2.1 Data

Haze pollution occurs frequently in autumn and winter over eastern China, and therefore, the winter half years (September to February) were chosen as the study period for the years 1961–2013. The daily visibility observations from 876 sites of the China Meteorological Administration (CMA) were used to characterize the air pollution status over eastern China (Figure 1), which are available in http://data.cma.cn/site/index.html. Based on the visibility observation rules of the CMA, the visibility data were recorded in the form of scale before 1980; those after 1980 were recorded in the form of kilometer. Their corresponding relationships are presented in Table 1. Because of the inconsistent methods of recording data for visibility before and after 1980, it was necessary to undertake procedures to achieve consistency across the data. Following the method of Ding and Liu (2014), all the visibility data from 1980 to 2013 were converted into the visibility-scale form. A hazy day is a day on which the average visibility (visibility level) is less than 10 km (6) and the relative humidity is less than 90%; the present weather code excludes natural events such as precipitation, dust, fog, mist, and gale conditions (Che et al., 2009; Schichtel et al. 2001; Wu et al., 2010). To check the robustness of the typical polluted and clean types of synoptic circulations (Zheng et al., 2015), the main data set used to describe air quality was the daily averaged visibility levels (excluding natural events) at each station during synoptic circulation events.

The corresponding surface and low-level synoptic circulation analyses performed in this study were based on the results of meteorological reanalysis products available from the National Centers for Environmental Prediction and the National Center for Atmospheric Research, which are 2.5° × 2.5° latitude-longitude grids produced on a daily basis (http://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis.html). The sea level pressure (SLP) field, which is closely related to the meteorological factors at surface, was selected to characterize a certain synoptic episode, and geopotential heights at the 850-hPa (HGT₈₅₀) level were selected to characterize atmospheric circulation in the upper PBL or near the bottom of the troposphere. In addition, monthly SLP, surface temperature, and wind field at the 850 hPa were also used to explore the patterns of circulation anomalies, when Arctic sea ice area obviously declined in autumn. The Arctic sea ice data are available at the website https://www.metoffice.gov.uk/hadobs/hadisst/data/download.html.

Annual total emissions data for eastern China were provided by the National Bureau of Statistics of China (http://data.stats.gov.cn/staticreq.htm) and were used to identify the emission trend of recent years. The emission data included power emissions, industry emissions, residential emissions, and transportation emissions.

2.2 Method

2.2.1 Persistent Synoptic Circulation

This study used the typical persistent synoptic circulations (PSCs) related to air pollution identified by Zheng et al. (2015) as examples. Figure 2 represents six polluted and three clean types of SLP patterns with HGT₈₅₀ circulations in the PBL. The polluted-type circulations in the winter half year over eastern China are associated with two possible systems: (1) they are dominated by a uniform surface pressure field (related to T2, T3, and T4), and (2) they occur before a cold front passes (related to T1) or are associated with anticyclonic circulation at a 850-hPa level (related to T2, T3, T4, and T5). The clean-type circulations, however, are mainly associated with the passing of a cold front that has strong, cold winds (related to T7 and T9) or the rear of a weak high-pressure system carrying clean air from the sea (related to T8). Occurrence frequencies of PSC events were calculated by dividing the total duration days for all PSC events by the total number of days in each winter half year. Durations of PSC events were defined as the number of consecutive days (2, 3, 4, and >5 days) for each of the PSC events in each winter half year during 1961–2013. The percentage of PSC events by duration of 2, 3, 4, and greater than or equal to 5 days was defined as the ratio of events of each duration to total PSC events for polluted types and clean types in each winter half year during 1961–2013.

Vertical velocity, averaged from 1,000 to 100 hPa, and divergence of the wind field in the lower troposphere were used to quantify the diffusion conditions in each circulation type. A pollution episode commonly occurs when both the mean downward motion (less than 2.56 × 10⁻² Pa/s) and the divergence of low-level winds (less than 1.79 × 10⁻² s⁻¹) are strong. Otherwise, a clean episode is likely.

2.2.2 Similarity Method

A similarity method is usually applied to describe the difference between data samples in weather and climate research (He & Shao, 2014; Li, 1986). A PSC is typically characterized by a certain synoptic weather process with a stable or highly similar pattern on two or more consecutive days. To screen for PSCs from long-term data, criteria were required to determine the similarity of two circulation fields. The morphological and numerical similarity of a certain meteorological variable (e.g., pressure or geopotential height) should be considered simultaneously when analyzing the similarities between two atmospheric circulation fields. Here a correlation coefficient and a Hamming distance were used to analyze the similarities among PSCs.

A correlation coefficient characterizes the morphological similarity for a meteorological variable at two different times, and Hamming distance is used to describe the numerical difference between two samples (Li, 1986; He & Shao, 2014; Hou et al., 2012). The correlation coefficient and Hamming distance were respectively calculated in equations 1 and 2 as follows:

(1)

(2)

where R_ij is the correlation coefficient for a meteorological variable x (e.g., pressure or geopotential height) between two time periods, i and j. k is the sequence number (k = 1, 2, …, n) of the grid point contained in the study area, and x_ik indicates the value of the kth grid at time i. / is the mean value of x correlation coefficient R_ij that is mainly related to the dispersion of x at each grid to the mean value at the corresponding time in the study area, which is a valuable criterion for morphological similarity identification. The value range is −1 ≤ R_ij ≤ 1. R_ij > 0 indicates a positive correlation between two samples at different times, and R_ij < 0 indicates a negative correlation. The larger the absolute value of R_ij is, the more favorable the pattern of similarity is. When the patterns of two samples match exactly, the value of R_ij is 1.

H_ij is Hamming distance. The value range of H_ij is related to specific variables, and thus, it was not convenient for us to compare the numerical similarity of different variables by using their Hamming distances. A specific variable was therefore normalized as equation 3:

(3)

where x_ik and indicate the original and normalized values of the kth point (k = 1, 2, …, n) at time i, respectively. x_kmin and x_kmax are the minimum and maximum values appearing in historical samples of the kth point, respectively. The value range is . The average Hamming distance was calculated from the normalized data according to equation 4:

(4)

Similar to H_ij, is also spatially averaged, and its values ranged from 0 to 1 when maintaining its meaning. The smaller the is, the higher the degree of numerical similarity is. When the values of two samples are exactly equal, the value of is 0.

Following the method of Wang and He (2012) and Wang et al. (2015), the East Asian winter monsoon index (EAWMI) was defined by the mean geopotential height at 500 hPa in the area of (25°–45°N, 110°–145°E) to describe the East Asian trough that is closely associated with the EAWMI and cold surge activity. The larger the EAWMI number is, the weaker the East Asian winter monsoon is. The autumn Arctic sea ice index (ASI) was calculated using the accumulated autumn sea ice extent in the region north of 45°N. The larger the ASI number is, the greater the extent of the autumn Arctic sea ice is. The cold activity index (CAI) was defined as the total counts of every winter half year, which are from the National Climate Center of China (http://cmdp.ncc-cma.net/Monitoring/cn_index_130.php). The larger the CAI number is, the higher the cold activity is.

To match the dimension of each index, EAWMI, ASI, and CAI were normalized as follows:

(5)

where i refers to a year during 1961–2013, x_i is the EAWMI/ASI/CAI at year i, x_max and x_min are the maximum and minimum EAWMI/ASI/CAI during 1961–2013, respectively, and f(x_i) is the normalized EAWMI/ASI/CAI for each year, which is within a range from 0 to 1.

3.1 Statistical Analysis of Thresholds in the Criterion of Similarity of Circulation

Figure 3 shows the probability density function (PDF) distributions of correlation coefficients and Hamming distances of SLP and HGT₈₅₀ over eastern China and surrounding areas (105–127°E, 22–45°N) between two consecutive days in the winter half year during 1961–2013. For SLP, the shapes of PDFs for correlation coefficients were similar between autumn and winter, with values ranging from −0.2 to 1 (Figure 3a); values over 0.6 occurred with high frequency. Negative values were rare, which implied that the occurrence of opposite morphological patterns of SLP between two consecutive days was rare. The Hamming distances of SLP mainly ranged from 0 to 0.15. The shapes of PDFs for Hamming distances were narrower in autumn than in winter (Figure 3b). Hamming distances in autumn were smaller than those in winter, which indicated that the values of SLP were more stable in autumn. For HGT₈₅₀, both the shapes of PDFs for correlation coefficients and Hamming distances were similar between autumn and winter. The correlation coefficients and Hamming distances were, respectively, 0.2–1 (Figure 3c) and 0–0.15 (Figure 3d).

To determine similar circulation patterns, criterion thresholds of the similarity indexes were defined as the 50th percentile (i.e., the median quartile) of SLP and HGT₈₅₀. For instance, we calculated four similarity indexes and present their time series during 1 September 1993 to 28 February 1994 in Figures 4a–4d. When the correlation coefficients of SLP (SLP-R) and HGT₈₅₀ (HGT₈₅₀-R) reached or exceeded their corresponding median quartiles, the days were determined to have had the patterns that exhibited a high morphological similarity between the two consecutive days. It was the same determination for numerical similarity based on Hamming distances of SLP(SLP-H) and HGT₈₅₀(HGT₈₅₀-H).

Furthermore, circulation in the PBL was determined to have numerical and morphological similarity between two consecutive days when at least three of four similarity indexes met their criterion thresholds at the same time. Figure 4 presents the days having similarity of circulation in the PBL from 1 September 1993 to 28 February 1994. To validate the rationality of the selected threshold, we randomly chose the SLP-HGT₈₅₀ series of 1–10 September 1993 for a case study. Figure 5 displays the distributions of SLP and HGT₈₅₀ during that period, and the corresponding correlation coefficients and Hamming distances between two consecutive days are presented in Table 2. After comparing the similarity indexes and their criterion thresholds, similar circulations in the PBL were determined to have mainly appeared on 1–2 and 6–8 September 1993, which matched well with the actual situations presented in Figure 5.

For analyzing circulation in the PBL, it is generally appropriate to use four similarity indexes to discriminate and ensure the morphological and numerical similarities of the SLP and HGT₈₅₀ situations in the PBL. After preliminary calculations, 1,754 PSC events occurred during the winter half years of 1961–2013, and the total number of days of PSC events in the PBL was 5,993, accounting for roughly 52% of the total days of the entire set of winter half-year periods.

3.2 Long-Term Trends in Typical Polluted and Clean Types of PSCs in the PBL

In addition to emissions, a stagnant synoptic circulation, especially under a high atmospheric stability condition, contributes to air pollution episodes in eastern China (Y. Zhang et al., 2016; Zheng et al., 2015). It is critical to understand how the variation features of PSCs relate to air pollution over eastern China. Figure 6 presents the mean spatial patterns of visibility levels for these nine PSC types, which are presented in Figure 2. For polluted-type circulations (T1–T6), lower visibility levels (<7.0) were present in most areas (Figure 6a), and higher visibility levels (>7.0) were present in most areas (Figure 6b) for clean-type circulations (T8–T9). The spatial patterns of lower (higher) visibility levels are well matched with those of higher (lower) aerosol optical depth from derived from Terra's Moderate Resolution Imaging Spectroradiometer measurements (Zheng et al., 2015, see Figure S1 in supporting information). These results confirmed that the empirical classifications of weather types were robustly correlated to air quality with various circulations patterns.

Using the criterion of similarity method for PSC (section 3.1) to determine the similarity of 1,754 PSC events with the nine PSC types, a total of 236 polluted-type and 167 clean-type PSC events, occurring on days 868 and 625, respectively, were identified in the PBL from 1961 to 2013. To validate these results, we analyzed the averaged visibility-level patterns of the polluted and clean events and their difference patterns during 1961–2013 (Figure 7). The visibility level at most stations was less than 6.9 in the polluted-type events (Figure 7a) and dropped under 6.5 in areas with especially high air pollution—Hebei, Henan, and the Yangtze River Delta—which exhibited significantly decreased visibility levels (Figures 7d and 7e). The visibility level in clean-type events was greater than 8.0 (Figure 7b), except those in aforementioned high air pollution areas (Figure 8d). The averaged visibility levels in the polluted-type events were 0.1–0.8 smaller than in the clean-type events over almost the entirety of eastern China (Figure 7c). These results are consistent with the conclusions of Zheng et al. (2015). The comparison proved the reliability and robustness of the classifications of PSC events for polluted and clean types.

For the classification of PSC events, the long-term variations and trends in the polluted and clean PSC events were addressed through analysis of total duration days and occurrence frequency of persistent polluted and clean events, and the frequencies of different visibility levels only for PSC period (Figure 8), which were key features for exploring the variations in PSC with their effects on air quality.

First, in past decades, the total duration days of PSC events for polluted types generally tended to become longer, especially for the periods of after 2005 (Figure 8a). The variations in occurrence frequencies of PSC events for polluted types in the PBL exhibited varying trends in the past decades over eastern China, an increase in 1961–1979, no obvious change in 1980–1999, and a rapid increase since 2000 (Figure 8b), which are well matched with the variation in trends of haze days (Wang & Chen, 2016). Correspondingly, the trends in the frequencies of lower visibility levels 5 (visibility: ~2–4 km) and 6 (visibility: ~4–10 km) presented the similar variation patterns in these three different decades and also increased significantly (Figure 8c), indicating a strong relationship with occurrence frequencies of polluted-PSC events (0.650 and 0.654, respectively; p < 0.001), while the trends in the frequency of higher visibility level 7 (visibility: ~10–20 km) decreased significantly (Figure 8c), exhibiting a significantly positive correlation (0.444; p < 0.001) with occurrence frequency of clean-PSC events. The long-term variation in mean visibility levels only for the polluted-type events was highly related to that in occurrence frequencies of polluted-PSC events in PBL (correlation coefficient was −0.443; p < 0.01). As results, the mean visibility levels only for the polluted-type events were obvious decreasing and dropped under 6.8 during the periods with especially high emissions (e.g., 1996, 2005, and so on; Figure 9a).

However, the different trends of the emissions and the visibility levels in these regions in early 2000 and after 2005 (Figure 9a) indicate their disagreements, which implies that the use of emissions alone cannot fully explain the variation in air quality in the region. For instance, even though the total emissions of most provinces were decreasing during 1995–2000 and 2005–2012, the mean visibility levels in the polluted-type events and in the entire period of winter half year were declining, except those in the clean-type events (Figure 9a). Particularly, when there were higher occurrence frequencies of PSC events for polluted types in the PBL in winter half year of 2012, the mean visibility levels in the polluted-type events and in the entire period of winter half year declined under 6.5 and 6.7, respectively, due to the contribution of the persistent severe fog-haze event over eastern China in January 2013 (Wang et al., 2014; Zhang et al., 2014). Therefore, all these evidences suggest that emissions alone cannot explain fully the long-term trends of air pollution events, as the long-term trends and variation of PSC may play an influential role in air pollution.

Moreover, to check out the impacts of polluted-PSC events on air quality in the entire period of winter half year, Figure 9b depicts long-term trends in frequencies of different visibility levels in the entire period of winter half year from 1961 to 2013. A significantly negative correlation (−0.403; p < 0.01) was also found between the mean visibility level and polluted PSC (Figure 9a). The occurrence frequencies of lower visibility levels 5 and 6 in the entire period of winter half year increased significantly, also demonstrating a close relationship with pollution-related PSC (0.409 and 0.373, respectively; p < 0.01; Figure 9b). All these above significant correlations still imply that the polluted-PSC events contributed to the variations in the mean visibility levels for the entire period of winter half year, which cannot be ignored. By contrast, severe haze pollution events, which are reflected in the frequency of level 5 visibility (Figures 8b and 9b), had a more significant relationship with pollution-related PSC. These results suggest that the occurrences of haze pollution, especially for severe haze pollution events, are significantly modulated by variations in PSC in the PBL.

Finally, Figure 10 shows the percentages of PSC events with four duration scales (2, 3, 4, and greater than or equal to 5 days) in each winter half year during 1961–2013. For PSC events for polluted types in the PBL, the percentages of event durations of 2 days and greater than or equal to 5 days increased generally, while the percentages of events having durations of 3 and 4 days slightly decreased in the past decades (Figure 10a). By contrast, for clean types of PSC events in the PBL, the percentages of durations of 3, 4, and of greater than or equal to 5 days decreased gradually, and the frequency of durations of 2 days increased generally (Figure 10b). These results demonstrate that the durations of PSC events for polluted types and clean types have tended to become longer and shorter, respectively, over the past decades in the PBL, which has been conducive for haze maintenance and prolonged durations of haze pollution days over eastern China. We conclude that mean air quality may have improved because of emission control policies. Nevertheless, the frequency and duration of severe air pollution episodes have increased because of the increase in pollution-related PSC events after 2005.

3.3 Possible Linkages of Long-Term Trends Between PSC in the PBL and Large-Scale Circulation

It is critical to understand the main cause of the long-term trend and its variation in PSC events for polluted types in the PBL over eastern China in the past decades. Many studies have reported that variations in large-scale atmospheric circulations, such as the weakened East Asian winter monsoon and its interannual variation, can significantly influence air pollution in China by directly affecting the pressure distribution, precipitation, humidity, and wind field near the surface and then affecting the advection of air pollutants (e.g., Cai et al., 2017; Ding & Liu, 2014; Jia et al., 2015; Li et al., 2016; Z. Zhang, 2016; Zhu et al., 2012). In addition, Wang et al. (2015) further discovered that a reduction in Arctic sea ice can intensify the haze pollution over eastern China by weakening Rossby wave activity and surface winds. We therefore employed the East Asian winter monsoon and Arctic sea ice indexes to explore the possible evidence of linkages of long-term trends between PSC in the PBL and large-scale circulation.

Figures 11a–11c present temporally normalized long-term variations in occurrence frequency of polluted-PSC events in the PBL in the winter half year, EAWMI, CAI, and the autumn ASI and their corresponding correlation coefficients, respectively. The long-term variation in pollution-related PSC events in the PBL was related to variation in the EAWM (correlation coefficient was 0.228; p < 0.10; Figure 11a), and the CAI had a significantly negative correlation (−0.317; p < 0.05) with pollution-related PSC events (Figure 11b). The correlation between the ASI and polluted-PSC events in the PBL was the most significant (correlation was −0.465; p < 0.01). The ASI had a significantly positive correlation (correlation was 0.362; p < 0.01) with the CAI, which implied that a smaller CAI was associated with less sea ice extent in autumn and cold activities over eastern China. This can be explained by the fact that the reduction in the autumn ASI has led to positive SLP anomalies in midlatitude Eurasia, a northward shift of the track of cyclone activity in China, and weak Rossby wave activity in eastern China south of 40°N during the winter season (Wang et al., 2015; Wang & Chen, 2016). Consequently, less cold activities reached middle-lower-latitude regions.

To address how the Arctic sea ice in autumn can significantly influence local PSC via variation of cold activity, we chose two typically weaker (2007 and 2012 in Figures 11b and 11c) and stronger (1966 and 1969 in Figures 11b and 11c) ASI/CAI winter half years and then calculated the mean surface temperature/SLP/wind field at 850 hPa during winter half year in the four events, respectively (see Figures S2–S5 in the supporting information). Figures 11d–11f showed differences in the combined mean surface temperature/SLP/wind field at 850 hPa between weaker and stronger years. The results show that when ASI and CAI were weaker, surface temperature increased (Figure 11d), SLP transformed to negative anomalies (Figure 11e), and wind field at 850 hPa presented southeasterly anomalies (i.e., northerly wind was weakened in Figure 11f) over eastern China, indicating that the atmosphere was more stable, and persistent synoptic circulations increased frequently in PBL.

Therefore, it can be deduced that a reduction in Arctic sea ice in autumn was favorable for less cold activities and more persistent pollution-related synoptic circulations in the PBL, with higher frequencies of lower visibility levels in eastern China, especially for events lasting more than 5 days; this has contributed positively to the formation and maintenance of persistent haze pollution in the region.

4.1 Discussions

The correlation coefficient and the Hamming distance can be used to analyze the similarities of atmospheric circulations at all pressure levels. Here only SLP and HGT₈₅₀ were selected to explore the similar circulations in the PBL that were closely related to air pollution (e.g., Z. Q. Li et al., 2017; Miao et al., 2017). Moreover, many types of synoptic circulation related to air pollution were identified by other studies (e.g., Chen et al., 2008; Miao et al., 2017; Wang et al., 2014; Yin & Wang, 2017; Y. Zhang et al., 2016). This study only examined the parts of circulations in the PBL—typical examples—to reveal long-term variations in PSC event trends on the basis of previous work (Zheng et al., 2015). Typical examples can serve as proxies for most PSC types to foster the understanding of the contributions of unfavorable long-term trends in PSCs to haze pollution over eastern China. Many more PSC types will be explored in our future work.

Studies (Chen et al., 2008; Ding & Liu, 2014; Li et al., 2016; Wang & Chen, 2016; Wu et al., 2017; Y. Zhang et al., 2016) have reported the effects of climatic conditions on haze or particulate pollution for the purposes of predicting air quality, identifying air pollution causes, and developing emission reduction measures; linkages among the climate index, meteorological factors (e.g., rainfall, wind, and humidity), and hazy days were analyzed. This study systematically explored possible evidence of linkages among haze pollutions, PSC in the PBL, and large-scale circulation during 1961–2013. Our findings demonstrated that the use of PSC with its variation in the PBL could be a useful vehicle for enhancing relationships between synoptic and climatic scales for identifying long-term trends in haze pollution over eastern China.