Since January 2013, the Ministry of Environmental Protection has been monitoring the PM2.5 mass concentrations in real time at over 1000 environmental monitoring stations established in different regions of China. In this study, we used the hourly PM2.5 mass concentrations provided by the Ministry of Environmental Protection from 1 December 2016 to 10 January 2017 and the respective averaged PM2.5 mass concentrations of all the urban stations in Xi'an, Yuncheng, Shenyang, Chengdu, Wuhan, Nanjing, Shanghai, Jinan, Guangzhou, and Qingyuan. The in situ monitoring data of the hourly concentrations of PM2.5, PM10, CO, NO2, SO2, and O3 were acquired from the national air quality real-time publishing platform (http://106.37.208.233:20035, last access: 4 March 2019).

We also used an unmanned aerial vehicle to observe the PM2.5 mass concentration at different heights at 1 km intervals every 3 h in Nanjing from 3 to 4 December 2017. The data were obtained with the CEEWA X8 UAV platform (Nanjing CEEWA Intelligent Technology Co., Ltd., Nanjing, China), a six-rotor industrial UAV carrying a highly reliable, triple-redundant FC-IU3 flight control system (Zhou et al., 2018). A multiparameter atmospheric environment detector was equipped on the platform, which was developed by Shenzhen Tengwei Measurement and Control Technology Co., Ltd., Shenzhen, China (Zhou et al., 2018). The experiment was performed on the Xianlin campus of Nanjing University. This campus is located in the eastern suburb of Nanjing and is surrounded by farmland, residential areas, and small patches of forest and chemical plants. The experimental site was located on a playground which was not surrounded by any tall structures.

In China, 120 stations observe vertical meteorological factors using L-band sounding radars. Their accurately positioned radar systems collect reliable meteorological data each second; thus, these data have high spatial and temporal resolutions (Tao, 2006). In this study, we used the L-band sounding radar data from the meteorological stations in Xi'an, Shenyang, Chengdu, Wuhan, Nanjing, Shanghai, and Qingyuan; these stations observe several meteorological factors, including wind, temperature, and RH, twice each day at 08:00 and 20:00 BJT (Beijing time) from 1 December 2016 to 10 January 2017. The meteorological factors were analyzed in detail below a height of 3 km. The heights from the surface to 1 km, from 1 to 2 km, and from 2 to 3 km are termed low-level, mid-level, and upper-level, respectively. In addition, due to the lack of meteorological radiosonde observations in Guangzhou, we supplemented related observations from an adjacent city, Qingyuan.

Since 2001, national weather stations have been conducting automatic hourly observations. Starting in 2001, some of the stations began to record observations at 5 or 10 min intervals. This study used the hourly meteorological observation data, including temperature, pressure, RH, wind, and visibility from the national automatic weather stations (AWS) provided by the National Meteorological Information Center of China Meteorological Administration (NMICMA). The time period of the data selected is from 1 December 2016 to 10 January 2017.

We also used an hourly radiant exposure data set of national meteorological radiation factors (V2.0) provided by the NMICMA. This data set contains 104 radiation stations, including the following: primary stations with global, direct, scattered, reflected, and net radiation; secondary stations with global and net radiation; and tertiary stations with only global radiation. These radiation stations have recorded hourly basic radiant exposure data and the corresponding station information (i.e., latitude, longitude, and altitude) since 1993. In this study, we used the global, direct, and net radiant exposure from 1 December 2016 to 10 January 2017.

Based on the definition and calculation formula of PLAM (Parameter Linking Air-quality to Meteorological conditions/haze; X. Y. Zhang et al., 2009; Wang et al., 2012, 2013; Zhang et al., 2015), we obtained the PLAM index values in Xi'an, Nanjing, Wuhan, Qingyuan, Chengdu, and Shenyang using surface meteorological factors. PLAM includes two separate major factors: (1) the initial meteorological conditions α(m) associated with the atmospheric condensation processes, and (2) an effective dynamic parameter associated with the initial contribution of air pollution β(c′): PLAM=α(m)×β′(c). This calculation mainly indicates regional atmospheric stability and air condensation ability. The details of the calculation have been presented in previous studies (Wang et al., 2012, 2013).

ERA-Interim is the ECMWF's latest global atmospheric reanalysis, which extends back to 1979 and continuously updates in real time (Dee et al., 2011). It is produced with a four-dimensional variational data assimilation scheme and advances forward in time using 12 h analysis cycles (Thépaut et al., 1996; Dee et al., 2011). Before assimilation, all data are subject to gross, redundancy, and background quality controls, which results in a large drop between the total number of data available and the number of data used in the assimilation. The mean daily usage rate of radiosondes is no more than 50 % over the entire period (Poli et al., 2010). In addition, although the effect of aerosols on radiative transfer in the atmosphere is modeled based on prescribed climatological aerosol distributions (Dee et al., 2011), it does not consider the two-way feedback mechanism between the cumulated aerosol pollution and deteriorating meteorological conditions (Simmons, 2006). Therefore, the magnitude of the feedback mechanism could be statistically reflected by the gaps between the ERA-Interim reanalysis and the meteorological radiosonde observations. Therefore, the disparities have been used to present the observational evidence of aerosol–planetary boundary layer interactions in Beijing (Ding et al., 2016; Ding et al., 2016).

In this study, we used ERA-Interim data with a horizontal resolution of 0.125∘ × 0.125∘. Its mandatory pressure levels include 1000, 975, 950, 925, 900, 875, 850, 825, 800, 775, 750, and 700 hPa. According to these pressure layers, we interpolated the radiosonde observations and calculated the vertical temperature differences between the ERA-Interim reanalysis and the interpolated sounding data at 20:00 BJT.

To the north of the Loess Plateau and the south of the Qinling Mountains, the Guanzhong Plain has a narrow and closed terrain (Fig. 2), and its climatic and meteorological conditions are distinctive from those of the surrounding areas. Compared with the plateau to the north, the Guanzhong Plain is less affected by northerly cold, clean winds, and these conditions favor the accumulation of pollutants. However, because the Loess Plateau is lower in elevation than the Hengduan Mountains and the Daba Mountains located to the northwest of the Sichuan Basin, the barrier effect of the plateau on the northerly cold air is weaker than that of the abovementioned mountains (Figs. 4b, 12b). Because the North China Plain is bordered to the west by the Taihang and the Lüliang Mountains (Fig. 2), the Guanzhong Plain is rarely affected by pollutant transport from the North China Plain; however, air pollution is highly correlated among the different cities in the Guanzhong Plain. To the west of this plain, Xi'an lies north of the Wei River and the Loess Plateau and south of the Qinling Mountains (Fig. 2). Due to its enclosed topography, Xi'an frequently experiences heavy urban aerosol pollution.

From 1 December 2016 to 10 January 2017, two HPEs occurred in Xi'an and persisted for more than 7 days with peak mass concentrations greater than 400 µg m-3 (Fig. 4a, dark blue lines). During HPE1-2, we observed a striking two-way feedback mechanism between the deteriorating weather conditions and the cumulated aerosol pollution (Fig. 4, red and white boxes). When the near-ground PM2.5 accumulates to a certain extent, the particles scatter more solar radiation back to space, which substantially reduces the surface radiation (Fig. 4e, red boxes) and consequently lowers the near-surface temperature (Fig. 4c, white boxes). Under slight or calm winds (Fig. 4b, red boxes), the temperature reduction induces or reinforces inversions, which further weaken turbulent diffusion and suppress the diffusion of water vapor and pollutants (Zhong et al., 2017, 2018a); these conditions also decrease the near-ground saturation vapor pressure to increase the RH (Fig. 4d, red boxes), which further enhances aerosol hygroscopic growth and accelerates liquid-phase and heterogeneous reactions (Cheng et al., 2016; Fang et al., 2016; Tie et al., 2017). This type of two-way feedback mechanism leads to declining meteorological conditions and elevated PM2.5 mass concentrations.

To quantify the magnitude of the two-way feedback during TSs and CSs in polluted Beijing and Xi'an, we obtained the air temperature difference between the radiosonde observations at 20:00 BJT affected by the two-way feedback (where the temperature profiles were more affected by aerosols blocking in solar radiation transfer compared with observations at 08:00 BJT) and the ERA-Interim reanalysis data without the feedback. We found that the temperature profile was modified by aerosols during both the TSs and the CSs in Beijing and Xi'an (Fig. 5a, b). However, a comparison of aerosol-induced temperature modification in the TSs and the CSs indicates that the lower cooling bias was more striking in the CSs, and that this bias also became increasingly substantial at distances closer to the surface. From TSs to CSs, the negative temperature difference at 1000 hPa increased from -0.8 to -4.3 ∘C and from -1.3 to -2.7 ∘C in Beijing and Xi'an, respectively. Using the near-ground temperature reduction (1000 hPa) as an index to evaluate the magnitude of the two-way feedback from TSs to CSs, we found that during TSs the aerosol-induced cooling bias was 18.6 % and 48.7 % of the difference during CSs in Beijing and Xi'an, respectively. This is expected because aerosol pollution worsened from the TSs to the CSs with increasing radiative cooling effects. Moreover, although relatively strong winds in the TSs were conducive to pollution transport, they were unfavorable for the formation and maintenance of stable stratification, in which aerosol self-induced pollution deterioration frequently occurred.

During HPE1-2, we also observed an increase in the PM2.5 mass concentration caused by pollutant transport. The aerosol pollution in Xi'an might have been aggravated by the transport of pollutants from the eastern polluted plain area with heavily polluted cities, including Yuncheng and Linfen. To reveal the effects of air pollutant transport from the eastern plain on the aerosol pollution in Xi'an, we compared the trends in the variation of the PM2.5 mass concentrations in Xi'an and Yuncheng under lower northwesterly winds (Fig. 4a, b). We found that during TSs (Fig. 4, orange boxes), low-level northwesterly winds transported pollutants below the BL which maintained or aggravated the aerosol pollution in Xi'an when Yuncheng was heavily polluted; however, when the air quality in Yuncheng was good, the aerosol pollution in Xi'an was lighter or even eliminated.

In addition to the scavenging effect of clean northwesterly winds on aerosol pollution, pollution elimination mainly depends on lower strong northwesterly winds and mid-upper level southerly winds. Because the Loess Plateau north of Xi'an is sparsely populated with rare air pollutant emissions, lower strong and clean northwesterly winds would blow aerosol pollutants away in Xi'an, causing a subsequent rapid improvement in the air quality (Fig. 4a, b). As the mid-upper level southerly winds transport water vapor to Xi'an from the area south of the city, the mid-upper (or whole-layer) RH level is considerably enhanced (i.e., greater than 96 %; Fig. 4b, d – brown boxes), which causes the PM2.5 to enter the fog-cloud phase and possibly produces precipitation that eliminates pollutants through wet removal (Fig. 4d, blue dots represent precipitation).

Located in the lower reaches of the Yangtze River, the Yangtze River Delta is a triangle-shaped metropolitan region. It covers an area of 211 700 km-2 and is home to more than 150 million people as of 2014 (http://www.ndrc.gov.cn/zcfb/zcfbghwb/201606/t20160603_806390.html, last access: 4 March 2019). The urban buildup in this area has given rise to what may be the largest concentration of adjacent metropolitan areas in the world. The Yangtze River Delta has a marine monsoon subtropical climate with cool, dry winters. Situated in the Yangtze River Delta, Nanjing is the second largest city in the eastern China region. The southern, northern, and eastern sides of the city are surrounded by the Ningzheng Ridges (Fig. 2), whereas the Yangtze River flows west of the city and along part of the northern margin.

From 1 December 2016 to 10 January 2017, four aerosol pollution episodes occurred in Nanjing (Fig. 6a, blue boxes). One of these episodes lasted for less than 3 days and exhibited light pollution, whereas the other three episodes persisted for more than 5 days and had peak mass concentrations greater than 150 µg m-3; thus, these three episodes were termed HPEs (Fig. 6a). During these three HPEs, although the PM2.5 mass concentration was much lower than that in Beijing, the aerosol pollution formation was similar to that in Beijing, including earlier TSs and later CSs (Zhong et al., 2017, 2018a). During the TSs in the HPEs, strong northerly winds transported aerosol pollutants from the polluted North China Plain to the Yangtze River Delta region below and above the BL (i.e., long-distance pollution transport), which induced a striking increase in the PM2.5 mass concentration in Nanjing and a reduction in the PM2.5 mass concentration in Jinan, a regional center city representative of the pollution conditions in the NCP (Fig. 6a, b). To some extent, based on the PM2.5 mass, the two-way feedback mechanism was activated during the CSs, in which we observed surface radiation reductions, near-surface inversions, low-layer RH enhancement, and increased PM2.5 mass concentrations under slight winds (Fig. 6). Due to the lighter aerosol pollution in Nanjing, the two-way feedback mechanism was weaker than that in Beijing (Figs. 1, 6a). In addition, the mechanism might have been weakened by relatively strong lower winds (compared with those in Beijing) (Figs. 1, 6b), which are unfavorable for the accumulation of aerosols.

To reveal the regional pollutant transport patterns from the North China Plain to the Yangtze River Delta region, we calculated the concentration difference in the PM2.5 mass between the start time and the end time of the TSs in HPE1,2,4 (Fig. 7). We found that the southern area of the North China Plain experienced a substantial reduction in its PM2.5 mass concentration, whereas an increase occurred in the middle and lower reaches of the Yangtze River, including the Two Lakes Basin and the Yangtze River Delta region; these results indicate the process of regional pollutant transport from northern China to eastern China under strong northwesterly winds. In the winter of 2017, we also observed this pollution transport (Fig. 6, orange boxes). As shown in Fig. 8, we observed mid-level pollutant transport at the beginning of the TS, after which near-surface PM2.5 reaches the highest concentration with downward mixing. Subsequently, another northerly pollutant transport reached middle layers over Nanjing. Afterward, persistent northerly winds blew pollutants away (Fig. 8, purple boxes). In addition to the “blowing effect” of persistent northerly winds, eliminating pollution in Nanjing mainly depends on strong southeasterly winds, which transport warm, humid, clean air from the Yellow Sea and the East China Sea; this air also blows the aerosol pollutants in Nanjing away (Fig. 6b–d). In addition, transported water vapor increases the RH (Fig. 6b, d), which causes the PM2.5 to enter the fog-cloud phase and possibly produces precipitation that eliminates pollutants through wet removal (Fig. 6d, blue dots represent precipitation).

Consistent with the results observed in Nanjing, Shanghai also experienced long-distance pollution transport below and above the BL under northwesterly winds (Fig. 9a, b – orange boxes). After PM2.5 had accumulated to some extent, we observed a two-way feedback mechanism, including reduced radiation, near-surface inversions, RH enhancement in the lower parts of BL, and an increase in the PM2.5 mass concentration under slight or calm winds (Fig. 9a–e, red and white boxes); however, the magnitude of the feedback was weaker than that observed in Nanjing (Fig. 6). Because Shanghai is closer to the sea than Nanjing, it is more susceptible to warm, humid southeasterly winds from the sea, which carry more water vapor to Shanghai than to Nanjing (Figs. 6 and 9b, d).

The Two Lakes Basin is in the middle reaches of the Yangtze River. With the Sichuan Basin bordered to the northwest by the Daba Mountains (Fig. 2), the Two Lakes Basin is rarely affected by pollutant transport from polluted cities in the Sichuan Basin. The north side of the Two Lakes Basin is connected to the North China Plain via the “Suizhou corridor” and the Nanyang Basin (Fig. 2); thus, the Two Lakes Basin is vulnerable to pollution transport from the North China Plain, which experiences the heaviest aerosol pollution in China (Fig. 3). As a large exorheic basin surrounded by low ridges or mountains, the Two Lakes Basin more frequently exchanges air masses with its surroundings, with wind speeds much higher than those in the Sichuan Basin. Situated in the eastern Two Lakes Basin, Wuhan is the most populous city in central China. The Yangtze and Han rivers wind through this city, which has a southern hilly and central flat terrain (Fig. 2).

From 1 December 2016 to 10 January 2017, four aerosol pollution episodes occurred in Wuhan (Fig. 10a, blue boxes). Three of these episodes lasted longer than 5 days and had peak mass concentrations greater than 150 µg m-3, and are termed HPEs (Fig. 10a). During these three HPEs, we observed a two-way feedback mechanism (red boxes in Fig. 10), including surface radiation reductions, near-surface inversions, low-level RH enhancement, and increases in PM2.5 mass concentrations under slight or calm winds. Similar to the conditions observed in Nanjing, Wuhan experienced lighter aerosol pollution than Beijing (Figs. 1, 10a); thus, the two-way feedback mechanism is weaker than that observed in Beijing.

Figure 7 shows the regional pollutant transport from the North China Plain to the Two Lakes Basin, which also aggravates the PM2.5 pollution in Wuhan. As shown in the orange boxes in Fig. 10, the lower northerly winds transported pollutants from the north of Wuhan to below Wuhan and sometimes above the BL, which resulted in increasing PM2.5 mass concentrations. Therefore, favorable northerly winds establish a pollution linkage between the North China Plain and the middle and lower reaches of the Yangtze River (including the Yangtze River Delta and the Two Lakes Basin), which have low and flat terrains (Fig. 2). However, if the northerly winds are persistent and strong enough, they blow the aerosol pollutants out of the North China Plain entirely and then transport clean, cold winds to Wuhan; under these conditions, the PM2.5 mass concentration first increases and then decreases dramatically. This phenomenon was observed from 12 to 14 December 2016 and is shown in Fig. 10.

In addition to the blowing effect of the strong, persistent northerly winds, clearing the pollution in Wuhan mainly depends on the mid-upper level southerly winds, particularly the southwesterly winds, which transport water vapor to Wuhan from the south, substantially enhancing the RH (over 96 %) (Fig. 10b, d, brown boxes); these conditions cause the PM2.5 to enter the fog-cloud phase and often produce precipitation that eliminates pollutants through wet removal (Fig. 10d, blue dots represent precipitation).

Located in southeastern Guangdong Province, the Pearl River Delta is one of the most populous and densely urbanized regions in the world. This low-lying area is surrounded by the Pearl River estuary, where the Dong, Bei, and Xi rivers converge to flow into the South China Sea. With the South China Sea to its south, the Pearl River Delta region is often influenced by southerly sea winds; however, with the mountainous area in northern Guangdong to the north (Fig. 2), the Pearl River Delta region is less affected by northerly cold, clean winds. Situated at the heart of the Pearl River Delta region (Fig. 2), Guangzhou is the most populous city of Guangdong Province. However, due to the lack of a meteorological radiosonde station in Guangzhou, we had to use the sounding observations from Qingyuan, a city with similar PM2.5 variation trends (Fig. 9a); Qingyuan is located approximately 60 km north of Guangzhou.

From 1 December 2016 to 10 January 2017, the PM2.5 mass concentration in Guangzhou and Qingyuan is ∼50 µg m-3, which is much lower than that in Xi'an, Nanjing, Wuhan, Chengdu, and Shenyang. During this period, only one HPE occurred, and it lasted for more than 8 days with a peak mass concentration of approximately 150 µg m-3 (Fig. 11, blue line). During this episode, we observed surface radiation reductions, near-surface inversions, low-level RH enhancement, and increases in the PM2.5 mass concentration under slight or calm winds (Fig. 11, red/white boxes below the blue line), which suggest that a two-way feedback mechanism exists in the region. Except for this episode, we found that the PM2.5 mass concentration increased during slight or calm winds but was still below the threshold (Fig. 11, the red boxes before 1 January 2017) (Zhong et al., 2019); thus, no inversion or increased RH occurred because the two-way feedback mechanism was not effectively activated.

Clearing pollution from Qingyuan depends on the lower strong northeasterly winds, which transport dry, cold, clean air to decrease the temperature and RH and blow aerosol pollutants away from Qingyuan (Fig. 11b, d, purple boxes). In addition to the blowing effect of the cold northeasterly winds, the aerosol pollution in Qingyuan is also affected by the mid-upper level sea flows, which enhance the atmospheric RH and cause the PM2.5 to enter the fog-cloud phase in addition to possibly producing precipitation that eliminates pollutants through wet removal (Fig. 11d, blue dots represent precipitation).

Located in the upper reaches of the Yangtze River in southwestern China, the Sichuan Basin is a lowland region surrounded by mountains on all sides (Fig. 2). Abutting the eastern edge of the Tibetan Plateau to the west and northwest and the Daba Mountains and the Wu Mountains to the east and northeast, respectively (Fig. 2), the Sichuan Basin is rarely affected by cold northerly winds, which are blocked by the high mountains. On the southern and southeastern sides, the Sichuan Basin is flanked by the lower Yungui Plateau (Fig. 2), which is frequently affected by warm, humid southwesterly and southeasterly airflows from the Bay of Bengal and the southeastern sea. Transported water vapor from the south is blocked by the tall northern mountains and then accumulates in the Sichuan Basin. Located at the western edge of the Sichuan Basin, Chengdu is surrounded by the highlands to the south, the high and steep Longmen Mountains to the northwest, the Qionglai Mountains to the west, and the low Longquan Mountains to the east. The enclosed topographical features lead to a lower wind speed and a higher RH in Chengdu than in other parts of the Sichuan Basin.

From 1 December 2016 to 10 January 2017, three HPEs occurred in Chengdu (Fig. 12, blue boxes), and these episodes lasted for more than 10 days and had peak mass concentrations greater than 200 µg m-3 (Fig. 12a). During these three episodes, we observed thick mid-upper level fog/cloud above Chengdu (Fig. 12d), which was blocked by the surrounding mountains and upper-level inversions. The mid-upper level cloud competes with the near-surface aerosols for solar radiation, i.e., as more solar radiation is reflected by the mid-upper layer cloud, the near-surface aerosols receive less solar radiation. Therefore, with cloudy mid-upper layers, more solar radiation is reflected to cool the atmosphere below the clouds, and this condition suppresses the two-way feedback mechanism between the unfavorable weather conditions and the near-surface aerosols. Consequently, the two-way feedback was weak and nearly no near-ground temperature inversion was observed (Fig. 12c). Despite the lack of a two-way feedback mechanism to aggravate aerosol pollution, the increase in the PM2.5 mass concentration is still dominated by slight or calm winds under stable stratification(Fig. 12, red boxes). Comparing the RH variations in the two processes of increasing PM2.5 (Fig. 12 red boxes) during the HPE from 26 December 2016 to 6 January 2017, we found that the PM2.5 mass concentration increases in connection with lower RH.

In addition to the near-surface weak winds, persistent aerosol pollution is a result of temperature inversions caused by the warm southwest advection (Fig. 12b, c, brown boxes). The ground of the Qinghai–Tibet Plateau is a heat source throughout the year (Ye and Gao, 1979); thus, it heats the near-surface ambient air (Fig. 13). When the relatively warm air moves east across the Tibet Plateau under southwesterly winds, it forms an inversion above the basin (Fig. 12c, brown boxes), which caps the convective layer and induces the accumulation of aerosols and water vapor.

Effective pollution clearing rarely occurs in Chengdu because the Sichuan Basin is less affected by the cold, clean northerly winds as a result of the surrounding high northern mountains. However, as soon as aerosol pollutants and water vapor are cleared, aerosol pollution forms again due to more longwave radiation lost from the ground. For example, during the 4–7 December 2016 period, the fog/cloud dissipated, and the PM2.5 mass concentration dropped to a low value on the 5th December (Fig. 14a, d). Due to the absence of cloud/fog blocking, more longwave radiation from the ground was emitted into space on the night of 6th December, and the surface net radiant exposure decreased from -0.58 on the 5th to -1.45 on the 6th December (2.5 times; Fig. 14e). The significant reduction in the surface radiation cooled the near-surface atmospheric temperature, which formed an inversion layer of approximately 50–100 m (Fig. 14c). Capped by the inversion layer, the PM2.5 mass concentration doubled after the night of the 6th December to form another aerosol pollution event (Fig. 14a).

Pollution removal in Chengdu mainly relies on northeasterly winds to blow pollution away. The winds also carry water vapor to add humidity to the atmosphere above Chengdu, which converts pollutants into fog/cloud drops or produces precipitation that removes pollutants through wet removal (Fig. 12d, blue boxes).

The Northeast China Plain lies north of the Liaodong Gulf, west of the Changbai Mountains, east of the Greater Khingan Range, and south of the Lesser Khingan Range (Fig. 2). Due to the low mountains to the northwest, the Northeast China Plain is susceptible to cold, dry northerly air from Siberia in winter. As the largest city in northeastern China with respect to its urban population, Shenyang is located in the southwestern Northeast China Plain (Fig. 2), where the warm, humid southwesterly flows are transported from Bohai Bay.

From 1 December 2016 to 10 January 2017, six aerosol pollution episodes occurred in Shenyang (Fig. 15a, blue boxes), four of which persisted for more than 3 days with peak PM2.5 mass concentrations greater than 200 µg m-3 (Fig. 15a). During these HPEs, we observed surface radiation reductions, near-surface inversions, low-level RH enhancement, and increases in the PM2.5 mass concentration (Fig. 15a–d, red and white boxes) under slight or calm winds (Fig. 15b, red boxes); these conditions indicate the occurrence of the two-way feedback mechanism in Shenyang.

To quantify the magnitude of the two-way feedback during TSs and CSs in Shenyang, the air temperature difference between the radiosonde observations at 20:00 BJT and the ERA-Interim reanalysis data was obtained. Similar to Beijing and Xi'an, the temperature profile was also modified by aerosols during both the TSs and the CSs (Fig. 16a, b). However, the magnitude of near-ground cooling bias was lower than that in Beijing and Xi'an, which is due to relatively light aerosol pollution and windy conditions. From TSs to CSs, the negative temperature difference at 1000 hPa increased from -0.26 to -0.93 ∘C. The mean aerosol-induced cooling bias (1000 hPa) in the TSs was only 28.2 % of that in the CSs.

Compared with those in Xi'an, Nanjing, Wuhan, Qingyuan, and Chengdu, the speeds of the southeasterly or northwesterly winds are strikingly higher in Shenyang. Relatively strong mid-lower southeasterly winds originate from Bohai Bay, and these winds transport warm, humid air that heats and adds humidity to the mid-upper layer above Shenyang (Fig. 15c, d). This air also transports aerosol pollutants to Shenyang because it carries pollutants from populated and polluted southwestern industrial regions, including Anshan. Lower strong northwesterly winds carry dry, cold air from Siberia to remove pollutants in Shenyang (Fig. 15a, b).

As previously mentioned, the weak two-way feedback mechanism in the Sichuan Basin is weakened by the cloudy mid-upper layers, which compete with the near-surface aerosols for solar radiation. However, the mechanism occurred in the Guanzhong Plain, the middle and lower reaches of the Yangtze River, the Pearl River Delta region, and the Northeast China Plain. Dominant scattering aerosols will scatter solar radiation to cause near-ground temperature reduction; absorbing aerosols will absorb solar radiation to heat the upper aerosol layer and subsequently cool the near-ground layer (Boucher et al., 2013a). Previous studies have found that aerosol profiles play an important role in the radiative cooling effects of aerosols, which results in vertical differences in meteorological factors (Wilcox et al., 2016; Wang et al., 2018). To quantify the magnitude of the two-way feedback in these haze regions of China, we obtained the vertical air temperature difference between the radiosonde observations affected by this two-way feedback and the ERA-Interim reanalysis data without feedback in the regional center cities, including Beijing, Xi'an, Shenyang, Wuhan, Nanjing, and Qingyuan (as a substitute for Guangzhou). A previous study established a threshold value for the PM2.5 mass concentration (100 µg m-3) that effectively activates the two-way feedback in HPEs; additionally, a lower threshold value (71 µg m-3) has been identified for lighter HPEs (Zhong et al., 2019). Therefore, based on the diurnal mean PM2.5 mass concentration (from 08:00 to 17:00 BJT), the temperature difference is further classified by the criterion of 100 µg m-3 in the more polluted North China Plain, Guanzhong Plain, and Northeast China Plain and by the criterion of 71 µg m-3 in the less polluted Two Lakes Plain, Yangtze River Delta, and Pearl River Delta.

By comparing the air temperature difference below and above these thresholds in the six cities, we found that the lower temperature profile was strikingly modified by the two-way feedback mechanism (Fig. 17). On the North China Plain, the Guanzhong Plain, and the Northeast China Plain, the lower temperature bias between the sounding observations and the ERA-Interim data was close to zero below the threshold of 100 µg m-3 but immediately became negative above the threshold (Fig. 17a–c). In the Two Lakes Plain, the Yangtze River Delta, and the Pearl River Delta, we observed a similar reduction in the temperature difference below and above the threshold of 71 µg m-3 (Fig. 17d–f). Overall, the magnitude of the two-way feedback mechanism was larger in the North China Plain, the Guanzhong Plain, and the Northeast China Plain than in the Two Lakes Plain, the Yangtze River Delta, and the Pearl River Delta.

For each representative site, the low-level cooling bias was more striking near the surface; additionally, as the PM2.5 mass concentration increased, the low-level cooling bias became more significant (Fig. 17). In Beijing, the negative temperature difference reached more than 2 ∘C with PM2.5 values in the range of 200–300 µg m-3 compared to approximately 1 ∘C in the range of 100–200 µg m-3. In Xi'an, the temperature difference decreased from approximately -1.5 ∘C in the range of 100–200 µg m-3 to 2.5 ∘C in the range of 200–300 µg m-3. In Shenyang, the cooling bias of approximately 0.6 ∘C occurred with the increase in PM2.5 from 100–200 to 200–300 µg m-3. Under the most polluted conditions, the near-ground cooling bias was greater than -4 ∘C, approximately -4 ∘C, and approximately -1 ∘C in Beijing, Xi'an, and Shenyang, respectively, which was substantially affected by the two-way feedback.

To quantify the feedback of deteriorating meteorological conditions on the increasing PM2.5 in the CSs, the PLAM index was used, which mainly reflects the stability of the air mass and the condensation rate of water vapor on aerosol particles. The squared correlation coefficients between the hourly PLAM and PM2.5 mass concentration in the typical processes of PM2.5 increase during the CSs were 0.71, 0.7, 0.72, 0.68, 0.64, and 0.63 in Beijing, Xi'an, Shenyang, Wuhan, Nanjing, and Qingyuan, respectively (Fig. 18a–f); these values exceeded the 0.05 significance level, which suggested that such a meteorological feedback on PM2.5 explained 60 %–70 % of the increase in the PM2.5 during the CSs.