3.1. Regional Validation of POLDER-3 Aerosol Retrievals
 Comparisons of POLDER-3 AOD against τfine(r≤0.3μm)AER are shown in Figures 2a, 2b, and 2c, respectively, for 865, 670, and 440 nm channels. The error bars on the y-axis represent the spatial standard deviation computed over the 0.5° × 0.5°-size window. Similar comparison for 440 nm is presented without error bars, since the standard deviation was very similar to that of 865 and 670 nm. Our results demonstrated overall good agreement between POLDER AOD and τfine(r≤0.3μm)AER. The correlation coefficients are 0.92, with the 95% confidence intervals (CI) from 0.90 to 0.94, for both of 865 and 670 nm. For 440 nm, the correlation coefficient is 0.908 and the corresponding 95% CI is 0.88–0.93. The slopes, calculated by the robust regression method on the basis of iteratively reweighted least squares, are 0.84, 0.81, and 0.7 for 865, 670, and 440 nm, respectively. The observed decrease of the slope with wavelength can very likely be explained by extrapolation validity as well as limitations in the retrieval assumptions such as departure between modeled and true absorption, particle shape, and residual surface contribution.
Figure 2. Comparisons between POLDER-3 AOD and AERONET fine-mode AOD (r ≤ 0.3 μm) at (a) 865 nm, (b) 670 nm, and (c) 440 nm over 14 sites located in East Asia from March 2005 to June 2008. POLDER-3 AOD values at 670 and 440 nm were computed from POLDER-3 AOD at 865 nm and the corresponding Ångström exponent. Error bars in Figures 2a and 2b show spatial variation of POLDER pixels.
Download figure to PowerPoint
 More detailed information (site by site) from robust regression is provided for the 865 nm channel in Table 2. To ensure the reliability of regression results, we skipped four sites having less than 5 days with available data, including Kanpur (4 days), Mukdahan (4 days), Pimai (4 days), and Taipei_CWB (2 days). Averaged AERONET and satellite AOD for each site are presented in the second and third columns in Table 2, along with the root mean square (RMS) indicating daily variation. All statistical data are arranged in order of increasing averaged AERONET AOD to partially reflect the aerosol pollution level for each site. For most of the sites, the averaged POLDER-3 AOD is rather comparable with the ground-based AOD. The results of robust regression analysis are also presented in Table 2, where R, together with its 95% CI in the fifth column, represents the correlation coefficient and N represents the number of days with both satellite and ground-based data. The SD provided in the sixth column of Table 2 is actually the residual standard deviation, suggesting the degree of dispersion from the actual data to the linear regression line.
Table 2. Statistics and Robust Regression Analysis Between PARASOL AOD and Fine-Mode AOD from AERONET Measurements at 865 nm (March 2005 to June 2008)
|Site||Mean Value||Robust Regression|
|AERONET||PARASOL||R||95% CI for R||SD||slope||intercept||N|
|Dalanzadgad||0.01 ± 0.01||0.02 ± 0.01||0.16||(−0.42, 0.65)||0.01||0.38||0.01||13|
|Anmyon||0.04 ± 0.01||0.04 ± 0.01||0.84||(0.40, 0.97)||0.01||0.77||0.01||9|
|Osaka||0.05 ± 0.02||0.05 ± 0.03||0.61||(0.12, 0.86)||0.02||0.97||0.00||14|
|Shirahama||0.05 ± 0.03||0.05 ± 0.04||0.72||(0.46, 0.88)||0.03||0.98||0.00||25|
|Xinglong||0.06 ± 0.04||0.04 ± 0.04||0.93||(0.73, 0.98)||0.01||0.78||−0.01||10|
|SACOL||0.07 ± 0.03||0.03 ± 0.02||0.67||(0.44, 0.82)||0.02||0.56||−0.01||37|
|XiangHe||0.13 ± 0.13||0.11 ± 0.11||0.96||(0.93, 0.98)||0.03||0.85||0.00||32|
|Beijing||0.13 ± 0.13||0.11 ± 0.10||0.93||(0.89, 0.96)||0.04||0.73||0.01||52|
|Taihu||0.14 ± 0.06||0.17 ± 0.09||0.87||(0.60, 0.97)||0.05||1.27||0.00||12|
|Bac_Giang||0.15 ± 0.14||0.17 ± 0.14||0.97||(0.78, 1.00)||0.03||0.95||0.03||6|
 For 9 of 10 sites, the correlation coefficients (R) are better than 0.6 and the majority of slopes are close to 1. Nevertheless, there are three sites with 95% CI for R wider than 0.5, which probably were induced by the few available data.
 For Dalanzadgad site, the correlation coefficient of 0.16 shows no strong correlation between POLDER-3 retrievals and AERONET inversions. The main explanation may be attributed to the very low level of AOD over this site, which is located in the Gobi areas of Mongolia, together with restrictively limited AOD range. The atmosphere is generally quite clear and clean, except during dust events that are estimated to occur 30 times per year [Mandakh and Khaulenbek, 2002; Qian et al., 2006]. The multiannual AERONET total AOD at 865 nm measured over Dalanzadgad was only 0.07; this is much lower than the total AOD recorded over Beijing, which was 0.41. In most cases, dust events generating mainly coarse, nonspherical particles are very weakly polarizing light, therefore making these particles almost nondetectable with POLDER current retrievals. The distribution of blue dots in Figure 2 also partly reflects the particular characteristic of aerosols over this site.
 Overall, our validation results show that POLDER-3 aerosol retrievals over land are very consistent with the ground-based fine-mode AOD for most of the AERONET sites of East Asia, when the fine mode is defined with a maximum radius value of 0.30 μm. We observed that 9 of 10 AERONET stations have correlation coefficients above 0.6 and four sites are above 0.9.
 These results indicate that POLDER aerosol products over land are quite relevant for analyzing spatial and temporal distributions of fine particles over East Asia. This analysis is developed in the next section.
3.2. Characteristics of POLDER AOD Distribution Over East Asia
 East Asia is an important source of both natural and anthropogenic aerosols because of its geographical characteristics and the rapid growth of its economy. However, our knowledge of the spatiotemporal distributions of aerosols is still limited because of the lack of long-term and large-coverage observations [Choi et al., 2009; Li et al., 2002; Xu et al., 2004; Park et al., 2005; Li et al., 2008]. We therefore considered POLDER-3 retrievals from March 2005 to February 2009 to characterize aerosol patterns over East Asia as well as seasonal variability.
 To provide a first overview of the aerosol distribution over this very large region, we show in Figure 3a a 4-year POLDER-3 averaged AOD. Information on the percentage of days with available aerosol retrievals is also given in Figure 3b. Percentages lower than 100% are due to cloud cover (e.g., the whole area, especially Southeast Asia and the southern part of China), to topographic features of the underlying surface (e.g., the Tibetan Plateau), or to missing data (the white color in the inland areas). Moreover, to aid interpretation, the locations of several typical areas mentioned below are marked in Figure 3c.
Figure 3. (a) Map of averaged AOD at 865 nm and (b) pixel frequency based on POLDER-3 observation over a 4-year period (March 2005–February 2009). The frequency is calculated as the number of days with successful PARASOL aerosol retrievals in clear sky divided by the total number of calendar days during the study period. (c) Illustration of typical areas mentioned in this paper, with the Sichuan Basin presented as a shadow. Country names are given in red, including red “1” and “2” for Laos and Bangladesh, respectively, while the names of provinces in China (shaded boundaries) are written in light green.
Download figure to PowerPoint
 As shown in Figure 3a, the highest average AOD mostly appears in the Sichuan basin and the southeastern region of China, particularly around the middle and lower valleys of the Yangtze River, the Delta of the Pearl River, and the North China Plain, which are densely populated, highly industrialized, and economically developed. Additionally, there are also several other areas with high AOD value, such as the southern Himalaya region in India and the areas around the border between China and Laos. According to our study, the maximum of POLDER-3 AOD value over East Asia has been estimated as 0.25, whereas the mean value is 0.05. Excluding the very clear regions over the Tibetan Plateau and the northwest part of East Asia, which are beyond the scope of this study, the low AOD values primarily appear over northeast China, including Heilongjiang (HLJ), Jilin (JL) Province, and Inner Mongolia (IM).
 Similar patterns have also been demonstrated by Luo et al.  on the basis of the analyses of total AOD retrieved from a network of 46 solar radiation stations as well as with MODIS total AOD within the August 2000 to April 2003 period [Li et al., 2003b].
 However, there are some areas with high AOD derived from MODIS in the western part of Tibet, the Tarim Basin of Xianjiang (XJ), and the Qaidam Basin in Qinghai Province (QH) mentioned by Li et al. [2003b] that are not highlighted by POLDER. This difference is likely related to the specificity of the POLDER-3 retrievals over land, which are not sensitive to the coarse mode fraction characterizing dust particles occurring in these regions. There is another visible feature in the Sichuan Province, where the highest AOD, appearing in the east, contrasts sharply with the lowest values retrieved in the western part of the province. This remarkable difference is likely due to the unique topography of Sichuan Province. The elevation in the west part is almost 10 times higher than the elevation in the east. The terrain with higher elevation certainly limits aerosol transport and therefore exhibits lower AOD.
 A more precise examination of the seasonal variation of aerosol patterns is presented in Figure 4, in which the four seasons, spring (MAM), summer (JJA), autumn (SON), and winter (DJF), are separated. The white color in the inland regions again represents areas in which the valid retrievals were not obtained.
Figure 4. Spatial distribution characteristics of POLDER-3 AOD at 865 nm over East Asia for (a) spring (MAM), (b) summer (JJA), (c) autumn (SON), and (d) winter (DJF).
Download figure to PowerPoint
 In spring (Figure 4a), the highest AOD values mostly appear in the Sichuan Basin, the Pearl River Delta in China, the areas around Stanovoy Range in Russia (the north of Heilongjiang Province in China) and Southeast Asia, especially in the northern part of Laos and its surrounding areas in Burma (B), Thailand (T), and Vietnam (V), which nearly link up with the high AOD value area in the southern Guangxi Province (GX) of China. The high AOD values in Sichuan Basin seem consistent with its high population density and its unique topography benefiting the retention and accumulation of aerosol pollutants [Li et al., 2003a], while the high values over Southeast Asia and Russia are very likely due to slash-and-burn farming and forest fires [Kim et al., 2004; Murdiyarso and Lebel, 2007; Padoch et al., 2007; Damoah et al., 2004; Lee et al., 2005; Kanaya et al., 2003]. Along with local aerosol emissions, fine-mode particles generated by the biomass burning in Southeast Asia may be transported to the south of China following the dominant westerly wind during the spring; this is likely the cause of the high AOD in Guangxi Province and the Pearl River Delta [Li et al., 2003b; Kim et al., 2007].
 In summer (Figure 4b), though, the AOD values significantly decrease around Sichuan Basin. We observed an increase of AOD in the eastern part of China, especially around the border of Shandong and Hebei, likely due to anthropogenic pollution. The decreasing AOD trends in the Sichuan Basin are possibly associated with the wet deposition caused by the frequent rainfall, which washes out the majority of ambient respirable suspended particulates and shortens their lifetime. High AOD values detected in spring in Southeast Asia are barely visible in summer because of the reduced amount of aerosol retrievals resulting from the heavy cloud cover during the monsoon period. Additionally, the strong rainfall, together with the high cloud cover over Southeast Asia during summer, prevents the biomass burning and also washes out pollution. In contrast, most of southern Guangxi Province still has high AOD during this time because of the aerosol emissions from the heavy industrialized areas and the relatively stagnant conditions over this region.
 In autumn (Figure 4c), AOD slightly increases in eastern and southern China, the Sichuan Basin, and the region from the north part of India to the south of the Himalayan Mountains, but decreases in Guangxi Province of China.
 During the winter (Figure 4d), AOD values increase and reach their maximum in the Sichuan Basin, the northeastern part of India along the Himalayan Mountains and Bangladesh. In the Sichuan Basin, the observed AOD increase can be explained by poor dispersion conditions and heavy local industrial and vehicle emissions. Near the base of the Himalayan Mountains in northern India and Bangladesh, a large AOD, possibly related to the pollution accumulation due to temperature inversion in the boundary layer, is observed. During the winter, cold air flows from the mountains down to the plains, making the air near the ground cooler than the air above it. This probably traps aerosols from agricultural fires and cities near the ground [Marshall, 2005; Di Girolamo et al., 2004].
 On the basis of the above analysis, we conclude that aerosol patterns detected by POLDER-3 over East Asia are clearly linked with human activities. Most of the regions characterized by large AOD values are located in southeastern China and northeastern India, areas characterized by developed industries and rapid economic growth. The north part of Southeast Asia and the Stanovoy Range in Russia only exhibit a high AOD during spring, when representative aerosol emissions are dominated by biomass burning activities.
 Seasonal fluctuations in POLDER-3 AOD vary between different geographical locations. For example, south-central China and the northeastern region of India along the Himalayan Mountains reach their maximum AOD during winter, while the maximal AOD in Southeast Asia and the Stanovoy Range in Russia occurs during spring. The North China Plain reaches its maximum of AOD in summer, while in southern China (the middle and lower valley of the Yangtze River), the highest AOD occurs in autumn. The main factors controlling regional variability are the differences in both local emissions and seasonal meteorological conditions in the different geographical regions [Kaufman et al., 2002].
3.3. Year-by-Year Evolution of Summer Aerosol Loads in the North China Region
 China, centrally located in East Asia, shows higher emissions of both natural and anthropogenic aerosols. The mean aerosol total AOD in China is about twice the global continental mean value [Li et al., 2002]. Beijing, the political and cultural center of China, is even recognized as one of the world's 10 most polluted cities and also shares with Mexico City the distinction of being the world's most polluted capital. During the past 30 years, Beijing has experienced severe aerosol pollution predominantly caused by rapid economic development, population expansion, and urbanization, as well as some secondary problems such as increasing traffic density, a high consumption of coal, flourishing construction activities, and dust storms from deserts [Sun et al., 2004; Chan and Yao, 2008]. As an indicator of air quality, Aerosol Optical Depth in Beijing has been revealed to reach its maximum during summer because of atmospheric stagnation events that are typically associated with high temperatures [Fan et al., 2006; Xia et al., 2006, 2007]. According to the previous studies, the severe aerosol pollution in Beijing is a result of not only internal emission sources within Beijing but also the atmospheric pollution of surrounding provinces [Xu et al., 2002, 2003, 2005, 2006]; this is particularly true during summer as a result of a combination of southerly winds and the topography of the surrounding regions [He et al., 2009]. Numerical simulations have also suggested that 34% of PM2.5 [Streets et al., 2007] and 40% of PM10 [Chen et al., 2007] in summer over urban Beijing could be generated by regional emissions. To improve the air quality in Beijing and ensure a healthier atmosphere for athletes and spectators during the 2008 Summer Olympic Games, China introduced extensive provisions and enforced emission reductions in Beijing and its surrounding areas before and during the event. Such an effort provided a unique opportunity to study the anthropogenic contribution to the atmospheric aerosol loads. Therefore, we now focus our analysis on aerosol variability during the summer in Beijing and its surrounding provinces, which are termed “North China” in this article. In the following analysis, we investigate interannual AOD variation from 2003 to 2009 as seen using POLDER instruments. More precisely, our area of interest, North China, includes all regions of China within the limits of 32°N–42°N in latitude and 110°E–120°E in longitude (Figure 5a), including Beijing (population 11.5 million) and Tianjin (9.3 million) municipalities [National Bureau of Statistics of China (NBS), 2004]. Surrounding provinces, including Hebei, Shandong, Shanxi, and Henan, which are heavily populated, urbanized, and industrialized, are also included. In the areas surrounding Beijing, emission controls on stationary sources and vehicles are not as stringent as in Beijing itself, and emission rates are therefore higher. Rural biomass burning has also been indentified as an important contributor to fine PM concentrations in Beijing [Duan et al., 2004; Streets et al., 2007]. Emissions from these nearby sources, as well as more distant ones, are subject to chemical reactions during transport on prevailing winds, forming secondary species that enter the entire region and are added to the local pollution of Beijing [Han et al., 2005; Hatakeyama et al., 2005; Luo et al., 2000; Mauzerall et al., 2000].
Figure 5. (a) Study area referred to as “North China” in this paper and (b) the map of Beijing showing main urban districts (shaded area), interior suburban districts (dark shaded boundary), the far north districts and counties (light shaded boundary), and the locations of Beijing and XiangHe sites (black points). The box in Figure 5b highlights the Beijing City defined and analyzed in this paper.
Download figure to PowerPoint
 Utilizing both POLDER-2 (2003) and POLDER-3 (2005–2009), we averaged, over the North China area, the AOD at 865 nm from June to August for each available year. The results are presented in Figures 6a–6f. Moreover, complementary statistical parameters such as median, mean, standard deviation, first and third quartiles, as well as the maximum and minimum, are given in Figure 7.
Figure 6. Spatial distribution characteristics of POLDER AOD at 865 nm over North China (a) in the summer of 2003 and (b–f) from 2005 to 2009.
Download figure to PowerPoint
Figure 7. Boxplots of averaged POLDER AOD at 865 nm in the summer of 2003 and 2005–2009 over North China. In each box, the central bar is the median, and the lower and upper limits are the first and the third quartiles, respectively. The error bars on the y-axis indicate 1.5 times the spatial variation (SD).The associated maxima and minima are indicated by asterisks. The square symbols indicate mean values.
Download figure to PowerPoint
 A significant interannual variation in AOD can be observed in both Figures 6 and 7. The maximum AOD was observed in 2003 (0.19), whereas the minimum AOD, which was roughly half of the maximum (0.10), occurred in 2006. The summer of 2008 appears to have been a moderately polluted season and was followed in 2009 by a new minimum AOD (0.08).
 A similar trend was observed upon investigation of the regions covered with high AOD values (Figures 6a–6f). During the summer of 2003, high AOD values covering almost the entire Shandong Province were linked with those measured over the Southern Hebei Province, Beijing and Tianjin municipality, the northeast region of Henan Province, as well as the northern Anhui and Jiangsu Province. The strip of plains in Shanxi Province between the Taihang (East) and Lüliang (West) Mountains also shows a higher AOD in the summertime. During the summers of 2005 and 2006, the area covered with AOD values greater than 0.20 decreased significantly and was mainly localized along the border of Hebei and Shandong provinces, Tianjin municipality, and the border area between Shandong and Jiangsu provinces. However, the surface covered with higher AOD levels expanded greatly into the southwest regions during the summer of 2007, indicating another heavily polluted season. In the seasons following, emission control strategies were enforced in Beijing and its surrounding provinces to ensure better air quality for the 2008 Olympic Games. The area covered with AOD values larger than 0.20 decreased slightly in the summer of 2008 when compared to 2007. The area covered with high AOD values decreased again in the summer of 2009, when high AOD values only appeared over the north part of Anhui and Jiangsu provinces.
 Overall, our analysis indicates that the fine-mode AOD in North China is quite variable with the extreme values separated by a factor of nearly 2 between years. In addition, our results suggest that the summer of 2008 was a moderately polluted season.
 Furthermore, the differences observed between the third and the first quartiles in Figure 7, as well as the standard deviations, are quite large, indicating the high level of spatial variability of AOD.
 As shown in Figures 6a–6f, the high AOD values appear predominantly over the southeast plains of our analyzed area, where many large urban centers with dense populations and developed industries are distributed, whereas the plateau and mountains in the northwest typically exhibit AOD values lower than 0.10. The plateau and mountains with high elevations presumably restrict the spreading of aerosol pollution. The same explanation can be given for the low AOD in the northwest part of Beijing City, which contrasts with the high values in the main southeast urban regions.
 To obtain a more detailed picture of the aerosol variability in North China during the summer months, we investigated the interannual evolutions of monthly mean of AOD at 865 nm for June, July, and August separately in Figure 8. The results in Figure 8 demonstrate that June is the month with the highest AOD value, as well as demonstrating the highest level of AOD variability. In July, the variability of AOD generally appears to be similar to that of June, but with reduced amplitude. It should be noted that AOD for both June and July start increasing in 2006 and do not decrease until 2009. In 2008, AOD for both June and July still show an increase, whereas a decreasing trend was observed for the summer as a whole, as shown in Figure 7. August shows a significantly different interannual AOD variation. The monthly mean AOD of August increases continuously from 2003 until 2008, when it appears strongly decreased. Moreover, in contrast to June and July, the mean AOD increased in August 2009. The interannual contrasts observed over the North China region for June, July, and August can be explained by the relative contributions of complex processes such as stagnant synoptic meteorological patterns, secondary aerosol formation, and hygroscopic growth of aerosols and smoke aerosols by regional biomass burning [Kim et al., 2007].
Figure 8. Interannual evolution of monthly POLDER Aerosol Optical Depth at 865 nm in June (hollow triangle), July (filled triangle), and August (filled circle) over the whole of North China from 2003 to 2009.
Download figure to PowerPoint
 Finally, we focused on smaller-scale analysis, specifically in Beijing City, which is defined by a 0.5° × 0.5°-size area centered on the Beijing AERONET site (latitude 39.98°N, longitude 116.38°E).
 At this new scale of analysis, ground-based total AOD and fine-mode AOD as measured and derived from AERONET data again served as a reference (Figure 9). Considering the sparse available data obtained in the summer of 2008 over the Beijing site, we also include the XiangHe AERONET site (latitude 39.75°N, longitude 116.96°E), which has been in operation since 2005 and is located about 70 km southeast of Beijing. This site is classified as a rural area with respect to surface cover; however, AOD and other aerosol properties (size distribution, absorption) over the XiangHe site are nearly the same as those over the Beijing site [Eck et al., 2005; Xia et al., 2005], though the surface properties are different. Combining these two sites enhances the reliability of our comparison. Our study area, covering the core city districts of Beijing, is highlighted by a box in Figure 5b, along with the locations of AERONET sites of Beijing and XiangHe. As described previously, we use AERONET Level 1.5 products to obtain more data. The ground-based parameters are plotted in Figure 9 result from an averaging of the XiangHe and Beijing sites. The mean AOD from each site is weighted by the number of available observation from each site. As shown in Figure 9, both POLDER AOD values and ground-based fine-mode AOD generally follow very similar year-to-year evolution to total AOD provided by AERONET. A very slight decrease in the fine mode AERONET AOD can, however, be observed in 2008, whereas the POLDER-3 AOD slightly increases along with the total AOD.
Figure 9. Year-to-year evolution of AOD at 865 nm from POLDER observations (filled pentagram) and AERONET measurements around Beijing areas considering both Beijing and XiangHe ground-based sites during summer months, where the total and fine-mode AOD are represented by filled circles and hollow pentagrams, respectively.
Download figure to PowerPoint
 The good agreement between AOD variability as detected by POLDER and the ground-based one demonstrates the potential of POLDER for monitoring aerosols. Moreover, the combination of POLDER, demonstrated to be sensitive to a fraction of the fine-mode aerosol, and MODIS, known to be sensitive to the entire size distribution [Remer et al., 1996, 2005], could enable derivation of aerosol coarse-mode optical depth in the context of the A-Train observatory.
 More detailed information for June, July, and August over Beijing City are presented separately in Figure 10 for both total AOD and fine-mode AOD. As shown in Figure 10, monthly POLDER AOD, as well as monthly AERONET total AOD and fine-mode AOD, varies strongly. Specifically regarding AERONET, the maximum total AOD was detected in June 2008, while the minimum was detected in August 2004 (the next is June 2009, more recently). The maximum fine-mode AOD appeared in June 2002 (June 2007 next), and the minimum appeared in June 2009.
Figure 10. Interannual evolution of monthly averaged AOD at 865 nm from POLDER (inner bar) observations in June (light shaded), July (dark shaded), and August (black) as well as the associated total (hollow bar) and fine-mode AOD (short line) from AERONET measurements for the available years from 2002 to 2009 over Beijing City.
Download figure to PowerPoint
 The lack of available POLDER data in 2002 and 2004 may partially bias the comparison of year-to-year evolution with AERONET AOD. Nevertheless, the monthly POLDER AOD in summertime showed a maximum in June 2003 and a minimum in June 2009.
 However, for the fine-mode AOD, the aerosol data did not show such a clear pattern. Focusing on the years around the Olympic Games (2007–2009), we observed that the AERONET total AOD in August varied in a trend opposite to that of June and July. The total AERONET AOD decreased very slightly from 2007 to 2008 and then increased from 2008 to 2009. The decrease of total AOD in 2008 has also been observed with MODIS [Cermak and Knutti, 2009]. For the fine-mode fraction, the agreement between POLDER and AERONET are good for June and July 2008, whereas our results show that POLDER AOD were an overestimation of the AERONET values in August. Finally, both ground-based and satellite fine-mode AOD values are consistent and clearly detect the strong AOD decrease in 2009 compared to 2008.