Data (n = 161 cities) from almost half of the total cities (n = 342) in China were analyzed ecologically and cross-sectionally using standard and robust analysis approaches.

In the first group of analyses, univariate analyses were used to describe the pattern of daily, weekly, seasonally and yearly AQI concentrations by city and regions. Seasons were classified as spring (March to May), summer (June to August), fall (September to November), and winter (December to February). Six regions are grouped: North China (Beijing, Tianjin, Hebei, Shanxi, and Inner Mongolia), Northeast China (Liaoning, Jilin, and Heilongjiang), East China (Shanghai, Jiangsu, Zhejiang, Anhui, Fujian, Jiangxi and Shandong), South Central China (Henan, Hubei, Hunan, Guangdong, Guangxi, Hainan, Hong Kong, and Macau), Southwest China (Chongqing, Sichuan, Guizhou, Yunnan, and Tibet), and Northwest China (Shannxi, Gansu, Qinghai, Ningxia, and Xinjiang) [20].

In the second group of analyses, spatial patterns of AQI and air pollutants were described using Arc Geographic Information System (Version 12, Redlands, CA) [21] and the correlation of AQI with heat index and spatial variations were examined using linear correlation and spatial autocorrelation analysis techniques. In Spatial autocorrelation, Moran’s I statistics were calculated using SAS 9.3 Proc variogram [22]. Moran’s I is a measure of spatial autocorrelation and it indicates how AQI values were measured on the basis of their location (point) that is identified by latitude and longitude of the surveillance station where air pollutants were measured. Let Z (s) represent the value of Z at point s. For locations s and s + h, the variance of the difference of the Z(s) and Z(s + h) is of interest. Moran’s I is calculated based on cross-products of the deviations from the mean：

I ( h ) = n N ( h ) ∑ i ∑ j w i j ( Z ( s i ) − Z ( s ) ¯ ) ( Z ( s j ) − Z ( s ) ¯ ) ∑ i ( Z ( s i ) − Z ( s ) ¯ ) 2

where w_ij are the elements of the weight matrix, and N(h) is the sum of the elements of the weight matrix.

Meanwhile, Geary’s C statistic was calculated, an alternative method to measure spatial autocorrelation, estimated using the following formula [22].

C ( h ) = n − 1 2 N ( h ) ∑ i ∑ j w i j ( Z ( s i ) − Z ( s j ) ) 2 ∑ i ( Z ( s i ) − Z ( s ) ¯ ) 2

In the third group of analyses, correlations between AQI and climate factors were examined using linear correlation and regression analysis. Because of the nature of the spatial-based cross-sectional and time-series data, we further applied Panel (data) analysis methods [23] [24]. We performed an independently pooled model, and a fixed effect model, because time-specific effects on AQI are not randomly distributed in the data.

In the fourth group of analyses, the association between AQI and total mortality were examined using simple and partial (adjusted) correlation analysis methods. In the partial correlation analysis, we adjusted for city-level socioeconomic status (assessed by gross domestic product per capital).

All statistical analyses were conducted using SAS software, version 9.3 (SAS Institute, Cary, North Carolina) [22]. The level of significance was set at a 2-sided test at p ≤ 0.05.

Table 1 shows that of 161 cities the mean (SD) and medial AQI were 95.69 (37.21), and 89.28 (quartile range: 70.62 - 112.31), and of them mean temperature in 50 cities (climate data available and matched for 50 cities) was 14.34 (11.00)˚C in 2014. The means (SD) of heat index, precipitation, sunshine hours and pressure are shown in Table 1 as well.

Figure 1 depicts the six geographic regions in mainland China (on the left side), and the monthly average AQI concentrations by regions from January to December in 2014 (on the right side). All six regions had higher AQI in the winter term, followed by earlier spring (March and April), and alter fall (October and November), and the lowest AQI in the end of summer and the beginning of fall (August and September). Of the four seasons, the highest average concentrations

a: IQR: Interquartile range; b: Of 161 cities with AQI measures, 50 cities had daily climate change measures.

(±SD) of AQI were in winter (119.58 ± 2.19), followed by spring (95.02 ± 1.35), fall (89.30 ± 1.37), and summer (78.86 ± 1.05) in 2014 (Table 2). The ANOVA test shows that compared with summer as the baseline, there were significant difference between spring and summer (p < 0.0001), fall and summer (p<0.0001), winter and summer (p < 0.0001). Table 2 also shows that among 6 regions, the highest average concentrations (±SD) of AQI were in north (117.36 ± 2.73), followed by northwest (102.39 ± 2.79), east (99.23 ± 1.26), northeast (90.06 ± 1.89), south central (85.18 ± 1.66), southwest (81.54 ± 2.72).

Figure 2 depicts a significant variation in five major air pollutants (PM2.5, PM10, CO, SO2, 1-Hour Ozone, and 8-Hour Ozone) across the country. Figure 3 depicts the annual AQI concentrations across the nation (on the right-side). Spatial autocorrelation analysis technique was further applied to test whether cities with higher AQI were clustered in certain areas. The result indicates that

*AN0VA: Analysis of Variance. Rank test used Mann-Whitney U test.

Moran’s I coefficient (0.22) was greater than 0 (p < 0.0001), and Geary’s C coefficient (0.80) was less than 1 (p < 0.0001). Both statistics indicate that the values of AQI measures across cities had positive spatial autocorrelation, meaning the spatial distribution of high AQI values were more spatially clustered rather than dispersed.

In 2014, Of 161 cities, more than one third (37.27% - 39.75%) had AQI 100-149, and 67 cities (41.61%) had AQI 150 and higher in January (Table 3).

Of the 161 cities with AQI measures, 50 cities had available data for meteorological measures. Among these cities, simple correlation analysis indicated that monthly average AQI concentrations were significantly and negatively correlated with temperature, heat index, precipitation, and sunshine hours, but significantly and positively associated with pressure (Table 4).

a: Air quality level: Excellent: AQI < 50; Good: AQI ≥ 50 and AQO < 100; Light polluted: AQI ≥ 100 and AQI < 150; Moderately polluted: AQI ≥ 150.

Similar correlation relationships between QAI and meteorological factors were observed using panel pooled analysis. It indicates that 28.8% (R2) of the variations in AQI could be explained by the differences in meteorological factors (Table 5). As AQI varies by time, using panel fixed time (month) effect models, the results suggest that 39.2% of the variation in AQI could be explained by the differences in meteorological factors and time (months) in China 2014.

a: Beta: Regression coefficient. B: R-Square: Coefficient of determination.

Of 161 cities with AQI data, we are able to collect reliable total mortality data from 110 cities in 2014. Annual average AQI concentrations were positively and significantly correlated with total mortality (r = 0.2, R² = 4.0%, p = 0.034). After adjustment for socioeconomic status (assessed by gross domestic product), the association between AQI and total mortality in 2014 remained statistically significant (r = 0.21, R²=4.4%, p = 0.036).

China is facing a great challenge of controlling air pollution, especially in urban cities and certain regions because of their great differences in geo-physical and socioeconomic environmental status. Our study shows that about more than one third (32% - 39%) of the study cities had AQI ≥ 100 (classified as slightly polluted) in seven months of 2014 (Jan-March, May and Oct.-Dec.), and significantly higher proportions of cities had AQI ≥ 150 (classified as moderately polluted) in Jan (41.61%), Feb (16.15%), Nov (7.45%), and Dec (6.83), respectively. The variations in AQI were significantly and spatially clustered across the country. Overall, North and Northwest China had higher AQI than the rest of the country. Although several individual studies examined the differences in air quality across the cities, the present analysis is one of the first to use a larger sample size (161 cities) to test the geographic differences in AQI across the county. In Huang and his colleagues’ study, they examined air pollution (assessed by PM2.5) from 13 air-monitoring stations in the city in 2013-2014 in Xi’an City, Shannxi province of China, and observed a significant clustering distribution of neighborhoods with elevated PM2.5 concentration across the 13 air-monitoring stations [25]. Kai et al. evaluated air quality in Guangzhou (located in south central China), compared with additional 11 large cities of China. They used air quality data monitored from 2000 to 2005. Their findings suggested that the average air quality in Guangzhou over 6 years was higher than that in Beijing, Tianjin, Nanjing, Hangzhou and Shanghai, and lower than that in Shenzhen, Zhuhai and Shantou [26]. Findings of our present study added new evidence by using a much larger sample size and applying more vigorous analysis to the body of the research literature and highlighted a serious variation in the burden of air pollution across the country.

Climate change can be caused in part by increased atmospheric concentrations of carbon dioxide and other green-house gases. It is likely to result in changes in temperature, humidity, amount, distribution, and intensity of precipitation events and the intensity and frequency of certain extreme weather events [27]. In our study, significant and negative associations of AQI with heat index, precipitation, and positive association with atmospheric pressure were observed. A higher concentration of AQI in winter and lower concentration in summer has also been reported by other studies in China. For example, the concentration of PM2.5 was higher in winter and spring, but lower in summer reported in previous studies in Xi’an and Guangzhou cities of China [25] [26]. We confirmed this seasonal difference by using data across the country. We further examined the correlation between AQI and four major climate factors using panel fixed effect model (fixed for yearly-month effect), and panel pooled model (without fixed yearly-month effect). It suggests that on average, heat index was negatively correlated with AQI in China. A significant variation in air quality by seasons may be attributable to the natural climate changes, (such as cold months but less wind in winter, and lots of rainfall in summer), and use of different types for heating. In winter, coal is the main source of heating in most cities of China, which is considered the main factor for a poor air quality in cold months. A dry and dusty spring may account for a relatively higher concentration of air pollutants in spring. While in summer, there is lots of rainy and windy days, which can increase the sedimentation ability of the air and help the diffusion of air pollutants. In fall, a relatively low temperatures and winds are very common, both are not beneficial for the diffusion of air pollutants [24] [25]. Furthermore, cities in North China have experienced a longer history of industrialization as compared to South China. It consumes much more energy from coal or other sources than South China, which may be partly attributable to a higher concentration of air pollutants in the North versus the South. Although we are unable to examine details of the complex associations of air pollution with climate and energy use because of the lack of the data availability, findings from our study add to new evidence of the overall association between air quality and climate change and calls for more detail studies.

The association between air pollution and public health has been examined in several Chinese cities. These studies have found a positive association of air pollution with years of potential life lost, mortality, cardiovascular disease and respiratory diseases mortalities [4] [6] [7] [17] [24] [28] [29] [30]. The importance of controlled air pollution in urban cities was highlighted, although the data used for these studies were either local or from only a few major cities. For example, one of the previous studies found the significant association between ambient air population and adult respiratory mortality. Their data was from 32 Chinese cities [24]. Our study has an advantage that utilized data from multiple sources released by the Chinese national air quality and health surveillance systems and was able to test for and confirm a significant and positive association between AQI and all-cause mortality. This finding further addresses the serious impact of air quality on urban public health.

There are several limitations that should be kept in mind when interpreting the major findings of this study. First, the associations between AQI and climate factors were analyzed cross-sectionally. It is not necessary to interpret as a cause-effect association, although changes in regional distribution of temperature show that a warming trend was more significant in West, East and North China than in South China [6]. Second, data for the analysis of the association between AQI and meteorological factors were only available in 50 cities of the 161, which reduced the statistical power when we tested the association between AQI and mortality with adjustment for climate factors (data not shown). Third, more detail health related data at individual city levels were not available. Therefore, further studies with multiple-sits and multiple disciplinary collaboration are needed. Nevertheless, the present study also has several advantages. First, as compared to all previous publications, the present study had the largest sample size on air quality which was measured in 161 major cities, almost half of the total cities of China. Second, the accuracy and quality of data on AQI, meteorological and health outcomes are reliable as they were consistent and reported by Chinese authority agencies, which offer a comparable base for further studies. Third, using the spatial autocorrelation technique, the present study is able to quantitatively test and confirm a spatial clustering of AQI across the regions of China. The panel fixed analysis is one of the first to address the collection between AQI and meteorological factors by fixed-months effect (differences in temperature across the regions by months). The study provides further evidence to public health policy makers on air quality control and health outcomes promotion.

In conclusion, findings of this study indicate that the burden of air pollution remains a serious public health issue. AQI significantly varies across the country, and is significantly associated with climate factors, and is positively and significantly associated with total mortality.