Consistency of the aerosol type classification from satellite remote sensing during the Atmospheric Brown Cloud–East Asia Regional Experiment campaign



[1] The Atmospheric Brown Cloud–East Asia Regional Experiment (ABC-EAREX) was conducted under the UNEP/ABC-Asia project to intercompare the aerosol and gas measurements in springtime from various instruments from late February to April 2005 at the Gosan Supersite on Jeju Island, Korea. Satellite instruments such as the Moderate Resolution Imaging Spectroradiometer (MODIS) and the Ozone Monitoring Instrument (OMI) provide a large-scale regional view of the aerosol during the ABC-EAREX period. This study shows the temporal and spatial distribution of four major aerosol types (dust, carbonaceous, sea salt and sulfate) retrieved by MODIS-OMI Algorithm and Four-Channel Algorithm utilizing data from MODIS and OMI over east Asia during the ABC-EAREX campaign. Results from two different retrieval show that a complexity of aerosol types and sources exist over east Asia: Some aerosols are emitted while others are transported. Nevertheless, the results show reasonable consistency in the distribution according to aerosol type. The agreement of aerosol type classification for each aerosol type ranges from 32% to 81% depending on the type. These results were compared with the results from a three-dimensional aerosol transport radiation model, SPRINTARS. Dust type aerosol is usually found to be mixed with carbonaceous type aerosol. It implies that the dust type aerosol is loaded and transported with polluted air mass. The evidence that polluted air masses in the continent can be transported long distance is also captured; that is, sea salt type mixed with the sulfate aerosol is detected over a remote ocean.

1. Introduction

[2] Asia holds more than 60% of the population of the world. However, only recently has it been recognized as an important aerosol source region that impacts the climate on a global scale. Because of the persistent dust storms and growing air pollutants, the air mass in this region includes natural aerosols mixed with variable sources of anthropogenic aerosols [e.g., Chu et al., 2005]. The Atmospheric Brown Cloud–East Asia Regional Experiment (ABC-EAREX) was an important basic step toward the long-term goals of the ABC project. The experiment was designed to intercompare the characteristics of different ground-based and satellite-borne instruments used to collect aerosol and gas data at Gosan, Korea (along with Maldive, one of the two supersites in this program), from February to April 2005. Satellites such as Moderate Resolution Imaging Spectroradiometer (MODIS) and Ozone Monitoring Instrument (OMI) provided a large-scale regional view of the aerosol during the ABC-EAREX period in 2005. With the improved accuracy of satellite observation, these observations have played a significant role in understanding the global characteristics of aerosol, for example, MODIS [Kaufman et al., 1997; Tanré et al., 1997; Remer et al., 2005], AVHRR [Higurashi and Nakajima, 1999; Mishchenko et al., 1999; Ignatov and Stowe, 2002], and TOMS [Herman et al., 1997]. It is a challenging work to classify aerosol in this area where growing and variable source of natural and anthropogenic aerosols are mixed throughout the year. ABC-EAREX, conducted at Gosan, Korea located to the east of China, provides an excellent opportunity to pretest the satellite capability in classifying aerosol types during the long-term ABC project.

[3] Characterization of the chemical aerosol type is an important factor in evaluating the climate forcing by the aerosol. It is also the basic step in distinguishing anthropogenic from natural aerosol. An aerosol-cloud-radiation model requires the chemical and microphysical nature of aerosol as input to the model [cf. Ramanathan et al., 2007]. In spite of its importance, the spatiotemporal distribution of aerosol types is poorly known at both regional and global scales. For these reasons, significant satellite-based efforts have been expended to monitor and predict the source, sink and transportation of aerosols.

[4] There have been limited researches in classifying aerosol types from satellite remote sensing compared to obtaining aerosol amounts and optical properties such as optical thickness, single scattering albedo, effective radius etc. Studies in aerosol types by remote sensing were limited mainly because of the difficulty in classifying chemical components from the columnar measurements of radiances from space. Higurashi and Nakajima [2002] developed a four-channel algorithm to classify aerosols into four major aerosol types, that is, soil dust, carbonaceous, sulfate, and sea salt aerosols, using the four-channel data of SeaWiFS over the ocean. The advantage of this algorithm is its near real-time capability in classifying aerosols from the simultaneous measurements at four wavelengths (412, 443, 670, 865 nm). With their analysis, they showed that the East China Sea region was loaded with different aerosols in a highly mixed condition. Jeong and Li [2005] developed an aerosol classifying algorithm by utilizing two instruments, Total Ozone Mapping Spectrometer (TOMS) and Advanced Very High Resolution Radiometer (AVHRR). They combined aerosol size information (Ångstrom exponent) from AVHRR and aerosol absorption(aerosol index) from TOMS to infer the aerosol types of biomass burning, pollution, dust, sea salt, and mixture of different types. The advantage of the TOMS-AVHRR algorithm is its adaptability to other satellite platforms, which opens the possibility to expand the analysis to include past satellite data for climate research. Hsu et al. [2004] developed the deep blue algorithm, which uses three channels to distinguish smoke and dust over bright-reflecting source regions. There have been other attempts to infer aerosol types, for example from the multi angle observation of the MISR [Martonchik et al., 1998], and the polarized observation of the POLDER [Bellouin et al., 2003].

[5] In this study, we classify aerosols over the east Asia region into four major types, that is, dust, carbonaceous, sea salt and sulfate by using the MODIS-OMI algorithm (MOA hereafter) and the four-channel algorithm (4CA hereafter) [cf. Jeong and Li, 2005; Higurashi and Nakajima, 2002] during the ABC-EAREX campaign. These results were then compared with the results from a three-dimensional aerosol transport radiation model, SPRINTARS [Takemura et al., 2000], in terms of optical depths of each aerosol types. The comparison of remote sensing results with SPRINTARS model results is intended to supplement the interpretation of aerosol distributions, not to validate satellite data retrievals. The consistency between the two different algorithms is also analyzed and discussed.

2. Methods

[6] MOA and 4CA are used to classify aerosols into four major types over ocean and possibly over land. To classify an aerosol, we need two basic properties of aerosol, that is, its size and radiation absorptance [Higurashi and Nakajima, 2002]. In this study, MODIS aerosol optical thickness (AOT), fine mode fraction (FMF) and OMI aerosol index (AI) are used for MOA, while reflectance data from MODIS channel 1 (640 nm), channel 2 (860 nm), channel 8 (410 nm) and channel 9 (440 nm) are used for 4CA.

2.1. MODIS-OMI Algorithm (MOA)

[7] To classify an aerosol, MOA uses MODIS AOT, FMF and OMI AI data. FMF is the ratio of fine mode AOT to the total AOT,

equation image

ranging from 0 to 1, where τ550,fine is fine mode AOT and τ550 is total AOT at 550 nm [Remer et al., 2005]. According to this definition of FMF, FMF provides information on the representative size of the aerosol. If there is little coarse mode aerosol in the atmosphere, FMF is close to 1, which implies an anthropogenic origin. For nonanthropogenic dust, the value of FMF becomes small, approaching 0 in the extreme case.

[8] OMI AI is a measure of how much the wavelength dependence of back scattered UV radiation from an atmosphere containing aerosols differs from that of a pure molecular atmosphere. Quantitatively, the AI is defined as

equation image

where I360Meas is the measured 360 nm radiance and I360Calc is the calculated 360 nm radiance for a Rayleigh atmosphere [Herman et al., 1997]. Under most conditions, AI is positive for absorbing aerosols and negative for nonabsorbing aerosols. Information on aerosol size can be provided by the MODIS FMF, while radiation absorptance can be inferred from OMI AI by utilizing the sensitivity of absorptive aerosol in UV. Then from these two independent data from different instruments, an aerosol can be classified into one of four types, that is absorbing coarse mode (dust), nonabsorbing coarse mode (sea salt), absorbing fine mode (carbonaceous) and nonabsorbing fine mode (sulfate) [see Higurashi and Nakajima, 2002]. As the exact property of organic carbon is not certain in this categorization from the satellite remote sensing, the term “carbonaceous” refers to elementary carbon in this work.

[9] Figure 1 shows the flowchart of the MOA, similar to those from Jeong and Li [2005] which has been applied to AVHRR and TOMS. First, the OMI AI can be used to determine the radiation absorptance of the aerosol. Jethva et al. [2005] showed in their study in the Indo-Gangetic basin that for absorbing aerosols such as dust, smoke and biomass burning, the AI values lie between 0.5 and 3.0, while nonabsorbing aerosols such as water soluble and sea salt particles yield negative or low positive values of AI. In order to minimize the misjudgment of absorbing aerosol, this study uses an AI threshold value of 0.7.

Figure 1.

Flowchart of MODIS-OMI algorithm.

[10] Next, the MODIS FMF data can be used to determine the columnar aerosol size. Kaufman et al. [2005] determined the averaged FMF for sea salt, dust and anthropogenic aerosol over ocean as 0.32 ± 0.07 in clean marine region (50–120°E, 20–30°S) for whole year of 2002, 0.51 ± 0.03 over west of the African coast (15–20°W, 15–20°N) for June-October 2002 and 0.92 ± 0.03 over the western Atlantic (70–90°W, 40–50°N) for June and (60–80°W, 40–50°N) for July 2002 respectively. Thus, according to Kaufman et al. [2005], the representative FMF values can be taken as 0.32 for sea salt, 0.51 for dust, and 0.92 for anthropogenic sources(carbonaceous and sulfate). Anderson et al. [2005] analyzed the fine/coarse partitioning of aerosol over Korea and Japan from their airborne experiments and compared the results with MODIS retrieval. In their study, the average FMF for the dust (Dust Report by Y. S. Chun, Korea Meteorological Administration, private communication, 2006) was found to be 0.535, whereas values for the continental sources were close to 1. Chu et al. [2005] also showed similar statistics for the FMF: 0.47 in the cleanest regions far offshore, and 0.85 in nearshore regions dominated by smoke from biomass burning, during the ACE-Asia, which is also consistent with the current thresholds. Jethva et al. [2005] showed in the Indian region that the FMFs are less than 0.4 during the summer months with dust events, but are larger than 0.8 for the winter months with anthropogenic sources. Thus, to minimize the classification errors, rather than using single sharp threshold values for FMF, the ranges of 0.6–0.8 for absorbing aerosols and 0.5–0.7 for the nonabsorbing will be used. The FMF values between 0.5 and 0.7 for nonabsorbing aerosol (sea salt/sulfate mixture), and between 0.6 and 0.8 for absorbing aerosol (dust/carbonaceous mixture) are considered to be a mixed type of coarse and fine mode particles.

[11] Finally, from the AOT test, an aerosol is classified into four major types (dust, carbonaceous, sea salt, and sulfate) and four types of mixtures. Thus, from the previous classification criteria, a dust type aerosol has an FMF less than 0.6 and an AI greater than 0.7. If AOT is less than or equal to 0.2, a mixed condition of sea salt and dust is assumed reflecting the uncertainties in classification for the cases of low aerosol loading. The nonabsorbing, coarse mode aerosol is sea salt that has FMF less than 0.4 and AI less than or equal to 0.7. When FMF is greater than 0.8 and AI greater than 0.7, the type of aerosol is assumed to be carbonaceous. Sulfate type has FMF greater than 0.8 and AI less or equal to 0.7. These criteria can be summarized in Table 1.

Table 1. Threshold Values of Aerosol Type Classification From MOA
AI ≤ 0.7AI > 0.7
FMF < 0.5sea salt, sea salt + sulfate (if AOT > 0.2) 
FMF < 0.6 dust, sea salt + dust (if AOT ≤ 0.2)
FMF > 0.7sulfate 
FMF > 0.8 carbonaceous
0.5 < FMF < 0.7sea salt + sulfate 
0.6 < FMF < 0.8 dust + carbonaceous, sea salt + carbonaceous (if AOT ≤ 0.2)

2.2. Four-Channel Algorithm (4CA)

[12] This classification algorithm is based on the work of Higurashi and Nakajima [2002] which has been applied to the SeaWiFS data. This 4CA adopted the look up table (LUT) approach for computational efficiency. To compute TOA reflectance we assume a bimodal lognormal size distribution for the aerosol:

equation image

where V is the aerosol volume density, rm is the mode radius in μm, and s is the standard deviation of r, radius.

[13] MODIS channel 1 and 2 are used to retrieve the aerosol optical properties while channels 8 and 9 are used to determine whether the aerosol absorbs radiation or not. We compute LUT by using the Rstar5b radiative transfer model (RTM) [Nakajima and Tanaka, 1986] assuming absorbing and nonabsorbing aerosol type. The calculation of LUT is done at an interval of 2.5° from 0° to 70° for solar and satellite zenith angle, and at an interval of 5° from 0° to 180° for the relative azimuth angle. Eleven cases of aerosol volume peak ratio (c2/c1) are calculated ranging from 0.1 to 100.

[14] Figure 2 shows the Nakajima-King type LUT for specified satellite-Sun geometry, where similar patterns are obtained for other geometries except for the Sun glint region defined by:

equation image

where θs, θv, and ϕ are the solar zenith, the satellite zenith, and the relative azimuth angles (between the Sun and satellite), respectively [Remer et al., 2005]. For the severe aerosol loading case with AOT greater than about 1.5, MODIS channel 1 and 2 reflectance decreases with increasing volume peak ratio for the absorbing aerosol model. This implies that fine mode absorbing aerosol scatters radiation more effectively than coarse mode absorbing aerosol. On the other hand, MODIS channel 2 reflectance increases with volume peak ratio for the nonabsorbing aerosol model. This means that more radiation at channel 2 (860 nm) is reflected to the satellite by coarse mode aerosol than by fine mode, while the signal change in channel 1 (640 nm) is not so sensitive to the particle size. Using observed reflectance and computed reflectance, AOT and volume peak ratio are retrieved.

Figure 2.

Computed LUT by using Rstar5b RTM for different aerosol loading and volume peak ratio (c2/c1) assuming (a) absorbing aerosol model and (b) nonabsorbing aerosol model.

[15] To classify aerosols into the four major types, two important decisions on the size and radiation absorptance of the aerosol must be made. The difference between channel 8 and 9 reflectance (Δρ = ρ410–ρ440) is smaller for the absorbing aerosol model than the nonabsorbing aerosol model. Comparing the reflectance differences between channel 8 and 9 with those values for the absorbing or nonabsorbing model, the radiation absorptance of the aerosol is determined. As a next step, AOT and volume peak ratio can be retrieved by using MODIS channel 1 and 2 reflectance for the appropriate aerosol model. Detail flowchart for the 4CA is summarized in Figure 3.

Figure 3.

Flowchart of four-channel algorithm.

2.3. Cloud Masking

[16] Cloud masking is essential in satellite-based retrieval of aerosol properties, since cloud can generate large differences on the spectral radiation field as compared with the cloud-free condition. For the MOA, the cloud masking is from the science teams of MODIS and OMI. The details are given by Remer et al. [2005] and Stammes [2002] for MODIS and OMI, respectively.

[17] Cloud masking in 4CA basically applies the five kinds of tests for cloud masking of MODIS aerosol products over ocean [Remer et al., 2005]. The spatial resolution of channels 1 and 2 data used here is 500 m, while the resolution of channels 8 and 9 is 1 km. Thus one clear pixels of channels 8 or 9 in 1 km × 1 km resolution must have four clear 500 m × 500 m pixels, which masks out more pixels than the standard MODIS aerosol retrievals.

2.4. SPRINTARS Model

[18] As stated above, the SPRINTARS model is used in this study not for validating satellite retrievals, but for supplementing them to interpret aerosol distributions. SPRINTARS is a global aerosol climate model based on a general circulation model. The model simulates global distributions of the main tropospheric aerosols, i.e., black carbon, organic carbon, sulfate, soil dust, and sea salt. The simulated aerosol optical thickness, Ångström exponent, and single-scattering albedo were validated with dozens of observed values from both optical ground-based measurements and satellite remote sensing retrievals. The model successfully simulates aerosol distributions in the east Asian region including Asian dust events [Takemura et al., 2002a, 2003]. The aerosol direct effect can be calculated considering differences in refractive indices depending on wavelength, size distributions, and hygroscopic growth for each aerosol component. The cloud droplet number concentration is diagnosed using a parameterization based on the cloud microphysics to calculate changes in cloud droplet radius (the aerosol first indirect effect) and cloud lifetime (the aerosol second indirect effect). The detailed description of SPRINTARS is given by Takemura et al. [2000, 2002b, 2005].

3. Results

[19] In spring time, Asian dust from Taklamakan and Gobi desert are transported long distance across the Korean Peninsula to Japan, and sometimes across the Pacific Ocean to North America [Wilkening et al., 2000]. Figures 4 and 5show MODIS AOT, FMF, OMI AI, NCEP Reanalysis wind vectors at 850 hPa, and SPRINTARS dust AOT, with corresponding classified results in Figures 6 and 7on 20 April 2005 and 28 April 2005, respectively. Aerosol information only from MODIS on board the Aqua satellite are used in this study to closely match the overpass time with OMI measurements, both of which have equatorial crossing time near 1330 local time. By doing so, possible error sources due to the different overpass time between the two satellites are minimized. The retrieved areas for the aerosol products of OMI and MODIS are different because of differences in sensitivity to aerosol, coverage and cloud masking. These figures show the general situation of the aerosol events, not a quantitative comparison. On 20 April, air mass with high AOT (Figure 4a), small FMF (Figure 4b) and positive AI (Figure 4c) values have been observed from the east side of China across the Yellow Sea to the Korean Peninsula, while the mixture of absorbing and nonabsorbing fine-mode aerosol dominates the south eastern part of China below the latitude of about 35°N. Although the information from the MODIS is limited over land, the dust seems to be blowing out of the northern part of inland China, as shown by SPRINTARS model results as in Figure 4e. The feature of SPRINTARS dust AOT is similar with the NCEP/NCAR Reanalysis wind vector (Figure 4d). As shown in Figure 6a, the MOA detected this dust outflow from the east side of China to the west of the Korean Peninsula as dust (yellow color), while for the area in east inland China below 35°N MOA detects aerosols as carbonaceous (red color) and sulfate (sky blue color). Over the remote ocean region, coarse and nonabsorbing aerosols with small AOT are dominant, which implies existence of sea salt type over the remote ocean. In some of ocean areas, a mixture of sulfate and sea salt type aerosols is detected. Almost similar features have been detected from the 4CA of MODIS (Figure 6b), with dust over the Yellow Sea and the mixture of sea salt and sulfate over the ocean. Some of the carbonaceous aerosols mixed with weak dusts near the Sakhalin and east of Hokkaido, Japan are also noteworthy and consistent between the two algorithms. The MOA takes the daily values of AOT and OMI AI and thus shows low-resolution features compared to the 4CA from MODIS. Thus there exists an overpass time gap between the two instruments, which may account for some portion of differences between the two algorithms. Despite this time difference, the two algorithms show reasonably consistent performance.

Figure 4.

(a) MODIS AOT, (b) FMF, (c) OMI AI, (d) wind vector at 850 hPa from NCEP/NCAR reanalysis data set, and (e) SPRINTARS dust AOT on 20 April 2005.

Figure 5.

(a) MODIS AOT, (b) FMF, (c) OMI AI, (d) wind vector at 850 hPa from NCEP/NCAR reanalysis data set, and (e) SPRINTARS dust AOT on 28 April 2005.

Figure 6.

Comparison of aerosol classification from (a) MODIS-OMI algorithm and (b) four-channel algorithm on 20 April 2005. Each color represents the different aerosol type as shown below the images.

Figure 7.

Comparison of (a) aerosol classification by MODIS-OMI algorithm and (b) four-channel algorithm on 28 April 2005.

[20] The model results from SPRINTARS in Figure 4e also show similar dust outflow from the Sandung Peninsula across the Yellow Sea and Korea, reaching to the sea between Korea and Japan. Comparing the model results with OMI AI (Figure 4c), the dust transport across Korea and Japan to the Minami-tori Island (24°N, 154°E) in the Pacific Ocean shows quite similar features, although the cloud (near zero AI) and limited swath of the sensor constrain the display of the full features in OMI AI. The simulated AOT values are underestimated compared to the observation values from MODIS.

[21] A more severe dust event occurred on 28 April, with significantly enhanced MODIS AOT values up to 2 (Figure 5a) over the Yellow Sea, and AI over 3 (Figure 5c), when a “yellow sand warning” was announced publicly in Korea by the Korea Meteorological Administration (KMA). The FMF was quite reduced along the dust flow from China across Korea to the west coast of Japan as shown in Figure 5b. SPRINTARS simulated the results of transported dusts reasonably well as compared with the observations from MODIS and OMI as shown in Figure 5e. The location of the dust band is remarkable, and follows the wind field given in Figure 5d, but the absolute optical depth of the dust flow is underestimated.

[22] Both the MOA and 4CA detected this severe dust event nicely as shown in Figure 7. The dust outflow from the inner Mongolia across North Korea to Japan and another outflow from the Gobi desert to the Sandung Peninsula are detected in MOA, as compared to the OMI AI in Figure 5c. Similar performance and features were observed in the 4CA over the ocean between Korea and Japan. A mixture of sulfate and sea salt type is dominant over the remote ocean as shown in the two classification algorithms.

[23] The comparison of aerosol classification by the two different algorithms in Figure 6 and 7 shows reasonable consistency over most of the region considered. There exist complex types of aerosol over the east Asia region from emitted and transported sources. In the spring Asian dust season in particular, dust type aerosols are usually mixed with carbonaceous type aerosols. This implies that dust aerosols are modified by pollutants during their transport. In general, sea salt type aerosol mixed with sulfate is dominant over the remote ocean.

[24] In order to compare the noted consistency between the two algorithms with a consideration of model results in a more quantitative way, the classification results are displayed in the same low resolution and are forced into the four types based on FMF threshold values (0.6 for nonabsorbing and 0.7 for absorbing aerosols), and AI threshold of 0.7 in Figures 8a and 8b. Thus, for example, the mixture dust + carbonaceous which corresponds to intermediate values of FMF is forced into only one type (either dust or carbonaceous) of mixture, in quantitative comparison among different algorithms and model results. Figure 8c shows the dominant aerosol type in terms of optical depth from the SPRINTARS model. There exist remarkable agreement between the model and satellite observation in eastern part of China and Korea (dust), in China Sea (sulfate), and in Philippine Sea and Pacific Ocean below latitudes of about 25° (sea salt), except for the occasional carbonaceous type detected in the model result over the ocean. There are also large difference in the northern Pacific Ocean around the Kamchatka Peninsula between the model (sea salt) and satellite detection (sulfate). The degree of agreement, defined as the percentage of pixels with the same results out of a 3° × 3° box between the two satellite classification algorithms in Figures 8a and 8b are shown in Figure 8d. The degree of agreement ranges from 50% to 100% in most of the areas. Excellent correlations were obtained in the dust outflow in the Yellow Sea and over the remote ocean. The degree of agreement between the two algorithms was 67% over the whole region considered.

Figure 8.

Aerosol type classification from (a) MODIS-OMI algorithm and (b) four-channel algorithm in 1° × 1° resolution, (c) dominant aerosol types from SPRINTARS, and (d) agreement of aerosol type classification in 3° × 3° box from MODIS-OMI algorithm and four-channel algorithm on 20 April 2005.

[25] Table 2 summarizes the aerosol classification results from the MOA and 4CA of MODIS over east Asia on 20 April 2005. Comparisons were made at a 1° × 1° spatial resolution over the ocean only. The numbers in Table 2 represent the percentage of pixels counted as each aerosol type by 4CA with reference to the MOA. Thus the pixel percentages in each row sum to 1.0, while the summation of numbers in each column does not. The numbers in the diagonal of Table 2 (in bold type) represent the percentage of detecting the same type of aerosol from the two algorithms. In general the percentage to detect the same type of aerosol ranges from 47% to 89%. These numbers change to ranges from 21% to 89% when evaluated with reference to the 4CA of MODIS. The numbers in the parenthesis represent the number of 1° × 1° pixels which detected the corresponding aerosol type. Sulfate was detected most frequently in about 400 pixels while the number of pixels with dust and carbonaceous aerosols are much lower. The numbers off the diagonal represent the different detection of the aerosol type and are much lower than the numbers in diagonal except for some confusion arising from the fact that the carbonaceous type in MOA is detected as sulfate in the 4CA. This is due to the difference in absorptance tests for the fine mode. The difference could be attributed to the fact that the absorptance test in MOA is based on the UV measurements of AI in OMI, while the test in 4CA is based on the blue channel measurements.

Table 2. Agreement of Aerosol Type Classification From MODIS-OMI Algorithm (MOA) and Four-Channel Algorithm (4CA) Over East Asia on 20 April 2005a
Dust (54)Carbonaceous (91)Sea Salt (187)Sulfate (300)
  • a

    Numbers in parentheses represent the number of pixels detected over the ocean with reference to the 4CA. Boldface represents the agreement of aerosol type classification from two different algorithms for each aerosol type.

Dust (60)0.470.280.130.12
Carbonaceous (36)0.110.530.030.33
Sea salt (125)
Sulfate (411)

[26] In Figure 9, the detection frequency percentages of each aerosol type and averaged AOT are compared for the case of 20 April 2005. As mentioned above, sulfate is detected most frequently with percentage ranging from 50 to 65% compared to other types. In general, the statistics between the two different algorithms show reasonable agreement for the case of 20 April 2005, thus the consistency as well. By the influence of the dust event, the mean AOT of absorbing aerosol types is larger than for nonabsorbing types in this case.

Figure 9.

Frequency of each aerosol type from total aerosol types on 20 April 2005.

[27] Similar comparisons are shown in Figure 10 and Table 3 for the case of 28 April 2005. The dust outflow shown in Figures 10a and 10b has been discussed in the Figure 5. In terms of dominant optical depth, the SPRINTARS model results of Figure 5c show a large area of carbonaceous type aerosol over the ocean while the satellite detection shows only a limited area over the sea between Korea and Japan in both algorithms. Sulfate and sea salt type detected over the ocean also showed similar features. The model predicted the mixture of sulfate, carbonaceous and dusts over the sea between Korea and Japan. Over the whole area considered, the overall agreement was 68%, almost the same value as in the case of 20 April 2005. The statistics for the aerosol type detection are summarized in Table 3 for the case of 28 April 2005, which represent very similar characteristics for the case of 20 April 2005. The detection consistency was enhanced for dust in the case of the more severe dust event on 28 April 2005. Frequency distribution of each aerosol type and AOT are shown in Figure 11 for the case of 28 April 2005, which is also quite similar to the distribution of 20 April 2005. Because of the influence of the severe dust event, the mean AOT of retrieved dust pixels is increased up to 1.03 and 1.33 for the MOA and 4CA cases, respectively.

Figure 10.

Same as in Figure 8 except for 28 April 2005.

Figure 11.

Frequency of each aerosol type from total aerosol types on 28 April 2005.

Table 3. Same as in Table 2 Except for the Case of 28 April 2005
Dust (40)Carbonaceous (73)Sea Salt (221)Sulfate (312)
Dust (43)0.700.210.070.02
Carbonaceous (20)0.100.750.000.15
Sea salt (117)
Sulfate (466)

[28] Figure 12 compares the frequency distribution of each aerosol type at Gosan, Jeju Island. While the two-dimensional distribution of aerosol type detection showed reasonable consistency, this figure shows the limitations of applying aerosol classification results from the satellite remote sensing data at a fixed station. The dust type detection frequency was almost the same for the two algorithms, while the frequency distribution of the fine mode particles, i.e., sulfate and carbonaceous show large difference. There tends to be a greater probability for the sulfate type to be detected in MOA compared to 4CA, which also can be attributed to the different radiation absorption test in each algorithm. There may be some possibility to adjust the absorptance test in the MOA or 4CA algorithm to ensure better consistency. The limitation of MOA which relies on the satellite measurements for different overpass time of each location may have been reflected in these results.

Figure 12.

Frequency of each aerosol type from total aerosol types over Jeju Island (1° × 1°) during ABC-EAREX campaign.

[29] The overall consistency of aerosol type detection over east Asia for the whole period of ABC-EAREX campaign, from February to April 2005, is listed in Table 4, ranging from 32% for dust to 81% for sea salt. The detailed characteristics are very similar to the previous case study results shown in Tables 2 and 3. The frequency distribution of each aerosol type and AOT during the whole period of ABC-EAREX is shown in Figure 13. The AOT values for the dust is reduced compared to the previous case studies because of the limited number of dust event days during the campaign, but the AOT of absorbing aerosols is still larger than those of nonabsorbing. Overall, there seems to be significant similarity in the frequency distribution between the two algorithms.

Figure 13.

Frequency of each aerosol type from total aerosol types during ABC-EAREX campaign for three months from February to April 2005.

Table 4. Agreement of Aerosol Type Classification From MODIS-OMI Algorithm and Four-Channel Algorithm Over East Asia During ABC EAREX Campaign
Dust (3596)Carbonaceous (5504)Sea Salt (21916)Sulfate (18235)
Dust (2280)0.320.160.380.14
Carbonaceous (3268)0.110.490.100.31
Sea salt (14591)
Sulfate (29112)

4. Conclusion

[30] Comparison of aerosol classification by using MOA with 4CA is presented for the ABC-EAREX period together with the model results from SPRINTARS. In general, the results of the two different algorithms show reasonable consistency. Over the east Asian region, there exist complex types of aerosols from emitted and transported sources. The dust type aerosol usually occurred mixed with the carbonaceous type aerosol while the sea salt type mixed with sulfate type aerosol dominates the remote ocean regions. A three-dimensional aerosol transport model can be compared with the classified results only for the purpose of interpretation, but showed reasonably good agreement. Validation of aerosol classification results by using ground-based measurements is desirable to fine tune the detection efficiency, threshold values, and consistency from the satellite remote sensing.

[31] Although the analyses here are limited and preliminary, satellite observations show a promising future in understanding the global distribution and characteristics of aerosol type by using well-known, state-of-the-art MODIS and OMI, and possibly other relevant satellites. This work can be appreciated in the sense that the aerosol classification is carried out during ABC-EAREX in 2005 in a region where growing and variable source of natural and anthropogenic aerosols are mixed throughout the year. It also provided a meaningful basis to utilize the satellite capability in classifying aerosol types during the long-term ABC project.


[32] This research was supported by the COMS (Communication, Ocean and Meteorological Satellite) project of KMA/METRI, and the Eco-technopia 21 project under grant 121-071-055 by the Korea Ministry of Environment. This research was partially supported by the Brain Korea 21 (BK21) program for J. Kim and J. H. Lee.