Regression Analyses between Recent Air Quality and Visibility Changes in Megacities at Four Haze Regions in China

The Chinese government has put forward a series of aggressive control measures to tackle environmental problems, such as poor visibility, since the first year of its 11th five-year plan (2006–2010). Recently recorded visibility, air quality and meteorological data in four major megacities (Beijing, Shanghai, Guangzhou and Chengdu) in different haze regions (and climatic zones) of China were analyzed with the aim of evaluating the extent to which the control actions have affected these measures. The ambient concentrations of three major air pollutants (SO2, NO2 and PM10) in these cities all decreased in the years 2005–2009. However, improved visibility was observed only in Beijing and Guangzhou; it remained steady in Shanghai, and showed a decreasing trend in Chengdu. The results highlight the fact that the correlation between air quality and visibility is complex. Optimal empirical regression models were developed, based on measured air quality and meteorological parameter data, to better isolate possible causal correlations between visibility and air quality, as well as meteorological conditions. Our results show that the improvement in visibility in both Beijing and Guangzhou was mainly due to the reduced PM10 concentration. In Guangzhou, improved atmospheric visibility was also helped by a reduction in SO2 concentration in winter. In contrast, lower wind speed, together with possible changes in fine particle concentration and composition, could explain why no improvement in visibility trend was found in Shanghai or Chengdu.


INTRODUCTION
Atmospheric visibility degradation is a key issue in climatology and air pollution studies.It exerts adverse effects on humans' lives, such as in highway crowding and restricted aircraft movements.Recently, visibility in clear sky was found to be decreasing over land globally from 1973 to 2007 (Wang et al., 2009).In urban areas, visibility is a highly relevant visual indicator of air pollution level.Visibility reduction, other than on foggy days, is always associated with poor air quality caused by intensive emissions of air pollutants, which generate adverse health effects in humans (Watson, 2002).
Dramatic economic and industrial developments as well as vigorous urbanization in China have led to increased emission of various pollutants from urban areas, making visibility degradation one of the most severe environmental problems in such a rapidly changing country.Twenty-five years ' visibility data (1981-2005) recorded by well-trained observers at 615 meteorological stations across China revealed that visibility in the country was degrading rapidly due to increased fossil fuel usage (Che et al., 2007(Che et al., , 2009)).
Visibility impairment is generally accompanied by airborne particulate matter (PM), especially fine particles with aerodynamic diameters of 2.5 µm and less (PM 2.5 ), due to their light-scattering and absorption capabilities.Numerous studies around the world have focused on the relationship between visibility and aerosol composition (e.g., Dzubay et al., 1982;Sisler and Malm, 1994;Singh et al., 2008).In China, this relationship has also been widely investigated for individual cities.For examples, visibility degradation in Hong Kong has been found to be mostly due to fine sulfate particles (Lee and Sequeira, 2002;Cheung et al., 2005).Chemical speciation of aerosols in Guangzhou has indicated that fine particles, especially elemental carbon (EC), play a dominant role in visibility reduction (Deng et al., 2008).
However, most if not all of these studies were focused only on specific visibility degradation cases or seasonal variation.Although visibility is routinely monitored by the China Meteorological Administration (CMA), inter-annual variation is rarely examined due to the absence of long-term PM composition data.Routine observations of air quality and meteorological data are vital for interpreting the interannual changes of visibility and understanding its underlying causes.For instance, Tsai (2005) investigated the long-term trend of visibility in Taiwan (1961Taiwan ( -2003) ) by analyzing the relationship between meteorological parameters and air quality data, and found that the parameter of ln(PM 10 ) (Nature logarithmic concentration of PM 10 ) had the most significant impact on visibility.Zhang et al. (2010) found that the decreasing trend in visibility in Beijing from 1999 to 2007 was attributed to an increase in relative humidity.
Since the beginning of the 11th five-year plan (2006)(2007)(2008)(2009)(2010), aggressive steps taken by the Chinese government to tackle air pollution problems are expected to have a strong impact on regional air quality and atmospheric visibility.For example, the government encouraged the use of "cleaner" fuels such as liquefied petroleum gas (LPG) and low sulfur gasoline in industrial, transportation and domestic sectors The Flue-Gas Desulfurization (FGD) technique was widely adopted in power plants.These measures resulted in significant changes in atmospheric composition (e.g., Tang et al., 2008, Lu et al., 2010, Wan et al., 2011).Nevertheless, although decreasing trends of PM 10 concentration have been recorded by ground-based measurements over the five years, satellite observations continue to show an upward trend in aerosol optical depth (Lin et al., 2010).Such discrepancies highlighted the complexity of the air pollution problem in China.A crucial question is whether the control measures really do improve air quality and visibility.
Beijing, Shanghai, Guangzhou and Chengdu are four major megacities located in four economically and industrially developed regions of China with different climates (Fig. 1).
Beijing, as the capital of China, is the largest megacity, located at the center of the Beijing-Tianjing-Tangshan city cluster (BTT) in northern China.Shanghai and Guangzhou are the centers serving the Yangtze River Delta (YRD) in the east and the Pearl River Delta (PRD) in the south, respectively.The deltas are the most developed regions in eastern and southern China.Chengdu, located in the Sichuan Basin, is the most important megacity in southwestern China.Each of the four cities is also situated in a different haze region (Chang et al., 2009).The present study presents newly recorded data in these four locations over a recent five-year period (2005)(2006)(2007)(2008)(2009) to reveal the changes in visibility and air quality.An empirical regression model was developed to evaluate the effects of air quality and meteorological conditions on visibility, and to predict daily visibility based on air quality data and meteorological parameters.

Descriptions of Measurement Sites and Data
Urban meteorology stations provided data for Beijing (39°48'N, 116°28'E), Shanghai (31°24'N, 121°27'E) and Guangzhou (23°10'N, 113°20'E) were used for analysis.The station in Chengdu (30°42'N, 103°50'E) was situated in a suburban area.Its data was used because no continuous observation was carried out at an urban station during the period.Considering the well-mixed surface layer with a capping inversion in urban mountain basins (e.g., Pataki et al., 2005), we suggest that this suburban data adequately  The visibility data used for analysis in this study was recorded by trained observer at 0200, 0800, 1400 and 2000 (local standard time) every day throughout 2005 to 2009, in according with standard CMA procedures.Other meteorological data-temperature, relative humidity (RH), precipitation and wind speed (WS)-were also recorded at these times.A daily mean was calculated for analysis in this study.Air quality data, including daily concentrations of SO 2 , NO 2 and PM 10 , was measured at the air quality monitoring stations managed and operated by the Environmental Protection Bureau (EPB), by UV fluorescence method analyzers (Thermo Electron Corporation Model 43c in Beijing, Shanghai and Guangzhou, and an API 100E in Chengdu), dual-channel chemiluminescence analyzers (Thermo Electron Corporation Model 42c in Beijing, Shanghai and Guangzhou, and an API 200E in Chengdu), and R&P TEOM Series 1400a ambient particulate monitors.The detection limit was 1 µg/m 3 .The quality assurance (QA) and quality control (QC) procedures, including regular instrument calibration with Standard Reference Materials (SRM) or standards traceable to SRM, following Chinese government standard QA/QC requirements, were carried throughout the period at these stations.The data was stored as five-minute average values, from which daily average data was computed.
An overview of the general information of these four megacities is presented in Table 1.Care should be taken when comparing this data, especially visibility, with other previously reported observation results because of inherent systematic errors attributable to the limited number of stations in this study.The main focus was on the changes of visibility and air pollutant concentrations and their correlations in each individual city, rather than comparing the absolute values of visibility.

Time Series Analysis and Stepwise Regression Method
With an aim to better investigate the seasonal and interannual changes of visibility, a Kolmogorov-Zurbenko (KZ) filter was applied to the daily mean data set to separate the original time series into several components representing different time scale variation.A detailed description of KZ filter can be found in the literature (e.g., Rao and Zurbenko, 1994;Zurbenko et al., 1996;Eskridge et al., 1997).Briefly, the KZ filter is based on an iterative moving average that removes high-frequency variation from the data.In this study, a 15-day moving average procedure was repeated five times in order to remove all signals of less than 15√5 ≈ 33 days: that is, the high-frequency meso-and synoptic scales were removed in the KZ (15,5) time series and allows seasonal signals to be identified (Eskridge et al., 1997).Seasonal variations and the signals of smaller time scales were removed by KZ filtering for a window size of 365 days and 3 iterations.The resulting KZ (365,3) time series, containing signals with time scales of greater than 365√3 ≈ 632 days (1.7 year), and a fitted linear regression line were then used to reveal the inter-annual variation of visibility over the five-year period.
Visibility is affected by ambient air pollutant concentrations and meteorological conditions: in particular, degradation of visibility is directly proportional to the loading of airborne PM (light scattering and absorption); NO 2 also contributes due to its strong blue-light-absorbing capacity; and SO 2 , which itself does not affect visibility but can oxidize to form sulfates, some of the most important atmospheric species scattering light and degrading visibility.On the  2006-2009.d Exceeding rate is defined as the percentage of episode days with daily concentration exceeding Grade II of the Chinese National Ambient Air Quality Standard (SO 2 : 150 μg/m 3 ; NO 2 : 80 μg/m 3 ; PM 10 : 150 μg/m 3 ; GB3095-1996) within the study period.
other hand, meteorological conditions can also influence visibility by affecting air pollutant concentrations or their light-scattering capability.Specifically, aerosols grow in size when they interact with water vapor, causing increased scattering of sunlight in the atmosphere with high RH.High temperature and WS improve visibility by enhancing the dispersive capability of the atmosphere (via thermal and mechanical turbulence respectively) and reducing aerosol concentration levels.
In order to investigate the predominant factors affecting visibility and better predict visibility from air quality and meteorological data, an empirical model was developed incorporating the stepwise regression method.Generally, the stepwise procedure comprises an "enter" process and a "remove" process.The "enter" process allows each independent variable to be added into the regression equation only if it significantly increases the correlation of the equation.The independent variables are entered into the equation in descending order of their individual Pearson correlations with the dependent variable.In the present study, a significance level of 0.001 was adopted.After the new variable is entered into the regression equation, previously entered variables may lose some of their explanatory power.If the significance of a variable already in the regression equation falls below 0.005, it is removed from the equation.Daily visibility was set to be the dependent variable.Since visibility is generally associated with air quality (concentrations of SO 2 , NO 2 and PM 10 ), temperature, RH and WS, these parameters were selected as the independent variables (Tsai, 2005).Note particularly that days with RH > 90% or rainfall were excluded from the regression analysis (Tsai, 2005;Chang et al., 2009).

Characteristics of Air Pollution
Table 1 shows the "exceeding rates" of three major air pollutants (SO 2 , NO 2 and PM 10 ) in the four megacities during 2005-2009.Exceeding rate was defined as the percentage of episode days with daily concentration exceeded Grade II of the Chinese National Ambient Air Quality Standard (NAAQS; GB3095-1996) within the study period.In Beijing and Chengdu, PM 10 was the most dominant air pollutant, with exceeding rates (30% and 23.8% respectively) more than double those of the next most dominant air pollutant, NO 2 .On the other hand, Shanghai and Guangzhou were mainly affected by NO 2 (with exceeding rates of 20.4% and 17.4% respectively).The severe PM 10 pollution in Beijing and Chengdu is probably because these two cities are always affected by dust storms (e.g., Zhang et al., 2005Zhang et al., , 2010;;Tao, J., unpublished data).Low precipitation in these two cities may be another reason; this would prolong the lifetime of aerosols in the atmosphere (Table 1).In Chengdu the stable atmosphere reflected by low WS also favors the accumulation of particulate matter (Table 1).(Similar meteorological conditions and their effect on air pollution have also been found in Chongqing, another basin city near Chengdu (Yang et al., 2008).)The NO 2 pollution in Shanghai and Guangzhou highlights the potential role of photochemical pollution in these two cities.This result agrees well with previous satellite studies that showed that the YRD, PRD and BTT regions were the major NO 2polluted areas in China (van der A et al., 2006).

Recent Changes in Visibility and Air Quality
Fig. 2 shows the visibility time series of average daily KZ (15,5) and KZ (365,3) values, as well as a linear regression line fit to the KZ (365,3) data.Generally, visibility in Shanghai, Guangzhou and Chengdu presented more or less similar seasonal patterns, with maximums in summer and minimums in winter negatively related to those of air pollutant concentrations (not shown).Such seasonal cycles are widely observed in China and can be explained by the seasonal variation of meteorological conditions dominated by the East Asian monsoon (e.g., Chang et al., 2009;Wan et al., 2011).However, visibility in Beijing presented a different seasonal pattern, with a maximum in winter and a minimum in summer.Similar seasonal patterns can be found after excluding precipitation, and therefore we do not believe that a higher rainfall in summer (not shown) to be one of the important causes.In fact, the degradation of visibility in Beijing in summer has been observed in other studies, and was due to the transport of air pollutants contributed by industrial emissions, vehicular fossil fuel combustion and agricultural biomass burning from upwind region such as Hebei and Shandong Provinces (Li and Shao, 2009;Li et al., 2010).As illustrated by Wang et al. (2010), this unique seasonal pattern may also be partly owing to the high RH in summer (approximately 30% higher than that in winter; data not shown), while the RH in the other three cities were more or less consistent throughout the year.
Visibility in Beijing improved from 2005 to 2009 at a growth rate of 0.58 km/year (R 2 = 0.497).Although Chang et al. (2009) suggested that visibility in Beijing was decreasing with a rate of -0.78 km/decade between 1973 and 2007, the increasing visibility observed here is identical to the result of Che et al. (2009), who found that visibility in Beijing began to improve after 1995 due to the stricter environmental control measure implemented.Visibility in Shanghai has been improving since 1990 (Chang et al., 2009;Che et al., 2009) but, in recent years, the annual variation of visibility in Shanghai was relatively small and did not show a significant increasing trend (R 2 = 0.008) during 2005-2009.Although visibility in Guangzhou was dramatically degrading in the 1970s, it was maintained at a relatively steady level and seemed to begin improving since 2005 (Deng et al., 2008;Chang et al., 2009).In fact, with a notable growth rate of 0.83 km/year (R 2 = 0.825), the visibility in Guangzhou did improve from 2005 to 2009, which was the most pronounced increase in any of these four cities.Visibility in Chengdu has been degrading in the past three decades (Chang et al., 2009;Che et al., 2009).In contrast to the cases of Beijing, Shanghai and Guangzhou, visibility in Chengdu in the past five years did not stabilize or improve, but continued to decrease (-0.21 km/year, R 2 = 0.885).
Table 2 summarizes the annual averages of the meteorological parameters and major air pollutant concentrations in the four megacities.Most of meteorological parameters did not show significant trends except the decreases of RH in Guangzhou and Chengdu as well as WS in Shanghai and Chengdu.Such decreases are probably due to rapid urbanization, which would result in an urban heatisland effect (Brazel and Balling, 1986) and increased surface roughness (Vautard et al., 2010), although longer data records are still needed for further investigation.These results imply that the observed visibility changes are more related to the changes of air quality.
In fact, significantly decreasing trends were observed in all air pollutant concentrations (Table 2), suggesting that air quality was improving throughout the five-year period.2007.Such decreasing trends, especially for sharp decreases in 2007-2008, were probably due to the implementation of gas-desulfurization in power plants (Lu et al., 2010).Slight decreases in NO 2 concentration were also observed in all four megacities, in contrast to the increasing trend from 1997 to 2006 based on satellite measurements (Zhang et al., 2007).Marked decreasing trends of particulate pollution were attributed to the strong efforts of the Chinese government to control particulate emissions from various industrial sources (http://www.mep.gov.cn/).Recently, a similarly notable decreasing trend in PM 10 concentration was also observed in Foshan, an industrial city in the PRD (Wan et al., 2011).However, it is worthwhile pointing out that the concentrations of air pollutants, especially SO 2 and PM 10 , were still remarkably higher than those in most developed countries (Wan et al., 2011).These results emphasize the important role of the Chinese government's efforts in air quality improvement.Specifically, the hosting of international events, such as the 2008 Olympic Games in Beijing, the 2010 World EXPO in Shanghai and the 2010 Asian Games in Guangzhou, may have presented a precious opportunity for China to push and enforce the implementation of emission control measures to improve air quality.The global financial crisis and China's economic recession may also have exerted certain impacts on air quality improvement.However, the discrepancy between visibility and air quality changes in Shanghai and Chengdu highlight their complex relationship, which is discussed below.

Dependence of Visibility on Air Quality and Meteorological Conditions
As discussed, variation of visibility usually results from changes in both air quality and meteorological conditions.Fig. 3 presents the average visibility of the four megacities in various cases."Episode day" was defined as those days on which the concentration of air pollutant (SO 2 , NO 2 or PM 10 ) exceeded the recommended NAAQS II level.Other days were defined as "normal" days.For each of the meteorological parameters temperature, RH and WS, "high" days were defined as days on which the daily mean exceeded the 2005-2009 average, while the others were characterized as "low" days.
Fig. 3 shows that visibility on SO 2 , NO 2 and PM 10 episode days was noticeably reduced compared to normal days, emphasizing the effects of air pollution on visibility degradation mentioned above.The average visibility in various meteorological conditions is also presented in Fig. 3.The lower visibility in high RH cases highlights the impact of RH on the light-extinction capacity of aerosols.As discussed, high WS and temperature enhance the dispersion of air pollutant and thus improve visibility, which is also supported by the present results.Note the exception of Beijing, which experiences lower visibility on high-temperature days.This is probably due to the fact that visibility in Beijing has always been degraded in summer, as illustrated previously, since it is related to air pollutant transport and high humidity (Li & Shao, 2009;Li et al., 2010;Wang et al., 2010).
To better understand the dependence of visibility on air quality and meteorological conditions, it was necessary to calculate the correlation between them.In fact, the Pearson correlations between visibility and temperature, RH, WS, SO 2 , NO 2 and PM 10 concentrations in the four megacities were all significant at the p = 0.001 level.Therefore, a stepwise multiple regression method was applied to further examine their underlying relationships, and which would consider their mutual impact.A model using this approach was developed to simulate daily visibility based on air quality and meteorological parameter data.To enable the validity of the model to be evaluated, only data from 2005 to 2008 was used in its development (Table 3); the model was then interrogated by the newly recorded 2009 data (Fig. 4).Overall, with the exception of Beijing, the model tended to overestimate or underestimate the visibility in many cases (Fig. 4), suggesting that it was too ambitious to expect a single regression model to be able to predict the visibility Fig. 3. Visibility means in (a) air pollutants (SO 2 , NO 2 or PM 10 ) episode or normal days and (b) days with various meteorological conditions.("Episode" or "Normal" stands for the days with air pollutant concentration exceeding or below the daily NAQQS II respectively; "High" or "Low" stands for the days with daily mean of meteorological parameter higher or lower than the average of 2005-2009 respectively; "Temp", "RH" and "WS" stand for temperature, relative humidity and wind speed respectively.) in such a wide range of situations.Accordingly, an optimal empirical regression model was proposed and developed to produce a formula set containing two stepwise regression models, as described below.

Optimal Empirical Regression Models for Visibility
The optimal empirical regression models for each megacity are shown in Table 3.The original data for each megacity was grouped into two categories and the detailed grouping procedure appears in the Appendix.That is, each optimal model is a formula set containing two stepwise regression models as foreshadowed above.Overall, as revealed by the higher determination coefficients R 2 (where R 2 × 100% of data can be explained by the model), the optimal regression models provide better prediction capacity than the original models did (Table 3).
To further verify the validity of the optimal model, daily observed visibility data for 2009 were examined by linear regression, which regards simulated visibility as an independent variable and observed visibility as the dependent variable.The results confirm that the newly developed optimal empirical regression model simulated the observed visibility data more accurately than the original stepwise regression model did, giving a higher R 2 value, a slope approaching 1 and an intercept approaching 0 (Fig. 4).As examples, in some cases for Beijing (e.g., 2009-2-9 and2009-12-24) which the original model had underestimated, the optimal empirical model now gave improved results.For Shanghai, in some cases (e.g., 2009-6-18), the optimal empirical model performed better than the original model had done, although it was still far from exactly accurate.For Guangzhou, the optimal empirical model simulated visibility more accurately in summer.In Chengdu, although R 2 in the correlation between simulated and observed visibility was slightly less than that in the original model, both the slope and intercept were remarkably smaller.The original model had tended to overestimate the visibility in several cases (e.g., 2009-1-18), but the newly developed model did not.

Possible Causes of Visibility Changes
As well as the ability to better simulate daily visibility, the new optimal empirical stepwise regression model developed in this study also provided clues to answering the question of the extent to which air quality and meteorological parameters affect visibility.
In Beijing, visibility was affected by concentrations of PM 10 and NO 2 as well as RH, when RH was above the fiveyear average.In other cases, visibility was influenced only by PM 10 concentration and RH.Such results indicated that PM 10 concentration was the common factor in visibility degradation in Beijing.Considering that the increasing trend of improved visibility (0.58 km per year) coincided with a decreasing trend in PM 10 concentration (13.8 μg/m 3 per year), it is reasonable to conclude that the reduction of PM 10 has been the main reason for visibility improvement.In Shanghai, visibility was related to PM 10 concentration, RH and temperature.Although a deceasing trend was observed in PM 10 concentration, visibility did not show any significant improvement.In a recent study, Lin et al. (2010) suggested that PM 10 emission control policy had not been successful in reducing concentrations of fine particles, which are more effective in scattering light and degrading visibility, due to the formation of secondary fine particles from precursor species.Therefore, the steady level of visibility was probably due to the similarly steady level of PM 2.5 concentration, although the concentration of coarse particles decreased significantly; unfortunately, long-term observation data of PM 2.5 concentration was not available for the present study, and this hypothesis could not therefore be tested.Our results imply that long-term observation of PM 2.5 concentration and composition is of importance for improved studies of the causes of long-term visibility change.Note particularly that if the entrance and removal significant levels in stepwise regression method are set to be 0.05 and 0.10 respectively, WS is one of the significant independent variables included in the model of "Low WS" cases (with a coefficient of 0.639 km per m/s, not shown), highlighting the potential impact of WS on visibility: in other words, the decreasing trend of WS could be a possible factor in maintaining a steady visibility level.Nevertheless, the exact cause of visibility change in Shanghai was difficult to explain currently and needs to be further investigated.
In Guangzhou, visibility was affected by RH and concentrations of PM 10 and SO 2 when temperatures were below the five-year average.Otherwise, temperature exerted a positive impact on visibility, while no significant correlation was found between visibility and SO 2 concentration.This result suggested that high temperatures were of importance to the dispersive capability and of atmospheric visibility (usually in summer, average 29.0°C), but had negligible effect in other cases when the SO 2 concentration exerted a negative impact on visibility and the influence of PM 10 concentration was smaller.As mentioned before, SO 2 impairs visibility through the formation of secondary fine sulfate particles which scatter light efficiently, and therefore the observed significant impact of SO 2 on visibility degradation at low temperatures implied that the oxidation of SO 2 and formation of sulfates were probably enhanced in winter (average temperature 15.6°C).It is worthwhile noting that, from the PM 2.5 chemical speciation data observed in the period mid-2009 to early 2010, elevated concentrations of sulfate and high ratios of sulfate: (SO 2 + sulfate) were detected in autumn and winter (Tao, J., unpublished data), which supported this hypothesis and highlighted the crucial role of secondary sulfate formation on visibility degradation at low temperatures.This result also agrees with the fact that SO 2 dissolves more readily in water and oxidizes into secondary sulfates via heterogeneous reactions at low temperatures (Ota and Richmond, 2011).A similar relationship between SO 2 and visibility impairment was also found in Hong Kong, another megacity in the PRD (Lin et al., 2012).Overall, the observed improvement in visibility may reasonably be attributed to prominent decreases of PM 10 concentration.Significant decreases of SO 2 concentration in winter (from 64 μg/m 3 in 2005 to 37 μg/m 3 in 2009; data not shown) also exerted a positive impact on the improvement of visibility.
In Chengdu, visibility was affected by PM 10 concentration, RH, temperature and WS.In particular, the negative effect of RH and positive influence of WS on visibility were much more vital when RH and PM 10 concentrations were high.Although PM 10 concentration and RH both decreased in the past five years (from 130 μg/m 3 and 77.2% in 2005 to 112 μg/m 3 and 73.1% in 2009, respectively), visibility in Chengdu did not improve but continued to deteriorate.The observed   The decreasing trend of visibility in Chengdu was thus probably caused increased PM 2.5 concentration as wellwhich, however, could not be investigated in this study.Note particularly that, as mentioned above, the air quality monitoring station in Chengdu was located about 10 km from the meteorology observation station (located in the suburban area), which in itself might suggest another reason for the observed discrepancy.

CONCLUSION
Visibility degradation has been widely observed in China.In this study, the recent changes of visibility in four major megacities were investigated: Beijing, Shanghai, Guangzhou and Chengdu.Optimal empirical regression models were developed to investigate the underlying causal relationships between visibility, air quality and meteorological conditions.The models also provide a simple tool for predicting daily visibility with regard to air pollutants (SO 2 , NO 2 and PM 10 ) and meteorological data (temperature, RH and WS) in the future.
Because of the stricter air pollutant emission control regulations imposed by the government, air quality in terms of three major air pollutant concentrations was observed to have improved in these four megacities during the five years 2005-2009, although pollution levels were still higher than in other megacities in developed countries.Similar significant improvements in atmospheric visibility were observed in Beijing (0.58 km/year) and Guangzhou (0.83 km/year).However, visibility was slightly degraded in Chengdu (-0.21 km/year).In Shanghai, visibility decreased slightly before 2007, then increased somewhat after 2007.Results of regression analysis have suggested that the improvements of visibility in Beijing and Guangzhou were mainly due to decreases in PM 10 concentrations.Moreover, reduction of SO 2 concentration in Guangzhou also exerted a positive effect on visibility improvement there.On the other hand, the exact reasons for the lack of visibility improvement in Shanghai and Chengdu are not yet clear.The decreasing trend of wind speed, and therefore a weaker atmospheric dispersive capability, is probably one reason.Changes of PM 2.5 concentrations and compositions might also affect the visibility change in all of these cities, especially in Shanghai and Chengdu.These results highlight that policy measures which focus upon PM 10 reduction as the air quality metric may not achieve corresponding improvements in visibility.A longer time data set will help to more clearly identify the trends.

Fig. 4 .
Fig. 4. Time series of daily observed visibility (black) and daily visibility simulated by original stepwise regression model (green) and optimal empirical regression model (red) in 2009 in Beijing, Shanghai, Guangzhou and Chengdu.
decreases in visibility were probably due to the reduction of WS (0.75 m/s in 2005 to 0.53 m/s in 2009), although the coefficient was just 2.01 km per m/s, too low to trigger the notably decreasing trend in visibility.This indicates that visibility degradation in Chengdu was still affected by other unknown factors.As mentioned previously, visibility is much more related to fine particle concentration than coarse particle concentration due to their different scattering effects.
Table 1 provides a detailed description of the stations.

Table 1 .
General information, meteorological conditions and air pollutant concentrations in Beijing, Shanghai, Guangzhou and Chengdu in the period2005-2009.
b Average ± standard deviation.c Data in the period of

Table 2 .
Annual means and inter-annual trends of air pollutant concentration and meteorological parameters in four megacities.

Table 3a .
Summary of original stepwise regression models.a