2.1.

Biomass Burning Aerosols

As a typical example of absorbing aerosols, the BBAs were examined first. BBA plumes are generated by large-scale forest fires or agricultural burns. Though BBA events can occur anytime, typically, large BBA events occur in tropical forests in August and September. BBA events are detected using hot spots derived from the rapid-fire response system (MYD14 Collection 6) of the Terra/MODIS satellite. Figure 1 presents the global distribution of monthly accumulated hot spots (denoted by red dots) from August and September of 2003. These figures clearly depict the occurrence of severe forest fires in southeastern Africa and the Amazon region in August and September, and Indonesia in September.

Fig. 1

Global distribution of accumulated monthly hot spots (denoted by red dots) derived from Terra/MODIS (the Rapid-Fire Response System of MYD14 Collection 6) in 2003.

Based on the results of Fig. 1, we chose data from GLI measurements as BBA case study targets such that almost all selected areas were covered by the BBA event. From the southeastern Africa region (except for the Kalahari Desert), we chose data from August 1 to 2, 4 to 6, 9 to 11, 13 to 14, 17 to 22, 24 to 28, and 30, 2003. From the Amazon region (except for the Atacama Desert), we chose data from September 1 to 4, 7 to 8, 10 to 15, and 17 to 28, 2003. Lastly, in Indonesia, we chose data from September 1 to 2, 4, 8 to 13, 15, 17 to 18, 20, 23, and 27 to 30, 2003. These were regarded as proper candidates for the BBA feasible study. The number of total pixels was $\sim 3,000,000$. Note that the cloud pixels were excluded by the using GLI-cloud flag. Figure 2(a) shows a histogram of the AAIs from the three target regions and Fig. 2(b) provides the cumulative frequency diagram of the upper one. The mean value ($m$) and the standard deviation ($\sigma$) of the AAI histogram were 0.86 and 0.05, respectively. The cumulative frequency diagram [in Fig. 2(b)] showed that the domain of the AAI $\ge 0.78$ (denoted by the dotted line) accounted for 97% (by the solid arrows) of the total. The results in Fig. 2 indicated that the BBA-AAI values were usually was larger than 0.78.

Fig. 2

Histogram of AAI over BBA plume targets in southwestern Africa during August, the Amazon in September, and Indonesia in September of 2003 derived from the GLI measurements. (a) Pixel numbers and (b) the cumulative frequency (%).

In order to compare the data with preceding AAI, Fig. 3 presents a correlation between the GLI-AAI with the TOMS-AI²⁰ for the same target areas as those in Fig. 2. Here, the differences in the observing times and effective resolutions between the sensors should be considered. The time difference between the ADEOS-2/GLI and EP/TOMS was restricted within 30 min, and the mean of the GLI-AAI values within a corresponding TOMS-AI pixel was adopted. As a result, the available GLI data were decreased to about 4500, and the correlation coefficient ($\gamma$) had a value of 0.81. From Fig. 3, it is possible to conclude that the two indices tended to coincide with each other.

Fig. 3

Comparison of the GLI-AAI with the TOMS-AI over the same targets as in Fig. 2. The TOMS-AI data are from Ref. 20.

Naturally, satellites measure the vertically accumulated radiation, i.e., the satellite data are a mixing of the scattered radiation by atmospheric particles and the reflected radiation from the Earth’s surface. The contribution of the bottom surface reflection to the satellite data is considered to decrease with the opacity of the atmosphere, and hence the Earth’s surface reflection term becomes offset with the aerosol optical thickness (AOT). Here, we aimed to detect the aerosol characteristics without the interference of the Earth’s surface reflection. In the case of an AOT higher than 1.0, there is no significant reason to consider that the bottom surface reflection apparently contributes to the satellite measurements.²¹ Based on this assumption, we assessed the greatest impact of surface reflection on the satellite measurements considering the bottom surface reflection (surface albedo = 1.0), and the nadir viewing of the satellite at noon. In the case of AOT = 1, the satellite can observe about 10% of the incident solar radiation that is directly reflected by the bottom surface. The satellite observes multiply scattered light due to the occurrence of atmospheric particles and to the interaction of both of scattering and reflection; however, the contribution of reflection to the satellite measurements decreases with AOT. Accordingly, AOT = 1 was selected as suitable value for the measurement of the bottom surface reflection contribution, and the AAI values in the case of an $\mathrm{AOT}\ge 1.0$ were investigated here. The relationship between the AAI and the AOT was examined using AOT data derived from Terra/MODIS (MOD04_L2, Coll.6). The time difference between the ADEOS-2/GLI and Terra/MODIS was restricted within 15 min. Figure 4(a) presents the AAI values over the BBA-dominant areas versus MODIS/AOT ($0.550\mu \mathrm{m}$). The red dots represent the averaged AAI values at each AOT. It was found that the AAI values increased with the AOT and the AAI values converged around the average values beyond the AOTs ($0.550\mu \mathrm{m}$) = 1.0. The case of an AOT ($0.550\mu \mathrm{m}$) $\ge 1.0$ seemed to suggest that aerosols are optically thick enough for aerosol scattering itself to contribute to the satellite measurements with less interference from of the Earth’s surface reflection. Namely, the AAI values in the case of an $\mathrm{AOT}\ge 1.0$ just provided by BBAs had AAI values of $\ge 0.83$. In order to confirm this finding, Fig. 4(b) presented the histogram and cumulative frequency (%) of the AAI for the BBA. Namely, the GLI data used as the BBA in Fig. 4(b) were restricted within the right upper domain (consisting of about 9000 pixels) in Fig. 4(a). Figure 4(b) shows that almost all of the BBAs (99%) acquired the AAI values greater than 0.83.

Fig. 4

(a) Verification of the BBA-AAI dependency on the AOT ($0.550\mu \mathrm{m}$). The AOT values were obtained from Terra/MODIS (MODIS/MOD04_L2 Collection 6). And (b) presents the AAI histogram of the genuine BBA. (a) The pixel numbers and (b) cumulative frequency (%).

2.2.

Mineral Dust Aerosols (DUST)

Here, the mineral dust particles (DUST) were examined as another typical type of absorbing aerosol.^7,22 It is natural to consider that the main generation sources of mineral dust are deserts, and hence the desert areas given by the International Geosphere-Biosphere Programme (IGBP) were selected for the dust aerosol trial targets. Figure 5 is similar to Fig. 2, except for the intrinsic DUST, i.e., the selected mineral dust aerosols over the desert with an AOT ($0.550\mu \mathrm{m}$) $\ge 1.0$. The AOT values were obtained from the Terra/MODIS deep blue algorithm (MOD04_L2, Collection 6). The time difference between the ADEOS-2/GLI and Terra/MODIS was restricted within 15 min. The solid lines in Fig. 5 approximately represent the AAI values for the DUST. In other words, the solid lines in Fig. 5 correspond to Fig. 4(b) rather than Fig. 2. Figure 5(a) presents the histograms of the summarized AAI values recorded over the deserts from May to September, 2003. For reference, the AAI values over the optically thin ($\mathrm{AOT}<1.0$) desert areas are shown by the dotted line in the upper histogram. The corresponding GLI pixels were about 8,400,000 in total, and the DUST cases were about one-eighth of the total. It was evident that the DUST had higher AAI values than those of the BBA. Almost all of the DUST (97%) (denoted by the solid arrows) acquired the $\mathrm{AAI}\ge 0.9$ value (the dotted line); however, BBAs also were found in this domain with an $\mathrm{AAI}\ge 0.9$ (refer to Figs. 2 and 4). This fact suggests that a different index other than the AAI must be determined to effectively distinguish the BBAs from the DUST.

Fig. 5

AAI Histogram of DUST over the desert areas referred from the IGBP from May to September 2003 derived from the GLI measurements. The data were selected for AOTs ($0.550\mu \mathrm{m}$) $\ge 1.0$. The AOT values were obtained from the Terra/MODIS deep blue algorithms (MODIS/MOD04_L2 Collection 6). (a) The pixel numbers and (b) cumulative frequency (%).

The AAI as a reflectance ratio of $R$ $(0.412\mu \mathrm{m})/R$ ($0.380\mu \mathrm{m}$) was interpreted in this study for the detection of absorbing particles. Our AAI index was the ratio of the satellite measurements of two-channels in the near-UV that will be available soon for the JAXA/GCOM-C/SGLI satellites. Select data observed by ADEOS-2/GLI in 2003 were used in this study. Although obtaining accurate AAI values requires more precise treatments with the SGLI, the resultant threshold AAI value obtained in this study was $\mathrm{AAI}\ge 0.83$ to distinguish absorbing aerosols such as BBA and DUST from other types of aerosols. In addition, DUST may have higher AAI ($\ge 0.90$) values than BBAs as mentioned above, namely AAI-DUST $>\mathrm{AAI}\text{-}\mathrm{BBA}$. To the extent that has been discussed up to this point, Fig. 6 is an approximate AAI classification with respect to the continental aerosols, such as DUST, BBA, and other types of aerosols (others). The BBA denotes BBA for an $\mathrm{AOT}\ge 1.0$ as already mentioned in Fig. 4. The domain $\mathrm{AAI}\ge 0.9$ in Fig. 6 requires the separation of the BBA from the DUST by a method other than the AAI.

Fig. 6

Classification of continental aerosols in terms of the AAI numerical values (AAI, aerosol absorption index).

3.1.

Discrimination of BBA from DUST

As mentioned in the previous section, only the AAI was unable to distinguish BBAs from DUST in the AAI domain greater than 0.9. Referring to the dust detection algorithm for MODIS measurements proposed by Ciren and Kondragunta,²³ the $2.210\text{-}\mu \mathrm{m}$ band was adopted to detect DUST in this study. The availability of the shortwave infrared wavelengths for the detection of dust aerosols has already been established.²⁴ Then, in a similar manner to Eq. (1), a reflectance ratio ($R$) at a wavelength of $2.210\mu \mathrm{m}$ to that of $0.380\mu \mathrm{m}$ was defined as the DDI:

Figure 7 presents the DDI histogram for the same DUST data as in Fig. 5, namely with the selected mineral dust aerosol data for AOTs ($0.550\mu \mathrm{m}$) $\ge 1.0$. Figures 7(a) and 7(b) show the DDI histograms and the cumulative frequency, respectively, for the summarized data from May to September, 2003. It was found from Fig. 7 that almost all of the DUST (99% denoted by the solid arrows) acquired DDI values $\ge 0.73$. Simply speaking, the values of DUST-DDI were greater than 0.73.

Fig. 7

(a) The histogram and (b) cumulative frequency graphs of the DDI for DUST. The data are the same as the solid line in Fig. 5.

Next, the DDI for the BBAs was considered. Figure 8 shows the BBA-DDI, where the GLI data were the same as that adopted in Fig. 4. Figure 8(a) is the same as Fig. 4(a), except for the DDI. The BBA-DDI values acquired $\sim 0.7$ at low AOTs and rapidly decreased with the AOT until AOT = 1, and then the DDI values slowly changed beyond AOT = 1. At that point, it became possible to say that the BBA acquired lower DDI values than $\sim 0.7$. Figure 8(b) is the same as Fig. 4(b), except for the DDI. Namely, Fig. 8(b) was used to determine the precise DDI values for the BBA. Comparing Figs. 7 and 8 provided us with the conclusion that DDI = 0.73 was the threshold value to distinguish between the BBAs and DUST.

Fig. 8

Detection of the BBA-DDI. (a) BBA-AAI dependency on the AOT based on the same GLI data as that in Fig. 4. (b) Histogram of the DDI for the BBA within the right square domain surrounded by the thick dashed line in (a).

It was concluded from the above discussions that the combined use of the AAI and DDI indices enables us to separate absorbing aerosols into BBAs and DUST. In other words, the classification criteria for the aerosol types were obtained.²⁵

3.2.

Cloud Detection

Here, the difficult issue pertaining to detecting clouds in terms of our AAI and DDI was challenged in addition to discriminating the aerosol type. A comparison between both AAI histograms on a global scale with and without clouds suggested that clouds led to low AAI values in the previous study.¹⁸ To be more precise, the cloud AAI values must be less than those of the aerosols. If this is corroborated, clouds could simply detected be using the AAI; however, it appears that this may be more complicated in practice. The cloud AAI values depend on the season, location, height, type, and relative position of the sun. It should be noted more above all that satellite measures accumulate signals from space, and hence a variety of information is incorporated in the satellite data, especially the cloud heights.

First, the AAI and DDI values were examined for cloud flag pixels. Figure 9 shows two kinds of indices for the AAI and DDI in Figs. 9(a) and 9(b), respectively. In Fig. 9, the cloud flag and cloud optical thickness [COT ($0.670\mu \mathrm{m}$)] were derived from the ADEOS-2/POLDER measurements. The COT behavior of the ice clouds seemed to differ from the water ones, and the AAI and DDI values for the ice clouds converged faster with the COT than with the water clouds. This feature was clearly found in the histogram of Fig. 9(b), where the first and the second peaks corresponded to the ice and the water clouds, respectively. Also, mixed regions of clouds and aerosols are frequently found. In order to further explore issue, more precise handling of the clouds is needed. Focusing on the aerosol retrievals, the task was concentrated to detect regions clearly not attributed to aerosols. The structure and physics of clouds are complex; hence it is impossible to detect clouds, especially optically thin cirrus clouds, using only these two indices. If we ignored such a fine feature, both clouds indices showed stable behavior in the region with a COT ($0.670\mu \mathrm{m}$) $\ge 20$. The clouds with COT ($0.670\mu \mathrm{m}$) $\ge 20$ in Fig. 9 required $\sim \mathrm{1,500,000}$ pixels corresponding to 21% of the total clouds. Figure 9 suggested clouds thick enough to cover the underlying aerosols or the Earth’s surface, and the AAI and DDI acquired values less than equal to 0.65 and 0.6, respectively. In other words, the thick clouds with optical thickness COT ($0.670\mu \mathrm{m}$) $\ge 20$ were detected with AAI values $\le 0.65$ and DDI values $\le 0.6$. It could be say from Fig. 9 that these threshold values of AAI and DDI covered the rather thin clouds having $5\le \mathrm{COT}$ ($0.670\mu \mathrm{m}$) $\le 20$.

Fig. 9

(a) Cloud detection by AAI and (b) DDI from the GLI data in 2003. The top is the AAI and DDI versus COT from the ADEOS-2/POLDER-2 data, and the bottom is the histogram and cumulative frequency (%) for $\mathrm{COT}\ge 20$.

3.3.

Aerosol Classification Criteria

An aerosol type and cloud discrimination chart created by merging Figs. 6–9 in the two-dimensional coordinates of the AAI and DDI is shown in Fig. 10. The yellow, red, black, and light gray colors represent DUST, BBA, clouds, and other types, respectively. Note that “clouds” indicates those thick enough to cover the underlying aerosols or the surface, BBA represents the “genuine BBA” shown in Fig. 4, and “others” include the nonabsorbing aerosols, a mix of several types of aerosols, unknown aerosols, or unclear regions. These ambiguities were resolved to some extent in the following aerosol retrieval process.

Fig. 10

An aerosol type discrimination chart in two-dimensional space. The horizontal and vertical axes denote the AAI and DDI values, respectively.

As an example, Figs. 11(a)–11(c) show the true color composite image, consisting of color-space GLI data (R:0.678, G:0.545, $\mathrm{B}:0.443\mu \mathrm{m}$), the classified aerosol types on a global scale according to the criteria shown in Fig. 10 and the GLI-cloud flag for clouds, respectively, on August 20, 2003, from the ADEOS-2/GLI measurements. Because the discrimination chart in Fig. 10 was the first and fast guide-posting to the precise aerosol characteristics, further retrieval processing was absolutely necessary. However, in Fig. 11(b), the BBA regions seemed to be overestimated although August was the season for the biomass burning (BB) episodes. One of the reasons might be caused by the cloud detection criterion. The AAI and DDI threshold values for cloud classes were restricted in the case of optically thick clouds with $\mathrm{COT}\ge 20$ in Fig. 10 as previously mentioned. As a result, thin clouds might be misinterpreted as BBAs or other types. This fact was suggested from the color composite image in Fig. 11(a). To confirm, Fig. 11(c) presented the global classification of aerosols according to the criteria in Fig. 10 with the exception of cloud classes determined by the GLI cloud flag. Comparing Figs. 11(c) to Fig. 11(b), the cloud portions increased and the BBA- and others-regions decreased. As such, the cloud regions in Fig. 11(b) were completely included in those in Fig. 11(c) and the former occupied 42% of the latter. It was found that with the cloud detection criterion proposed in Fig. 10 it was possible to catch nearly half of the clouds. From the standpoint of aerosol retrieval, it was efficient to ensure that nearly half of the cloud region could be pre-excluded. Through this process, the difficulty of cloud detection and the necessity of the following aerosol characterization were revealed.

Fig. 11

(a) The true color composite image consisting of data on August 20, 2003. Aerosol types were classified from the GLI data, observed on the same day and derived based on the criteria shown in Fig. 10, and those using the GLI-cloud flag to cloud data only, in (b) and (c), respectively.

4.1.

Aerosol Model

The characteristics of aerosols can be represented using several parameters. The most basic parameter is the spectral $\mathrm{AOT}(\lambda )$ at wavelength $\lambda$.²⁶ The Ångström exponent (AE) is derived from the spectral $\mathrm{AOT}(\lambda )$ and is closely related to the aerosol size.²⁷ In general, the AE values from $\sim 0$ to 1.0 indicate large particles (e.g., sea-salt aerosols and soil dusts), whereas values in the range $1.0<\mathrm{AE}<2.5$ indicate particles such as sulfates and those associated with BB.²⁸ The detection of high AE values almost always indicates contamination by small anthropogenic particles. Several other aerosol parameters, such as the size distribution and refractive index, are also derived from the $\mathrm{AOT}(\lambda )$ and sky radiances.^29,30 According to the automatic classification of accumulated NASA/AERONET data, atmospheric aerosols are classified into the following six categories: (1) BB, an aged smoke aerosol consisting primarily of soot and organic carbon; (2) rural (RU), referred to as a clean continental aerosol; (3) continental pollution (CP), representing anthropogenic aerosols, including various species of sulfate- (${\mathrm{SO}}_4^{2-}$), nitrate- (${\mathrm{NO}}_3^{-}$), OC, ammonium (${\mathrm{NH}}_4^{+}$), and soot; (4) dirty pollution, consisting of the same aerosol types as CP, but at significantly higher levels; (5) desert dust, assumed to be mostly mineral soil; and (6) polluted marine, consisting primarily of sea salt with traces of CP.³¹

First, we investigated the aerosol sizes. The size distributions of these six aerosol types had two modes (fine and coarse) in a bimodal log-normal distribution of the particle volume with six parameters (volume concentration, mode radius, and the standard deviation of the fine and coarse mode particles). Too many parameters are excessive for the retrieval of the optimized aerosol sizes at a global scale. Therefore, a simplification was made for the aerosol size distribution function.²¹ As a result, the size distribution function for continental aerosols ($V$) can be approximately expressed in a simpler form, defined by the unique variable of the fine particle fraction ($f$) of the volume concentration³²:

The next aerosol characteristic of interest is the refractive index. It is reasonable to consider that a mixture of various aerosol types exists in nature. Therefore, many studies have been conducted with respect to the mixing of various particle types.^33–36 Although particle mixing is an issue to be considered as well as understood, a simple homogeneous internal two-component mixing model was adopted here, using the Maxwell Garnett mixing (MGM) rule.³⁷ The MGM rule provides a complex refractive index calculated as follows:

Figure 12 presents the block flow of our aerosol retrieval method from the satellite data. Part “C” represents the debated section of the aerosol type classifications based on the near-UV measurements mentioned in the previous section. The key point of this study is that the advance classification of aerosol types using near-UV data is efficient for aerosol retrievals. This is important because radiation simulations require long computing times for the various aerosol models at each satellite data pixel. Best of all, useful information can be obtained beforehand about aerosol types using just the simple AAI and DDI color ratios that provide a rough overview of the aerosol characteristic distribution and can immediately be applied on a global scale. It is reasonable to consider that aerosols are naturally mixed and heterogeneous, and hence part of “E” in Fig. 12 corresponds to an exception-handling procedure for cases that do not conform to the prepared aerosol types. This exception handling can be compensated for by using interpolations between some of the prepared aerosol types; however, further work is needed in order account for these exceptional cases to increase the accuracy of the retrieved aerosol properties.

Fig. 12

Block flow diagram for the satellite data aerosol retrievals. Section “C” represents the debated section of the aerosol type classification based on the near-UV measurements, and “E” is an exceptional nonconforming case for the prepared aerosol models.

4.2.

Retrieval of Aerosol Properties in Heavy BBA Events

During aerosol episodes, extremely dense aerosol concentrations can occur. In other words, very high AOT is assumed during heavy aerosol episodes because the AOT is proportional to the amount of aerosols in the atmosphere. The basic framework for aerosol retrievals from satellite data is roughly divided into two parts: satellite data processing and radiation simulation calculations in atmosphere of the Earth involving aerosols and molecules. It is reasonable to consider that AOT values are very high during aerosol episodes where the incident solar radiation induces multiple interactions with atmospheric aerosols due to the dense radiation field. Unfortunately, precise simulation of multiple light-scattering processes requires very large computation times. The RT simulations for such cases have been treated with a process called the MSOS.^12,40 The MSOS effectively calculates the upward intensity of radiation at the TOA (i.e., the reflectance of the optically semi-infinite atmosphere model as $\mathrm{AOT}\approx \infty$). Consequently, the ground-surface reflection property can be ignored because the optical thickness of the atmosphere is too high to be affected by the radiation interaction of the atmosphere with the ground surface. In practice, ADEOS-2/GLI measurements over the area around the border of Mozambique and South Africa on August 20, 2003, where typical BBA episodes were occurring, were selected to satisfy the MSOS conditions mentioned above. The target acquired values of AAI = 0.95 and DDI = 0.20. It was observed from Figs. 4(a) and 8(a) that the values of both indices satisfied the high AOT conditions. This fact suggested that the MSOS was applicable for radiation simulation in this target.

The aerosol retrieval targets were preferably selected around AERONET sites for later validations. As noted above, the BBA model is described by only the $f$ and $g$ parameters. We compared the GLI measurements (represented by dots) with the simulated reflected intensity values calculated by the radiation simulations in a two-channel diagram with wavelengths of 0.443 and $0.545\mu \mathrm{m}$ in Fig. 13. Therefore, the two $f$ and $g$ parameters, i.e., the aerosol size distribution and refractive index, respectively, were used for the radiation simulations based on the MSOS. Accordingly, the optimized values ($f*,g*$) were estimated from the two wavelength planes. Once the $g$ value had been determined, the refractive index was calculated using the MGM rule [see Eq. (4)]. A detailed description of this procedure was presented in previous studies.^6,21 The dots in Fig. 13 denote the ADEOS-2/GLI data over the target on August 20, 2003. Furthermore, the desired aerosol property of the imaginary part of the complex refractive index at $0.38\mu \mathrm{m}$ was derived from the comparison of the reflectance value by the RT simulations with the GLI/Band-1 ($0.380\mu \mathrm{m}$) measurements. The parameters and retrieved results of the aerosol characteristics are summarized in Table 1. The retrieved parameters values ($f*$) indicated small particle sizes. The obtained refractive index suggested the particles were weakly absorbing aerosols. Note that the imaginary part of the refractive index in the near-UV was rather large. In other words, the BBA absorption in the near-UV due to forest fires was observed.

Fig. 13

Calculated reflectance values based on the MSOS RT algorithm for our proposed aerosol models, described by two parameters ($f,g$) in a two-channel 0.443 and $0.545\mu \mathrm{m}$ diagram. The solid and dashed curves denote the results for various values of $f$ and $g$, respectively. The dots denote the ADEOS-2/GLI data for the case study target. The values ($f*,g*$) denote the retrieved values optimized for the GLI measurements.

Table 1

Retrieved aerosol characteristics based on the MSOS using the ADEOS-2/GLI measurements at 0.545, 0.443, and 0.380  μm.

The retrieved aerosol properties presented in Fig. 13 and Table 1 should be compared with the ground measurements for validation. However, the simulations were unfortunately limited to the severely hazy case, and therefore the ground measurements were not available. Then, reproduction calculations of the MSOS using the retrieved aerosol parameters presented in Table 1 were performed and compared with the original satellite data for validation of the present retrieval. The maximum error fell within 3%. The MSOS method has been applied to other aerosol events in previous works, such as a forest fire in the Amazon¹² and a severe air pollution case in east-central China.²¹ In these studies, the difference between the resimulation results and the MODIS measurements was within 3%. The validation of the retrieved aerosol properties around the target was examined together with the finite atmosphere treatments and discussed in the next section.

4.3.

Aerosol Retrieval with ADEOS-2/POLDER-2

Satellite remote sensing usually provides the spatial distribution of aerosol properties. Nevertheless, in the previous section, the aerosol retrievals were limited to point analyses due to the heavy aerosol concentration restrictions. In this section, the retrieval target was expanded in the vicinity of the target point. In addition to the results obtained in the previous section, we estimated two more aerosol properties, i.e., optical thickness (AOT) and AE, based on the polarization measurements at 0.670 and $0.865\mu \mathrm{m}$ observed by the POLDER-2 sensor on ADEOS-2. It has been accepted that polarization information is very effective for aerosol retrievals over land. The retrieval algorithm was similar to previous studies,⁴¹ and is referred to as the method of vector radiative transfer (MVRT) for convenience, and the polarized reflectance by the bottom surface was interpreted by the model proposed by Bréon et al.⁴² It was found that the refractive index values were the same in the visible wavelength except for the near-UV, but that the size of the aerosols differed from each other. The refractive index was fixed at $1.44\text{-}0.010i$ around the target (Table 1). Accordingly, two aerosol parameters (AOT, $f$) were retrieved in a two-channel polarized radiance of 0.670 and $0.865\mu \mathrm{m}$ provided by the ADEOS-2/POLDER-2.

Our algorithm was applied to the BB scene in the vicinity of the same point and time treated in the previous section. Figure 14 presents a composite image captured by Terra/MODIS on August 20, 2003, over southern Africa. It can be seen at a glance that the diagonal right half of the image is covered by smoke.

Fig. 14

True color composite image consisting of Terra/MODIS data (R:0.645, G:0.555, B:0.469 μm); MOD021KM collection 6.1) captured on August 20, 2003, over southern Africa.

The high AOT regions were widely distributed around the central region in Fig. 15(a). The AE value that was calculated from the AOT and $f$ around this high AOT region was greater than 2 [Fig. 15(b)]. This implied that the small particles were in the atmosphere. For validation of the retrieval results, the Terra/MODIS products (MOD04_L2 Collection 6.1) were inferred [see Figs. 15(c) (AOT at $0.5\mu \mathrm{m}$ and 15(d) (AE)]. The correlations between the AOT ($0.5\mu \mathrm{m}$) retrieved results with the ADEOS-2/POLDER-2 and MODIS products are shown in Fig. 15(e). Note that the AOT ($0.5\mu \mathrm{m}$) was derived from the retrieved AOT ($0.55\mu \mathrm{m}$) and the AE. This event was also measured by the Skukuza AERONET station [white circle in Fig. 14(a)]. The instrument measured direct sun light and provided highly accurate AOTs as well as AE information.²⁹ The lower right panel of Fig. 15(f) shows the AERONET-AOT ($0.5\mu \mathrm{m}$) time series and AE before and after the BB event. The event started on August 19 and ended on the 20th. The red filled triangular symbols in Fig. 15(f) show the averaged AOT and AE values from POLDER-2 that passed over the station around 8:17 (UTC). The error bar for each symbol represents the standard deviation of the retrieved results within $\pm 0.1\mathrm{deg}$ around the AERONET station. Our retrieved results maintained a 10% accuracy, over the Skukuza AERONET station. The retrieved results were comparable to the ground AERONET measurements. The red filled circles in Fig. 15(f) show the MODIS products as shown in Figs. 14(c) and 14(d). Compared with the AERONET measurements, the MODIS-AOT results were better than our POLDER-2 results, but our AE was better than the MODIS-AE. The aerosol retrieval algorithm used by MODIS is different from ours. The comparison of the results by different approaches with different data was considered more interesting than a validation.

Fig. 15

(a) Retrieved distribution of the AOT ($0.5\mu \mathrm{m}$) and (b) AE based on the MVRT simulation in a two-channel 0.670 and $0.865\mu \mathrm{m}$ polarized radiance provided by the ADEOS-2/POLDER-2 over southern Africa on August 20, 2003. The Terra/MODIS products (MOD04_L2 Collection 6.1) are presented in (c) AOT ($0.5\mu \mathrm{m}$) and (d) AE. The correlation between the AOT ($0.5\mu \mathrm{m}$) retrieved results with the ADEOS-2/POLDER-2 and MODIS products are shown in (e). (f) The AERONET-AOT ($0.5\mu \mathrm{m}$) time series (in the upper graph) and AE (lower) before and after the BB event. The red triangles and the circles denote the retrieved value with the MRVT from the POLDER-2 and MODIS product, respectively.