the presence of aerosol. In order to correct aerosol effects in NDVI calculation, the aerosol effect was first estimated for channel 1 (412 nm) with an assumption of a black surface, which means that the surface reflectance is equal to 0. Generally, spectral surface reflectance in the short-wavelength region decreases with wavelength. Because of the decreasing reflectance with decreasing wavelength, reflectance at 412 nm gave the best results for AOT, which was then used to calculate AOT at channels 6 (670 nm) and 8 (865 nm) using the wavelength-dependent Ångström exponent. The global aerosol climatology by AERONET observation indicates the representative values of the Ångström exponent to be from 1.2 to 2.5 in urban areas and from 0.1 to 0.9 in desert areas [Dubovik et al., 2002a]. Since the typical Ångström exponent obtained from the AERONET sites over the study area was ∼1.0–1.3, it was initially assumed to be 1.0 for all aerosol cases. Then aerosol reflectance at channels 6 and 8 was calculated. These aerosol reflectances (plus Rayleigh scattering) were used to calculate the aerosol corrected NDVI. However, it is possible that the assumption of black surface can overestimate the aerosol effect. Also, the effect of the surface is reduced when the aerosol effect is relatively large, as in the severe dust storm case, since the extinction of coarse particles shows enhancement in the red channel. The result of underestimation of dust AOT over land can be explained by this reason. NDVI affected by a dense aerosol plume shows underestimated vegetation, which leads to overestimation of surface reflectance in our retrieval. Increased surface reflectance can decrease TOA reflectance in our separation technique. Thus decreased aerosol reflectance is interpreted as low AOT. A typical Ångström exponent for Asian dust was reported to be ∼0.2–0.5 [Tanaka et al., 1989] as confirmed by AERONET data (which resulted in values of around 0.5 in Table 3). Therefore, in our aerosol retrieval, α = 0.5 was assumed to be the correct NDVI for the Asian dust event day.
 Then the surface reflectance (ρλmixing) was determined by a linear mixing model with NDVI as
where ρλveg is vegetation reflectance and ρλsoil is bare soil reflectance, and
where F is a scaling factor defined as F = ρ670corr/ρ670mixing to adapt the level of the surface reflectance to that required within the satellite scene. The ρ670corr is the satellite-based TOA reflectance determined by subtracting Rayleigh scattering and atmospheric aerosol reflectance obtained for channel 1 (412 nm) under the assumption of a black surface. The scaling factor contributes much to a stabilization of the solutions and reduces the regional variability over the land surfaces caused by different surface types. After spectral aerosol reflectance is obtained using equation (2), spectral AOT is then determined from LUTs.
 Finally, the application of the constraints in an iterative procedure that minimizes the root-mean-square deviation (RMSD) of the spectral AOT should be checked so as to have smooth spectral dependence of AOT according to the Ångström power law. von Hoyningen-Huene et al.  have adapted three conditions for the iteral constraints, which are (1) the requirement that the wavelength dependence of the AOT be a smooth nonlinear function of wavelength, (2) a weighting parameter to ensure the convergence during iteration, and (3) linear mixing of the surface spectral reflectance from vegetation with that of the Earth.
 The smoothness is estimated as the RMSD, which is determined from the individual AOT (τ(λ)) and the value represented by Ångström power law, (λ).
where N is the number of channels used. If RMSD >0.01, then the weighting parameter is used to facilitate convergence for next iteration. For the ith iteration, surface reflectance is given by
where w(λ) is a weighting factor taking into account the spectral variability of the surface reflectance and the relative deviation of the AOT (di(λ) = [τ(λ) − (τ)]/τ(λ)) from the smooth behavior. The weighting factor is fixed with a value of 0.15 for 412 nm and increases to 0.30 for 670 nm. Iteration to obtain sufficient smoothness of spectral AOT continues until the convergence criteria RMSD ≤ 0.01 is met. Finally, spectral AOT is determined after iteration ends.
 For the ocean surface reflectance calculation, BAER uses seawater spectra given by the HIRES-ES radiometer [Zimmerman, 1998] measurements over the Baltic Sea. Ocean surface reflectance can be calculated by mixing this spectrum with the clean ocean water. However, highly turbid waters exist near Yangtze and Yellow river discharge in the study area, which can increase surface reflectance and exert impact on aerosol retrieval over the ocean surface. In order to separate this reflectance from TOA reflectance, turbid water spectra should be known. Rocha  used the oceanic turbidity identifier to estimate the spatial distribution and qualitative intensity of the ocean turbidity using SeaWiFS spectral reflectance. Spectral reflectance (wavelength in nm) for turbid water is plotted in Figure 3, which was obtained from SeaWiFS under relatively clear-sky conditions on 4 April 2001. The spectral reflectance of the East China Sea (Figure 3a) shows lower values at channels ∼4–8 than that of the Yellow Sea (Figure 3b) since the optical characteristics of pigment of the East China Sea differ from those of the Yellow Sea. The ocean surface reflectance was determined by a linear mixing model of clear water and turbid water. The portion of turbid water (rpig) was calculated by using a statistical method that divides the ratio of channel 6 and channel 7 by the ratio of channel 3 and channel 5:
where n is SeaWiFS channel number; ρTOAn and ρRayn (λ, θ, p(z), M0, MS) are the TOA reflectance and Rayleigh reflectance at channel n, respectively.
 The relationship between the satellite-detected aerosol reflectance and AOT can be determined by radiative transfer calculations. The satellite-detected aerosol reflectance was calculated as a function of the wavelength for Rayleigh atmosphere and for aerosol-containing atmospheres. Then the relationships between various AOT and aerosol reflectance are approximated with second-order polynomials and used to construct LUTs. The AOT is determined using this relationship between the aerosol reflectance at the TOA and the AOT. LUTs were computed using the Second Simulation of the Satellite Signal in the Solar Spectrum (6S) radiative transfer code [Vermote et al., 1997]. The 6S code can calculate the satellite signal between ∼0.25 and 4.0 μm. The 6S code uses the successive order of scattering (SOS) method to compute the scattering properties of the aerosol and Rayleigh scatters without taking account of polarization. Comparison between the scalars SOS versus the vector SOS results shows that the error is small, which can justify the use of the scalar code when molecules and aerosols are mixed. However, when only Rayleigh scattering is considered, polarization is taken into account through empirically adjusted coefficients [Su et al., 2002]. A Rayleigh depolarization factor of 0.0279 [Young, 1980] has been used (see also http://modis.gsfc.nasa.gov/data/atbd_mod08.pdf; Vermote and Vermeulen ; MODIS reflectance algorithm).
 The first step is to select aerosol size distribution to be used as input to radiative transfer model. Bimodal lognormal distribution aerosol size distribution was used in this study as in Table 1. Mie scattering calculation was performed for an aerosol size distribution expressed as
where n is the mode number of particles and Vn, rmm, and σn are the volume concentration, median radius, and standard deviation of nth mode, respectively. Table 1 shows the parameters of aerosol size distribution averaged from three different AERONET sites during the ACE-Asia Asian dust storm period. For Asian dust cases, aerosol nonsphericity can cause changes in phase function signature. Dubovik et al. [2002a, 2002b] showed that the use of spherical particle phase functions had a strong tendency to produce the retrieval artifacts when dust was present during Sun photometer measurement. Therefore spheroid volume size distribution data were used in our retrieval method in order to minimize the effects of nonsphericity of aerosol. Total averaged values are rfm = 0.12 μm, rcm = 2.56 μm, σf = 0.49, and σc = 0.64, respectively. The median radius of the fine mode, consisting mostly of pollution particles, was 0.12 μm, and that of the coarse mode, consisting of dust particles, was 2.56 μm. Although the complex refractive index of Asian dust particles is not known clearly, the refractive index of dust particles was assumed as 1.53−0.008i [d'Almeida et al., 1991] in this study.
Table 1. Aerosol Volume Size Distribution From AERONET Sun Photometer During ACE-Asia IOPa
Each value was averaged after collecting the data for Asian dust cases.
Here, vp, volume portion of fine and coarse mode divided by total volume.
 In their paper, von Hoyningen-Huene et al.  performed sensitivity analysis to investigate the effects of surface reflectance and geometrical conditions. An error of ±0.02 in the surface reflectance would lead to the selection of a wrong LUT, resulting in an error of ∼0.1 in the AOT [von Hoyningen-Huene et al., 2003].
 This sensitivity study was done in order to investigate the impact of aerosol microphysical parameters such as refractive index and size distribution on the satellite-received signal. For the real part of the refractive index of dust particles, the recommended value is ∼1.50–1.55 [Moulin et al., 1997]. Three representative values, 1.50, 1.53, and 1.55, were selected to calculate the TOA aerosol reflectance. Also, values of the imaginary part of dust were selected as 0.005, 0.008, and 0.010. The impacts of various refractive indices on the aerosol retrieval process are compared in Figure 4, where each curve shows the relation between TOA aerosol reflectance and AOT at 550 nm. The figure shows that imaginary part is relatively more important than the real part in calculating satellite AOT. In Figure 3 the relationship between each part of the refractive index and AOT change shows that change in the imaginary part by ±0.003 leads to error in the AOT of 15% and change in the real part leads to AOT error of ±5%.
Figure 5 shows the results of a second sensitivity study of the dependence of aerosol size distribution on satellite-derived AOT at the wavelengths of SeaWiFs channels 2 (443 nm) and 6 (670 nm). The effect of aerosol size distribution on satellite signal was calculated for three representative median radii for the fine mode, 0.10, 0.12, and 0.14 μm, and for the coarse mode, 2.0, 2.5, and 3.0 μm, respectively. It was found that there is a negligible difference among the three coarse modes but a larger difference among fine modes. If the fine-mode radius is changed by ±0.02 μm, the error of AOT is up to 10%. The error resulting from changing the coarse-mode radius is negligible. Under the same aerosol loading (AOT = 1.0 at 550 nm), error in AOT was calculated to be ±0.1 at 443 nm and ±0.04 at 670 nm, respectively, depending on aerosol size distribution. The shorter-wavelength AOT is more sensitive to fine-mode distribution. The AOT is less sensitive to the coarse model distribution that dominates the Asian dust storm particles.
 The relationship between aerosol reflectance and AOT was calculated using four different aerosol models in Table 1. Figure 6 shows the dependence of aerosol reflectance on the satellite-derived AOT at 550 nm for those models. If the averaged aerosol model is used for AOT calculation, the relative error for the AOT is less than 10% except for the Beijing aerosol distribution, which has a higher volume fraction of the coarse mode.
 On the basis of the results above, an aerosol model with rf = 0.12 μm and rc = 2.56 μm, which were averaged from AERONET Sun-photometer-measured size distribution parameters during Asian dust storm periods, was used in radiative transfer calculation. The SeaWiFS radiance was simulated with the 6S code for that aerosol model assuming it to be an approximation for the ACE-Asia dust aerosol. The LUTs of the six visible and two near-infrared SeaWiFS channel radiances were generated for five AOT values: 0.0, 0.2, 0.5, 1.0, and 2.0.
5. Results and Discussion
 During the ACE-Asia IOP the SeaWiFS data were collected over the study area excluding cloud-contaminated scenes. AOT retrieval results from these data were validated and compared with those from AERONET ground-based Sun photometer data. SeaWiFS-retrieved AOT data were also compared with TOMS AI, MODIS AOT, and air mass backward trajectory results to investigate the characteristics of spatial and temporal patterns of AOT.
 The BAER algorithm was applied to retrieve the spectral AOT from SeaWiFS data. It uses a reflectance threshold value of ρ412TOA = 0.2 for the cloud screening, which was selected below the minimum cloud reflectance value [Kokhanovsky, 2001]. However, in the case of heavy aerosol loading such as Asian dust events, this threshold value has to be increased. In order to eliminate the cloud effect, pixels where the reflectance exceeded 0.3 for four channels (412, 443, 490, and 510 nm) were excluded from data processing in this study. Although this can detect optically thick clouds effectively, it might not be sufficient for thin clouds and subpixel clouds because the dust effect can be shown in the same range. Although this visible threshold method may not be used elsewhere, it is appropriate in the ACE-Asia region. In this case, the aerosol product of SeaWiFS is a by-product. The spatial variability method [Martins et al., 2002] in the visible region can be applied to detect most of the clouds except cirrus. However, this method can only be applied over ocean. For effective cloud screening, the spatial variability method with IR test should be performed. Since SeaWiFS has no IR channel, the simple visible threshold method above was used in this study for cloud screening for aerosol retrieval over both ocean and land. Our retrieval, nevertheless, was limited to concurrent cases under clear-sky conditions, which were validated with AERONET cloud-screened and quality-assured data.
Figure 7 shows the correlation between SeaWiFS-retrieved AOT at two wavelengths (443 nm and 670 nm) and AERONET Sun-photometer-derived AOT at different sites: Gosan, Anmyun, and Beijing. The SeaWiFS scans the study area at ∼0300–0400 UTC; the AERONET measurements nearest in time were taken for the comparison. Although the comparison was done for a limited number of cases, there was good agreement between Sun-photometer-measured and satellite-retrieved AOT with linear fitting correlation coefficient (r) above 0.91 and the relative root-mean-square difference (RMSD) of 0.09 and 0.07 at 443 and 670 nm, respectively.
Table 2 summarizes the validation results obtained under various conditions considered in this study. At Gosan the mean values of the SeaWiFS-retrieved AOT at the two wavelengths 443 and 670 nm were 0.49 and 0.40, respectively, with standard deviations of 0.20 and 0.16, respectively. These AOT values are higher than those of Anmyun but lower than those of Beijing, as shown in Table 2. The comparison between the SeaWiFS AOT and AERONET AOT shows good agreement, with correlation coefficients over 0.91 and linear slopes over 0.84, as shown in Table 2. This means that points are not scattered and SeaWiFS AOT is slightly less than AERONET AOT. Lower SeaWiFS AOT can be mainly due to the surface reflectance effect. Low NDVI due to a dense aerosol plume can lead to the overestimated surface reflectance that caused underestimated AOT.
Table 2. Statistical Analysis Results of SeaWiFS-Retrieved Mean AOT, Which Was Compared With AERONET Sun Photometer AOT During April 2001
0.96x + 0.02
0.49 ± 0.20
0.86x + 0.10
0.40 ± 0.16
0.90x + 0.07
0.35 ± 0.14
0.87x + 0.07
0.24 ± 0.12
0.83x + 0.14
0.64 ± 0.31
0.97x + 0.08
0.49 ± 0.26
 Precipitation occurred on 4, ∼8–9, ∼16–17, ∼24–25, and ∼29–30 April 2001, and two major Asian dust storms were observed at Gosan supersite, ∼11–13 April and ∼25–26 April, during the ACE-Asia IOP [Huebert et al., 2003]. Figure 8 shows temporal variation of SeaWiFS AOT, AERONET AOT, and MODIS AOT at Gosan. AERONET AOT data at 500 nm were corrected to 550 nm using corresponding wavelength dependent Ångström exponents (τ550 = τ500(550/500)−α). Elevated AOT values were observed during two Asian dust storm periods. The highest AOT was observed during the first Asian dust storm observed at Gosan on ∼10–14 April 2001 and the second Asian dust storm on 25 April 2001. From the comparison of AOT data observed by Sun photometer and satellites, the SeaWiFS AOT data have good agreement with both AERONET AOT and MODIS AOT data. Since spatial resolution of MODIS was 10 × 10 km2, there existed relatively large discrepancies under nonuniform dust plume conditions.
Table 3 summarizes the day-by-day comparison between SeaWiFS-derived AOT and Ångström exponent and those from AERONET data selected within the ±30 min time window around the SeaWiFS overpass. AERONET AOT data measured at 500 nm are changed to AOT at 550 nm in Table 3 by using the Ångström exponent. The difference in AOT between the two methods was less than 10%, which can support the use of a single average aerosol distribution in the retrieval method.
Table 3. Comparison Between SeaWiFS-Retrieved AOT at 550 nm and Ångström Exponent and AERONET-Derived AOT and Ångström Exponent Over Gosan
12 April 2003
13 April 2003
14 April 2003
15 April 2003
16 April 2003
17 April 2003
26 April 2003
 The SeaWiFS-retrieved aerosol optical parameters are used to investigate spatial and temporal patterns of aerosol over the study area. TOMS AI data (http://toms.gsfc.nasa.gov) and MODIS AOT data (http://daac.gsfc.nasa.gov) collected for the ACE-Asia IOP revealed that dust plumes had passed over Gosan, Jeju Island, during major Asian dust storm events. On 13 April 2001 the SeaWiFS passed the study area at 0415 UTC. Figure 9 shows the SeaWiFS-retrieved AOT maps on 13 April 2001 in the middle of an Asian dust storm event and on 15 April 2001 after the event. In Figure 9, a SeaWiFS RGB image and AOT map on 13 April show that the Asian dust storm can be recognized by very well defined high AOT values (range of 0.6–1.7) and the dust plumes moving through the Yellow Sea. However, low AOT values (range of 0.3–0.6) are shown over the most northern part of the Korean peninsula. Even if the AOT near Anmyun is ∼0.2–0.3, the AOT of these dust plumes near Gosan was ∼0.7–0.8. Especially, the lidar-measured aerosol extinction profile showed that there were two aerosol layers over Gosan on 13 April 2001 [Hong et al., 2004] with high PM10 concentration close to 100 μg/m3 at the surface. Figure 10 shows the same area covered by TOMS AI and MODIS AOT, which have similar spatial patterns of aerosol. The TOMS-measured aerosol index is 2.2 over Gosan. The SeaWiFS-retrieved AOT in Figure 9 shows finer spatial resolution, enough for identifying the extent of dust plume. On 15 April 2001 an Asian dust storm was not reported at the surface on Jeju Island. However, a high-extinction aerosol layer was observed by lidar under 2-km altitude [Hong et al., 2004]. The TOMS AI indicates that the dust plume moved eastward and became more diffused as well. The SeaWiFS RGB image shows the blue color region over the Yellow Sea. The Ångström exponent of this aerosol is larger than that of the 13 April dust plume over Gosan. While AOT at 550 nm decreased to 0.30 on 15 April from 0.71 on 13 April, the Ångström exponent increased to 1.12 on 15 April from 0.52 on 13 April. In this particular case, it is likely that the aerosol consists of finer particles, such as pollution aerosol over the region. The TOMS AI shows a negative value, which means that it is nonabsorbing aerosol such as sulfate or sea-salt aerosol or an aerosol layer existing at low altitude.
 In order to investigate the sources and pathway of an air mass reaching Gosan, the air mass backward trajectory of the air mass reaching Gosan (126.1°E, 33.23°N) was calculated. The HYSPLIT-4 model was used to calculate backward trajectories for 0400 UTC arrival time each day. The 4-day (96 hours) backward trajectories at five heights of 500, 1000, 2000, 3000, and 4000 m were plotted in Figure 11. The top and bottom graphs of air mass trajectory analyses show the horizontal and vertical motion of air mass, respectively. These trajectory results show the large-scale atmospheric transport pattern, which is particularly informative for qualitative assessment of the air mass pathway. For the Asian dust case on 13 April 2001, trajectory lines that originated from the desert area passed through an industrialized area in northeast China. Vertical movement of the trajectory shows that the air mass can be affected by anthropogenic pollution near surface level. The difference between the Asian dust case on 13 April and the non-Asian-dust case on 15 April is direction and vertical movement of trajectory lines. Trajectory lines that originated from Lake Baikal at high altitude passed through northeast China and the Yellow Sea at low altitude on the second case day. These different air mass pathways can affect physical and optical characteristics of aerosol by different source contributions
 This work discussed mainly the retrieval of AOT from SeaWiFS data to estimate aerosol optical parameters in fine spatial resolution over northeast Asia during the ACE-Asia IOP. The main goals of this study were to validate SeaWiFS-retrieved AOT using the BAER algorithm over both ocean and land and to estimate the spatiotemporal pattern of aerosol optical parameters from satellite data. Using the same aerosol characteristics over a large geographical area, the errors introduced in AOT retrievals can be minimized. In order to apply the BAER algorithm to the study area, ocean surface reflectance had been corrected since ocean in the study area has high turbidity. New ocean surface reflectance determined by linear mixing model tuned by normalized spectral difference for pigments of water was used.
 The sensitivity study for TOA radiance change due to aerosol size distribution was also conducted to investigate the dependence of refractive index and particle size distribution on TOA reflectance. Validation of SeaWiFS-retrieved AOT was conducted by intercomparing with AERONET AOT data collected during the ACE-Asia IOP. This revealed a good agreement for most of the cases. The results are satisfactory with correlation coefficients above 0.89 and RMSD below 0.1 at the AOT for the wavelengths 443 and 670 nm.
 The results of SeaWiFS aerosol retrieval provide the regional aerosol pattern and information about aerosol optical characteristics. On 13 April 2001 a strong Asian dust event was observed. The SeaWiFS-retrieved AOT showed a high AOT value close to 0.71 and an Ångström exponent of 0.52 during this event over Gosan. Air masses that originated from the northwest Chinese desert regions had passed through the Chinese industrial and urban areas before arriving in Gosan. The spatial distribution of Asian dust aerosol could be identified from the SeaWiFS-retrieved AOT image. On 15 April 2001, after the dust event, a SeaWiFS-retrieved AOT of 0.30 was obtained with an Ångström exponent 1.12. These optical characteristics of aerosol indicate the dominance of small particles possibly from anthropogenic sources. Aerosol parameters retrieved from SeaWiFS data using the BAER technique showed results that hold promise for monitoring atmospheric aerosol loading and estimating its optical properties.
 This work was supported in part by the Korea Science and Engineering Foundation (KOSEF) through the Advanced Environmental Monitoring Research Center (ADEMRC) at Kwangju Institute of Science and Technology (K-JIST) and the Brain Korea 21 program from the Korea Ministry of Education and Human Resources Development. The author would like to thank NASA for the use of SeaWiFS, TOMS AI, and MODIS data. We thank the B. Holben, C. McClain, P. Gouloub, and H. B. Chen for their efforts in establishing and maintaining Jeju, Anmyun, and Beijing AERONET sites.