GRAPES_Meso5.1/CUACE is an online regional atmospheric chemistry model designed for both operational and research applications. There are two major parts of this model, namely a weather model (GRAPES_Meso5.1) and a chemistry model (CUACE). The former is a mesoscale weather prediction model that primarily consists of a fully compressible non-hydrostatical model core and a modularized physics package (Chen et al., 2008); the latter is an online chemistry model that is mainly composed of aerosol and gaseous chemistry modules with emission and dynamic processes (Gong and Zhang, 2008). Wang et al. (2022) established this updated model and provided a comprehensive description of the model. In this model, Peng et al. (2022) implemented the ARI mechanism for the two-way feedback between aerosols and weather processes by incorporating the real-time calculated aerosol optical parameters.

The model domain in this study is centered over the BTH region, covering an area of 33–45∘ N in latitude and 110–125∘ E in longitude (Fig. 1a). The model has a horizontal resolution of 10 km and 49 unevenly spaced vertical levels ranging from the near-surface to 33 km. The physical configuration options selected in this study include Thompson microphysics (Thompson et al., 2008), the Kain–Fritsch (KF) cumulus scheme (Kain, 2004), the rapid radiative transfer model (RRTM) longwave radiation scheme (Mlawer et al., 1997), and the Goddard shortwave radiation scheme (Chou et al., 1998), including the ARI mechanism, the medium-range forecast (MRF) boundary layer scheme (Hong and Pan, 1996), the MM5 surface layer scheme (Zhang and Anthes, 1982), and Noah land surface scheme (Ek et al., 2003). The chemical configuration options mainly include an emissions inventory treatment system, the second-generation Regional Acid Deposition Model (RADM2) gas-phase chemistry (Stockwell et al., 1990), and the CUACE aerosol model (Gong and Zhang, 2008; Wang et al., 2010, 2015a, b).

Five categories of data were used in this study. These include the global Final analysis (FNL) data, with a horizontal resolution of 0.25∘ × 0.25∘ (https://rda.ucar.edu/datasets/ds083.3/, last access: 10 November 2021), provided by the National Centers for Environmental Prediction (NCEP), which was used for the meteorology initial and lateral boundary fields of the model and the analysis of large-scale circulation in upper and mid-levels; multi-year climate average of chemical tracers used for chemistry initialization of the model (Wang et al., 2022); monthly anthropogenic emissions derived from the Multi-resolution Emission Inventory for China (MEIC) of 2017 (Zheng et al., 2018); hourly near-surface PM2.5 mass concentration measured by 149 state-controlled stations provided by the China National Environmental Monitoring Center (http://www.cnemc.cn/, last access: 15 November 2021) and 210 stations provided by the Hebei meteorological service; vertical meteorology data for three sounding stations in the BTH, i.e., Beijing (BJ), Tangshan (TS), and Xingtai (XT), including air pressure, temperature, and wind at 08:00 and 20:00 Beijing time (BJT) each day, as measured by the L-band radiosonde system.

The model simulation was conducted from 29 December 2016 to 31 January 2017, with a looping time of 72 h. The first 72 h simulations were considered to be the spin-up period. To evaluate the impacts of the ARI on the PM2.5 concentration, two numerical scenarios were performed in this study. The first is the controlling simulation (CTL), with the above model configurations and the ARI; the second is the sensitive experiment (EXP), which is consistent with the CTL but without considering the ARI. The same analysis data and emission inventory were used for both numerical scenarios.

Multiple model performance evaluation metrics were used, including the Pearson correlation coefficient (R), the mean bias (MB), the root mean square error (RMSE), the mean fractional bias (MFB), and the mean fractional error (MFE). The equations of these metrics are available in Boylan and Russell (2006).

Accurate reproduction of the aerosol concentration variations and the vertical structure of the atmosphere is a prerequisite for quantifying the ARI (Zhang et al., 2015) and local circulation. Figure 1 shows the distribution of the observed and simulated monthly mean PM2.5 concentrations in January 2017. The BTH region suffered from severe haze pollution in January 2017, with its regional monthly mean of the observed PM2.5 concentration reaching 130 µg m-3. Particularly at the eastern foot of the Taihang Mountains, the central and southern plains of the BTH, the PM2.5 values exceeded 200 and even 250 µg m-3. High anthropogenic emissions coupled with a stable atmosphere due to the mountainous topography lead to frequent and severe haze events in this area (Fu et al., 2014). The BTH is surrounded by the Yanshan and Taihang mountains from north to west, and this topography is not conducive to pollutant dispersion, since the mountains weaken the cold air from the north and west and block the transport of pollutants associated with easterly and southerly winds (Gao et al., 2017; Miao et al., 2015; Quan et al., 2020; Zhong et al., 2018b). A comparison of the observed and simulated PM2.5 results (Fig. 1c–d) shows that although both simulation scenarios reproduced the distribution of PM2.5 concentrations, the CTL results with the ARI were closer to the observations. Particularly in the most polluted central and southern BTH, the maximum PM2.5 concentration exceeded 225 µg m-3 in the CTL, while it was less than 200 µg m-3 in the EXP.

The hourly variation in the observed and simulated PM2.5 concentrations in the BTH was compared in Fig. 2. The model generally reproduced the temporal variation in the observed PM2.5. Compared to EXP, the model including the ARI (CTL) showed better agreements with observations, with higher R (from 0.71 to 0.74), smaller MB (from -40.2 to -16.4 µg m-3), and smaller RMSE (from 57.0 to 45.3 µg m-3; Table 1). Besides, both the MFB and MFE showed substantial reductions from -34.2 and 37.6 µg m-3 to -15.7 and 28.5 µg m-3, respectively (Table 1). According to the model performance goals for PM2.5 proposed by Morris et al. (2005), both the CTL and EXP simulation results achieved an average level (MFB ≤ ±60 % and MFE ≤ 75 %), with the CTL exceeding a good level (MFB ≤ ±30 % and MFE ≤ 50 %) and very close to an excellent level (MFB ≤ ±15 % and MFE ≤ 35 %). The results not only demonstrate the applicability of the model but also the necessity for considering the ARI for pollutant simulation. Moreover, the ARI effect strongly depends on the PM2.5 concentrations (Peng et al., 2021; Zhang et al., 2022). In this study, the ARI effect was significant when the PM2.5 concentration was larger than 100 µg m-3 (Fig. 2), implying that the ARI effect can significantly improve the model's underestimation of the PM2.5 peaks. Considering the essential influence of the ARI mechanism on PM2.5 extremes and the accuracy of the simulation results, three pollution periods were selected, namely 1–7, 16–18, and 23–26 January. Furthermore, since the short-term characteristics of the local circulation have a greater impact on the PM2.5 distribution than the long-term ones, it is necessary to select 1 d from each of the three periods to further investigate the link between the local circulation and ARI and how their interaction enhances heavy haze pollution. According to the criteria of the air pollution level (HJ633–2012) issued by the Ministry of Environmental Protection of China, heavy pollution is defined as the daily mean PM25 concentration larger than 150 µg m-3. Therefore, 6, 17, and 25 January, which reached the heavy pollution level and were in the rising stage, were finally selected as the representatives of heavy pollution days in three periods.

Given the important influence of atmospheric vertical structure, especially the temperature stratification, on the formation of pollutants, we further evaluated the model performance (CTL) in simulating the vertical profile of the potential temperature (PT) at BJ, TS, and XT by comparing the sounding observations. As shown in Fig. 3, the model simulations reasonably reproduced the vertical distribution of temperature in BJ, TS, and XT, including a good simulation of atmospheric warming during the pollution period. For the three pollution periods (1–7, 16–18, and 23–26 January), the mean values of R for PT below 2500 m were 0.94, 0.97, and 0.97 in BJ, TS, and XT, respectively; for the heavy pollution days (6, 17, and 25 January), the R values were 0.83–0.99 in BJ, 0.96–0.99 in TS, and 0.90–0.99 in XT. The accurate representation of near-surface PM2.5 concentrations and vertical temperature structure in the model provides a solid basis for clarifying the physical mechanisms of heavy pollution.

Previous studies have shown that persistent pollution is influenced not only by the PBL and surface meteorology but also by the configuration of the upper and lower-level circulation systems (Miao et al., 2015; Wu et al., 2017). Based on the distribution characteristics of pollutants in the BTH region and the good simulation performance of the model, the horizontal distribution of the upper, middle, and lower atmosphere and surface circulation field in the pollution days was examined in this section. Before the specific discussion of the 3 heavy haze days, we made a general analysis of the weather situation in the three pollution periods. Figure S1 (in the Supplement) shows the average distributions of geopotential height (GH), PT, and wind vectors at the 500 and 700 hPa levels from the FNL data. It can be seen that the BTH region was controlled by northwesterly airflow at upper and mid-levels during the pollution periods, and its PT increased with height. This is a typical atmospheric circulation that often accompanies pollution episodes. The distributions of simulated (CTL) GH and PT at 850 hPa were similar to those at 500 and 700 hPa (Fig. S2). The BTH region was dominated by a uniform pressure field in front of the high-pressure system. The weak pressure gradient resulted in weak westerly wind flows, which blocked dry and cold polar air into this region. Under such stable atmospheric circulation conditions, the PBL development over the BTH was suppressed, which led to a lower height of the PBL (PBLH) and near-surface wind speeds, contributing to the formation and maintenance of haze pollution (Fig. S3).

Based on the above results, we further compared the weather situation characteristics of the selected heavy pollution day in each period. The circulation patterns across the BTH differed considerably during these 3 d (Fig. 4). On 6 January, eastern China was in front of a weak north–south trough at 500 hPa, and the BTH region was controlled by southwesterly airflow and a slight temperature gradient. On 17 January, East Asia was dominated by a zonal circulation, and the BTH region was mainly controlled by westerly airflow. On 25 January, mainland China was dominated by a northeast–southwest high-pressure ridge, and the BTH region was controlled by westerly and northwesterly airflow over this ridge. Moreover, the circulation patterns at 700 hPa were consistent with that at 500 hPa. All of these synoptic conditions are generally considered to promote the deterioration of pollution since they impede the southward movement of cold air from the north and west and strengthen the downdrafts (Wu et al., 2017; X. Zhang et al., 2019).

Figure 5 displays the distribution of the simulated GH, PT, and wind vectors at 850 hPa level on the 3 d. On 6 January, most of the BTH region was between two weak high pressures. Influenced by the southwesterly and southeasterly winds, the warm and humid air masses from the south and the sea were brought to the central and southern BTH. The northern BTH was mainly controlled by low pressure to the northeast, and the northwesterly winds in the area were reduced due to the blockage of the Taihang and Yanshan mountains. On 17 January, the BTH region was between the low pressure in the northeast and the high pressure in the southwest, leaving most of the BTH under the control of northwesterly winds. On 25 January, eastern China was under the control of subtropical high pressure. The BTH region was located north of the subtropical high center, with southwesterly winds prevailing. Such a wind field is conducive to bringing warm and humid air masses from the south to the BTH.

The northerly or northwesterly airflow in the lower troposphere tends to form a sink motion on the leeward slope after being blocked by the Yanshan and Taihang mountains, which will lead to a decrease in PBLH and wind speed in the plains of the BTH region (Fig. 6). On the 3 d, the daily mean PBLH was generally below 300 m in the central and southern plains of BTH, and its low-value area corresponded well to the high-value area of near-surface PM2.5 concentrations. Moreover, due to the disturbance of the local circulation caused by the Yanshan and Taihang mountains, the near-surface wind field showed different distribution characteristics from the lower troposphere with wind speeds below 2 m s-1 in most areas. On 6 January, high PM2.5 concentrations were concentrated in eastern BTH, and a northeast–southwest transport channel was formed under the influence of northeasterly winds. This distribution of wind field was consistent with the average result during the three pollution periods (Fig. S3b). On 17 January, northwesterly winds prevailed in the northern Beijing area due to strong airflow in the lower troposphere. However, the blockage of the mountains weakened the airflow sharply in the plain area after it crossed the mountains, forming subsidence and a weak divergence, which led to a large accumulation of pollutants there. A similar transport channel was found on 25 January, although its wind field was completely different from that on 6 January. Most of the plains are controlled by southerly winds. The southerly winds formed a convergence zone with the northerly winds to the south of Beijing, which was not conducive to the outward dispersion of pollutants. Furthermore, the southerly winds brought warm and humid air masses from the south, which facilitated the formation of secondary pollution.

Under these weak weather-scale systems, the local circulation may dominate the distribution of pollutants and the development of haze. Figure 7 displays the daytime vertical circulation vectors and PM2.5 concentrations along the west–east cross section (to the east of the Taihang Mountains) and along the south–north cross section (to the south of the border between the Taihang and Yanshan mountains), respectively, on the 3 d. For the cross section along the west–east, a clockwise vertical circulation was formed in the lower level on 6 January. The westerly airflow sharply weakened after crossing the mountains, and the zonal wind speeds at lower levels (<500 m, where the pollution was most severe) were mostly below 2 m s-1; the differential heating between the mountain slopes and the plains caused the atmosphere on the slopes to rise with relative heating and the atmosphere on the plains to sink with relative cooling, resulting in a weak clockwise local circulation between the eastern Taihang Mountains and the BTH plain (119∘ E; Fig. 7a); pollutants then accumulated in the PBL through this recirculation. There was a similar vertical circulation on 25 January (Fig. 7c). However, this circulation was limited (between the eastern Taihang Mountains and 118∘ E) due to the control of lower and near-surface southwesterly winds. On 17 January, although a sinking motion occurred within the PBL, the zonal wind speeds were larger throughout the layer compared to the other 2 d, mostly ranging between 2 and 5 m s-1 (Fig. 7b). The stronger westerly winds made it relatively easy for pollutants to disperse eastward, thus the PM2.5 concentrations were lower than those on the other 2 d. For the cross section along the south–north, the PM2.5 concentrations on 6 January were significantly lower than those on 17 and 25 January. On 6 January, northeasterly winds prevailed near the surface of central and southern BTH, and the pollutants were transported from northeast to southwest via this channel (Fig. 6d). The airflow over the mountains formed a whole layer of subsidence near 38∘ N (Fig. 7d), which inhibited the upward transport of pollutants; at the same time, there was a vertical local circulation at 33–37∘ N, between 700 and 1500 m (Fig. 7d), which made pollutants recirculate in this region and not easily disperse to the outside. However, due to the high altitude of this circulation, its restrictions on pollutants were not as strong as the zonal circulations on 6 January (Fig. 7a) and 25 January (Fig. 7c). On 17 January, a wind convergence zone appeared in the lower level (<1000 m) near 37∘ N (Fig. 7e). The combined effect of southerly and northerly winds made the pollutants difficult to disperse outward, thus leading to their local accumulation. On 25 January, southerly winds prevailed throughout the layer below 1500 m (Fig. 7f), which is a typical meteorological condition leading to severe air pollution in this region (Huang et al., 2020; X. Zhang et al., 2019; Zhong et al., 2018b). The southerly winds facilitated the accumulation of local pollutants by weakening their horizontal diffusion and bringing in warm and humid air masses.

Given the considerable impacts of the local circulation and ARI on the distribution of PM2.5 extremes, we further analyzed the potential link between them. According to the location of the cross sections shown in Figs. S3 and 8, the vertical distribution of the wind field and PM2.5 with and without the ARI mechanism and the difference in the horizontal wind speeds induced by the ARI can be seen. For the cross section along the west–east, the ARI strengthened the clockwise vertical circulation near 500 m by simultaneously enhancing the westerly winds in the upper level (500–1000 m) and the easterly winds in the lower level (<500 m; Fig. 8a–c). For the cross section along the south–north, the ARI strengthened the circulation at high altitudes between 33 and 35∘ N by enhancing the upper (900–1500 m) southerly winds and the lower (<500 m) northerly winds (Fig. 8d–f). Moreover, the southerly winds in the circulation formed a wind convergence zone with the northerly winds on its northern side (Fig. 8d–e), and the ARI strengthened this zone by increasing the wind on both sides (Fig. 8f).

As the local circulation characteristics were different on the 3 heavy haze days (Fig. 7), it was necessary to investigate the impact of the ARI on each day. As shown in Figs. 9–12, the impacts of the ARI on circulations could be broadly classified into two types, namely strengthened local circulation and weakened horizontal diffusion. First, the ARI significantly strengthened the vertical zonal circulations on 6 and 25 January. The high aerosol concentrations concentrated on the BTH plain during the daytime substantially cooled the lower atmosphere by absorbing and scattering solar radiation, and the widening difference in the atmospheric heating between the mountain slopes and the BTH plain led to a simultaneous strengthening of the westerly winds in the upper level and easterly winds in the lower level. On 6 January, between the eastern Taihang Mountains and 119∘ E, the ARI increased the westerly winds (300–800 m) and the easterly winds (<300 m) by 0.8 and 0.3 m s-1, respectively, on average (Figs. 9a, 11a, d). On 25 January, the ARI increased the westerly winds by 0.6 m s-1 at the same location as on 6 January; meanwhile, the airflow below 300 m changed from weak westerly to easterly under the ARI effect, forming a closed circulation with the upper westerly winds (Figs. 9c, 11c, f). Moreover, the ARI could change the altitude of the circulation. On 6 January, the ARI shifted the vertical circulation downward by about 100 m, according to the wind speed minimum and the height of the wind shear (Fig. 11a, d). The stronger and lower vertical circulation caused a further accumulation of pollutants in the lower level, which then led to substantial cooling of the lower atmosphere (Fig. 9d, f) and weak vertical turbulent diffusion (Fig. 9g, i). The ARI also strengthened the meridional circulations on 6 January, although it did not cause the same downshift as the zonal circulations. Due to the ARI effect, the southerly winds in the upper level and the northerly winds in the lower level were strengthened simultaneously (Figs. 10a, 12a, d). The strengthened vertical circulation in this local area likewise traps the pollutants in the limited space, which further cooled the atmosphere (Fig. 10d) and weakened the turbulent diffusion (Fig. 10g) in the lower atmosphere.

Second, the ARI weakened the horizontal diffusion of pollutants. For the relatively lightly polluted northern BTH on 17 January, the ARI weakened the westerly winds below 300 m and east of 117∘ E (Figs. 9b, 11b, e), with a maximum wind speed reduction of 1 m s-1. In addition, the ARI enhanced the sinking of airflow near 116∘ E (Fig. 11b, e), promoting the accumulation of aerosols in the lower layer. For the heavily polluted southern BTH on 17 January, the ARI enhanced the wind convergence zone near 37∘ N by simultaneously strengthening the southerly winds south of the convergence line and the northerly winds north of the convergence line (Figs. 10b, 12b, e). At the same time, the ARI pushed this convergence line northward, causing southerly winds to prevail below 200 m over the plain. In addition, the ARI enhanced the southerly winds below 600 m along the south–north on 25 January (Figs. 10c, 12c, f), with an average increase of 0.6 m s-1. Pollutants accumulated locally due to the blockage of the Yanshan Mountains, which then led to the cooling of the lower atmosphere (Fig. 10e–f) and the weakening of the vertical turbulent motion (Fig. 10h–i).