Introduction

The rapid growth of industrialization and urbanization has led to severe and persistent air pollution in many cities of China in the last 30 years (Chen et al., 2017; Yu et al., 2020), which not only reduces atmospheric visibility (Liu et al., 2021), but also endangers human health (Ma et al., 2017). Thus, as one of the most serious environmental problems in urban areas, air pollution has attracted attention from the government, the general public, and researchers. Generally, persistent air pollution is caused by meteorological conditions combined with local and regional sources of pollution (Deng et al., 2019; Gowan & Andrew, 2008).

Many studies have shown that stable synoptic circulation patterns along with other adverse meteorological conditions promote the accumulation of pollutants in the atmosphere, which leads to persistent air pollution (Annalisa et al., 2021; Liu et al., 2020). Liu et al. (2013) found that haze events in Beijing tended to occur more frequently under weak anticyclone synoptic conditions. Yang et al. (2015) reported that the primary factors driving haze formation in Beijing were a lack of atmospheric circulation, while a temperature inversion and high relative humidity (RH) accelerated the formation of regional haze. In addition, many studies have pointed out that regional transport significantly influences air pollution (Chifflet et al., 2018; Kozáková et al., 2019). Wang et al. (2015) provided a rough estimation that 35.5% of PM2.5 in Beijing was attributed to long-distance pollutant transportation using a potential source contribution function (PSCF) model. Zhao et al. (2016) also found that the Yangtze River Delta region was an important source region of humic-like substances in Shanghai by using a backward trajectory clustering and a concentration weighted trajectory (CWT) model.

Lanzhou is an industrial city in northwest China with petroleum refining, chemical engineering, and machinery manufacturing industries as its pillar industries. The city is surrounded by Gaolan Mountain in the south and Baita Mountain in the north (Fig. 1). Specific valley terrain can often make it difficult for pollutants to become diluted and diffused (Zhang & Hu, 1998; Zhang & Li, 2011). In the recent years, anthropogenic emissions caused by the rapid development of Lanzhou have resulted in a serious increase in air pollution, especially in the heating season (Cheng et al., 2021; Guan et al., 2018). During this period, coal-fired emissions and the intensity of pollutant emissions increased significantly, resulting in frequent air pollution in Lanzhou (An et al., 2007; Chu et al., 2008a). Li (2018) found that in the non-heating season in Lanzhou, 20.9% of all days experienced significant air pollution from 2012 to 2016. Corresponding proportions of the heating season were 56.1%; the frequency of polluted days during the heating season was higher than during the non-heating season. From November 1, 2016 to March 31, 2017, several persistent air pollution events occurred, and one of them lasted from November 1 to 21, 2016.This persistent pollution in November 2016 was the worst air pollution event in recent years and had a wide impact and long duration. In order to alleviate air pollution, the local government urgently implemented an air pollution warning and launched traffic restriction measures based on license plate number for motor vehicles. The main goal of this study was to understand the formation mechanism involved in creating persistent pollution, explore potential sources of PM (PM2.5 and PM10), and quantify contributions of distant sources during the heating season from November 1, 2016 to March 31, 2017.

Fig. 1
figure 1

Location of Lanzhou within China: a aerial photo of Lanzhou showing four districts, Anning, Chengguan, Qilihe, and Xigu; b remote sensing photo showing the vicinity of Lanzhou and Beijing within China

Full size image

Data and methodology

Data

The concentration data for air pollutants (PM2.5, PM10, SO2, NO2, and CO) in this study were derived from five state-owned automatic ambient air monitoring stations in Lanzhou. The ground and sounding meteorological data were obtained from Gansu Meteorological Administration in China. Dust storm data came from the monitoring station at Zhongchuan Airport, Lanzhou. The NCEP/NCAR reanalysis data were downloaded from the HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory) website (available at ftp://arlftp.arlhq.noaa.gov/pub/archives/gdas1/).

Methodology

Method for calculating the height of the maximum daily boundary layer

The boundary layer height is an important parameter used to represent how vertical diffusion conditions affect pollutants. It influences the concentrations of air pollutants (Shi et al., 2020). Variations in pollutant concentrations can be caused by changes in the boundary layer height over time (Zhu et al., 2019). Currently, the adiabatic curve method put forward by Holzworth (1964) is the main method used to determine the daily maximum boundary layer height (BLHmax), which was applied in this study to estimate boundary layer height combined with the gradual approximation calculation method developed by Wang et al. (2000). The method was as follow by (1) calculation of the BLHmax using daily maximum temperature data based on the decline rate of the dry adiabatic temperature (γD = 0.976/100 m); it was calculated starting from ground level; the temperature was obtained every 50 m to an altitude of 3000 m; finally, a set of temperature values T1, T2, …, TM was found at equal intervals of height; (2) on the actual temperature stratification curve (08 Radiosonde temperature profile data at Yuzhong meteorological station), a set of temperatures TT1, TT2, …, TTM was also obtained by interpolation in the vertical direction with equal intervals of 50 m; (3) the temperature differences corresponding to T and TT were calculated at each height, and the temperature difference layer with the smallest absolute value was selected and identified as BLHmax.

PSCF and CWT models

TrajStat software was used for cluster and potential source analysis. This software incorporates trajectory statistics and a geographical information system (Wang et al., 2009); it allows users to customize the trajectory and potential source maps, which have been used in a number of prior publications (Huang et al., 2019; Bi et al., 2021). The 72-h backward air mass trajectories arriving at Lanzhou were calculated in this study by using a HYSPLIT model (Draxler & Hess, 1998). The NCEP/NCAR reanalysis dataset was used as meteorological data input to the model. The model was run at 00:00 UTC (08:00 local time (LT)) each day; the starting height was set at 500 m, which was chosen to diminish the effects of surface friction and to represent winds in the lower boundary layer (Hong et al., 2016). To compare the predominant transport pathways, backward trajectories were clustered into groups with similar transport pathways. The TrajStat software provides both angle and distance clustering methods. Here, we chose to use the angular distance, not the Euclidean distance, mainly because our interest was to use the trajectories to determine the direction from which the air masses that reached the site had been transported. The method used to calculate the mean angle between two trajectories was adapted from Sirois and Bottenheim (1995). A PSCF model was created to identify regional sources of PM2.5 and PM10 (Dimitriou & Kassomenos, 2018). The PSCF values for the grid cells were calculated by counting the trajectory segment endpoints that terminate within each cell. The zone of concern was divided into i × j small equal grid cells. The grid covered a domain between 28–80° north latitude and 40–140° east longitude with a 0.5° × 0.5° resolution. The PSCFij was calculated using Eq. (1):

$${\mathrm{PSCF}}_{ij}=\frac{{\mathrm{m}}_{ij}}{{\mathrm{n}}_{ij}}$$
(1)

where n(i,j) is the number of endpoints that fall in the ijth cell and m(i,j) is the number of endpoints that fall in the ijth cell, corresponding to trajectories with pollutant concentrations higher than an arbitrary criterion value. Here, the criterion values were set at 75 μg/m3 for PM2.5 and 150 μg/m3 for PM10. To remove the uncertainty in cells with small nij values, an empirical weight function W(nij) as shown in Eq. (2) was multiplied with the PSCF value (Zhang et al., 2013):

$${\mathrm{w}}_{i,j}=\left\{\begin{array}{c}1.00 {\mathrm{n}}_{i,j}>2.0{\mathrm{n}}_{\mathrm{avg}}\\ 0.70 {\mathrm{ n}}_{\mathrm{avg}}<{\mathrm{n}}_{i,j}\le 2.0{\mathrm{n}}_{\mathrm{avg}}\\ 0.42 0.5{\mathrm{n}}_{\mathrm{avg}}<{\mathrm{n}}_{i,j}\le {\mathrm{n}}_{\mathrm{avg}}\\ 0.17 0<{\mathrm{n}}_{i,j}\le 0.5{\mathrm{n}}_{\mathrm{avg}}\end{array}\right.$$
(2)

A PSCF model reflects only the proportion of pollution trajectories in a grid, and cannot reflect the trajectory of pollution levels. To compensate for this limitation, the CWT method developed by Hsu et al. (2003) was used to calculate the weighted trajectory concentration. The CWT value for the ijth cell was calculated by the Eq. (3):

$${\mathrm{C}}_{i,j}=\frac{1}{{\sum }_{l=1}^{\mathrm{M}}{\uptau }_{ijl}}\sum_{\mathrm{l}=1}^{\mathrm{M}}{\mathrm{C}}_{\mathrm{l}}{\uptau }_{ijl}$$
(3)

where Ci,j is the mean weighted concentration in the cell (i,j), l is the index of the trajectory, M is the total number of trajectories, Cl is the observed concentration related to the arrival trajectory l, and τijl is the time spent in the cell (i,j) by trajectory l. The arbitrary weighting function described above was also used in the CWT analyses to reduce the effect of the small values of nij.

Results

General meteorological conditions and air pollution characteristics

In this study, November 1, 2016 to March 31, 2017 was defined as the heating season. The daily mean concentrations of particulates (PM2.5, PM10) and gaseous pollutants (SO2, NO2, and CO) during the heating period are shown in Table 1. Except for CO, PM10 exhibited the highest concentration among all of the air pollutants, with a daily mean concentration at 177 ± 142 μg/m3, followed by PM2.5 (70 ± 32 μg/m3). About 53% of the daily PM10 concentration exceeded the Chinese Ambient Air Quality Grade II standard of 150 μg m−3 (GB3095-2012), while 37% of the daily PM2.5 concentration surpassed the Grade II standard (75 μg m−3). The daily mean concentrations of SO2, NO2, and CO were 29 ± 12 μg/m3, 64 ± 25 μg/m3, and 1.6 ± 0.71 mg/m3, respectively; the daily NO2 concentration exceeded the Grade II standard (80 μg m−3) 28% of the time; meanwhile, the daily SO2 and CO concentrations never exceeded the Grade II standard (150 μg m−3 for SO2, 4 mg/m3 for CO). This indicates that particulate pollution was extremely severe during the heating season while current situations of gaseous pollution were generally much better. Thus, PM (PM2.5 and PM10) levels are discussed in the next section in detail.

Table 1 Descriptive statistics of the air pollutants during the heating season between November 1, 2016 and March 31, 2017 in Lanzhou
Full size table

Table 2 shows the mean temperature, wind speed, RH, and BLHmax were 3.8 ± 5.5 °C, 1.1 ± 0.3 m/s, 46 ± 13.6%, and 871 ± 493.8 m, respectively. The mean value of temperature, wind speed, and BLHmax above the Grade II standard were lower than the mean value of temperature wind speed and BLHmax below the Grade II standard; however, the mean value of RH above the Grade II standard was higher than the mean value of RH below the Grade II standard. The linear relationship between the meteorological factors and PM was quantified using Spearman correlation tests (Table 3); the correlation coefficients between all of the meteorological factors except RH and PM were significantly negative, which varies between 0.46 and 0.65; the RH showed a significantly positive correlation with both PM10 (r = 0.41) and PM2.5 (r = 0.56) in this study.

Table 2 Statistical characteristics of meteorological parameters in the heating season, below and above the Grade II standard levels during the heating season from November 1, 2016 to March 31, 2017 in Lanzhou
Full size table
Table 3 Spearman correlation coefficients between meteorological factors and particulate matter (PM10, PM2.5) during the heating season from November 1, 2016 to March 31, 2017 in Lanzhou
Full size table

The formation and maintenance mechanism of persistent pollution

Classification of persistent pollution

The concept of a persistent pollution episode refers to a temporal process when pollutants accumulate high in the atmosphere where they remain for some time before cleaner air conditions return. The pollutant levels and durations of such episodes differed in several studies (Sun et al., 2019; Tan et al., 2018). In the present study, an Air Quality Index (as defined by China’s Ministry of Environmental Protection) above 100 for five or more consecutive days was defined as a persistent pollution episode. Figure 2 shows that five persistent pollution episodes were observed according to this definition; Table 3 provides the dates for these episodes. Among them, the second persistent pollution episode had the most intense and sustained period of pollution which lasted for 21 days from November 1 to 21, 2016. A prominent feature of this episode was that abrupt peaks of PM10 were observed without a notable increase in gaseous pollutants. For example, the PM10 concentration increased to 946 µg/m3 on November 10, 2016, which was the maximum daily PM10 concentration. This was mainly caused by a dust storm.

Fig. 2
figure 2

The variations of the concentrations of PM2.5, PM10, SO2, NO2, and CO during the heating season from November 1, 2016 to March 31, 2017 in Lanzhou. Dates for five typical persistent pollution periods (Ep1 to Ep5) are shown; Table 4 provides the dates for these periods

Full size image
Table 4 Characteristics of PM2.5 and PM10 concentrations and meteorological factors in five persistent pollution episodes (Ep1 to Ep5) during the heating season from November 1, 2016 to March 31, 2017 in Lanzhou
Full size table

The mean air pollutant concentrations and meteorological parameters in five persistent pollution episodes are summarized in Table 4. These episodes could be categorized into stagnant accumulation and explosive growth types based on their pollution and meteorological characteristics. The stagnant accumulation type was characterized by high PM2.5 and PM10 concentrations and a high PM2.5/PM10 ratio, which was mainly associated with persistently stable meteorological conditions, such as low temperature and calm wind. The second persistent pollution episode belongs to this type. The explosive growth type was caused by secondary particles in periods with stable meteorological conditions and a massive amount primary particles brought into the study area by dust storms. This type could be divided into two stages: stagnant accumulation (stage 1) and explosive growth (stage 2). Taking Ep1 as an example, the mean temperature and wind speed in stage 1 were low, which were 6.3 ± 1.4 °C and 0.7 ± 0.1 m/s, respectively; the mean relative humidity in stage 1 was relatively high at 65 ± 9%, indicated that the atmosphere was stagnant allowing pollutants to easily accumulate. The PM2.5 and PM10 concentrations were 106 ± 6 μg/m3 and 224 ± 19 μg/m3, respectively, and the PM2.5/PM10 ratio was relatively high at 0.48 ± 0.1. Also, the mean temperature and relative humidity in stage 2 were reduced compared with the previous stage at 5.2 ± 0.9 °C and 47 ± 7%, respectively; this indicated the arrival of cold air. The mean PM10 concentration suddenly increased in stage 2 because of dust particulate transportation accompanied by cold air, which was 503 ± 273 μg/m3. Meanwhile, the PM2.5/PM10 ratio decreased significantly to 0.27 ± 0.1.

Growth pattern of PM 10 and PM2.5 during persistent pollution episodes

Figure 3 shows the variations of the PM2.5 and PM10 concentrations and the PM2.5/PM10 ratio in a high time resolution (1 h) during two selected persistent periods of pollution (Ep1, November 1 to 21, 2016; Ep2, November 31 to December 14, 2016). The others belonged to the explosive growth type and the patterns were similar to that of Ep1, so they were ignored. The PM2.5 and PM10 concentrations had different growth characteristics in the two types (Fig. 3). They showed an asymmetric “saw-tooth cycle” growth in the stagnant accumulation type (Ep2). First, the PM2.5 and PM10 concentrations and PM2.5/PM10 ratio (red line in Fig. 3b) rose gradually, and maintained for some time. This was interrupted by an incomplete removal process. This was followed by a new cycle where the process was not ended until the weather conditions improved. Figure 3a shows that the explosive growth type (Ep1) was divided into two evolution stages; this was also an asymmetric “saw-tooth cycle” growth in stage 1. At the same time, PM10 had an explosive growth accompanied by PM2.5/PM10 ratio falling sharply in stage 2. Finally, as each dust storm weaken, the PM2.5 and PM10 concentrations decreased slowly, while the PM2.5/PM10 ratio also gradually increased.

Fig. 3
figure 3

The high time resolution variation of particulate matter (PM10, PM2.5) and its ratio during two persistent pollution emission periods, Ep1 (November 1 to 21, 2016) and Ep2 (November 31 to December 14, 2016)

Full size image

Synoptic circulation and characteristics of meteorological factors of persistent pollution

An analysis of the 500 hPa synoptic circulation over Lanzhou during the five persistent pollution episodes found that the circulation was primarily controlled by the zonal circulation (such as straight westerly airflow and weak frontal systems). The prevailing airflow was from the west and northwest; when the synoptic circulation shifted, Lanzhou was located at the front of a cyclone system or short-wave trough except for Ep2, indicating that cold air advection transited the region. Examples of the synoptic circulation for selected persistent pollution episodes are provided in Fig. 4. During Ep2, a high-pressure ridge covered the study area on December 30 when the atmosphere was stable (Fig. 4c, d). As the synoptic circulation shifted on November 12, Lanzhou was located behind the trough and in front of the ridge, indicating the atmosphere still remained stable. The synoptic circulation in stage 1 of Ep1 (Fig. 4a) was very similar to that in Ep2 on December 30. One notable difference was that a cyclone system developed in stage 2 of Ep1 (Fig. 4b) and was located to the west of Mongolia; Lanzhou lay on its southeastern edge, and experienced the arrival of cold air (Fig. 5).

Fig. 4
figure 4

Synoptic maps on two persistent pollution periods: a 18:00 local time (LT), November 2, 2016; b 18:00 LT, November 10, 2016 for Ep1 and c 18:00 LT, November 30, 2016; d 06:00 LT, December 12, 2016) for Ep2 and for 500 hPa

Full size image
Fig. 5
figure 5

Modeling of 72 h-backward trajectories during two persistent pollution periods a, b Ep1 (November 1 to 21, 2016) and c, d Ep2 (November 31 to December 14, 2016) using the US National Oceanic and Atmospheric Administration Hysplit Model

Full size image

Variable characteristics of surface meteorological factors and the boundary layer in persistent pollution

Figure 6 shows the daily variations in meteorological conditions for two selected persistent pollution episodes, Ep1 and Ep2. The temperature and wind speed were lower than 2.5 °C and 1.1 m/s during Ep2, respectively (Fig. 6c, d). The BLHmax was very low and varied between 286 and 975 m. The variations in meteorological conditions during Ep1 could be roughly divided into two stages (Fig. 6a, b). The meteorological factors during stage 1 (November 1 to 4, 2016) involved low temperature, wind speed, and BLHmax. Compared to stage 1, the most significant difference was the lower temperatures and RH, and higher BLHmax in stage 2 (November. 5 to 19, 2016).

Fig. 6
figure 6

Daily variations in a, c temperature, b, d Wind and RH, and BLHmax during two persistent pollution emission periods, Ep1 (November 1 to 21, 2016) and Ep2 (November 31 to December 14, 2016)

Full size image

Statistical analysis of potential source areas

Cluster analysis

In order to explore the influence of air masses on PM2.5 and PM10 in Lanzhou, TrajStat software was used to simulate and cluster the 72-h backward trajectories during the study period. All of trajectories were grouped into five clusters, which are described in Fig. 7b. Cluster 1 from the west had the highest frequency (64.6%) of the five clusters, followed by cluster 4 from the southwest with 11.6%, while cluster 2 contributed the least (4.6%) to the total trajectories. Table 5 shows corresponding daily mean PM2.5 and PM10 concentration in five clusters in this study. The highest PM10 concentrations were related to clusters 1 and 5, at 205 ± 176 μg/m3 and 148 ± 48 μg/m3, respectively (Table 5). The highest PM2.5 concentrations were observed in clusters 1 and 4 at 78 ± 33 μg/m3 and 73 ± 27 μg/m3, respectively (Table 5). The lowest PM10 and PM2.5 concentrations, which were 105 ± 27 μg/m3 and 41 ± 33 μg/m3, respectively, were found in cluster 2 (Table 5).

Fig. 7
figure 7

TrajStat software used to simulate and cluster a the 72-h backward trajectory distribution and b mean backward trajectory clusters during the heating season from November 1, 2016 to March 31, 2017 in Lanzhou

Full size image
Table 5 Summary statistics of PM2.5 and PM10 concentrations in Lanzhou associated with clusters 1 to 5 during the heating season from November 1, 2016 to March 31, 2017 in Lanzhou
Full size table

Weighted PSCF and weighted CWT

The weighted PSCF (WPSCF) maps of PM10 and PM2.5 in Lanzhou during the study period is shown in Fig. 8 The higher the value of WPSCF, the greater the impact on PM10 and PM2.5. Higher WPSCF values of PM10 were observed in the Kumtag Desert in the western Xinjiang, the Qaidam Basin and the Hehuang Valley of Qinghai, as well as in the western and eastern Hexi Corridor in Gansu (Fig. 8a). These values were mostly between 0.4 and 0.6, although those of the Hehuang Valley and the eastern Hexi Corridor were above 0.6. Atmospheric aerosols from these areas reached Lanzhou along clusters 1 and 5 and affected the PM10 concentration. Higher WSCF values of PM2.5 appeared in the Kumtag Desert, the Qaidam Basin and the Hehuang Valley of Qinghai, Linxia Hui Autonomous Prefecture in Gansu; their potential contribution to PM2.5 was above 0.4. The PM2.5 in these areas arrived in Lanzhou along with clusters 1 and 4 (Fig. 8b).

Fig. 8
figure 8

source contribution function maps of PM2.5 and PM10 in Lanzhou

The potential

Full size image

The distribution of the CWT shows the contribution of potential sources to PM10 and PM2.5 in Lanzhou (Fig. 9). The Kumtag Desert, Qaidam Basin, and Hehuang Valley, along with the eastern and western Hexi Corridor in Gansu contributed more than 120 μg/m3 to PM10 (Fig. 9a), indicating that these regions made the largest contribution the PM10 concentration in Lanzhou. The Hehuang Valley area and Linxia Hui Autonomous Prefecture contributed more than 60 μg/m3 to PM2.5, indicating that these areas were important sources of PM2.5 (Fig. 9b).

Fig. 9
figure 9

The weighted concentration weighted trajectory maps of PM2.5 and PM10 in Lanzhou

Full size image

Discussion

In this study, we found that particulate pollution is extremely severe during the heating season in Lanzhou, Gansu Province, China. Several important factors affect PM concentrations in Lanzhou, including high emissions from local pollution sources (Chu et al., 2008a), special basin topography, poor atmospheric diffusion conditions (Filonchyk & Yan, 2018; Zhang et al., 2001), and dust intrusions from upwind regions (Feng & Wang, 2012). These factors cause the worst air pollution to occur during the heating season.

Meteorological parameters play an important role in regulating PM concentrations (Li et al., 2019; Zhang et al., 2018). Our results demonstrated that daily temperature had a significantly negative effect on the daily PM concentrations. This might be attributable to weaker local diffusion of pollutants and higher pollution emissions in the heating reason (Chu et al., 2008b). Cold surface temperatures in this study led to a stable and shallow surface layer caused by strong radiative cooling at night, which suppresses the vertical dispersion of air pollutants (Li et al., 2017). In a similar study, Shakil et al. (2021) found a significant negative correlation between temperature and PM at an urban site in Srinagar, India. A significantly negative relationship was observed between the daily wind speed and the daily PM2.5 and PM10 concentrations. Some previous reports found that strong winds flush out pollutants from the atmosphere and slow winds allow pollution levels to rise (Pateraki et al., 2012; Zhou et al., 2019a). Due to the special basin topography, the daily wind speed in Lanzhou is mostly lower than 3 m/s; this low wind speed will often lead to stable atmospheric conditions and therefore, will limit the dispersion of PM leading to the elevated levels of PM observed in this study. This result agrees with a previous study that was conducted in Lhasa, China (Duo et al., 2018). We found that the PBLmax was also negatively associated with the daily PM2.5 and PM10 concentrations. The boundary layer height usually increases along with solar heating. There is less solar heating in the heating season, which contributed the low PBLmax in the heating season. Thus, a low altitude PBLmax hinders the diffusion of pollutants and create favorable conditions for the accumulation of PM pollutants (Murthy et al., 2020). As a result, the PM concentration gradually increased as the PBLmax decreased in this study. Similar results were also reported in other cities of China, such as Beijing, China and Bet Dagan, Israel (Xiang et al., 2019; Yuval et al., 2020). The relative humidity showed a significantly positive correlation with both PM10 and PM2.5 in this study. Hygroscopic growth is one of the most fundamental properties of atmospheric aerosols. Lou et al. (2017) observed that under low (RH < 70%) humidity, the accumulation of PM increased significantly. We also discovered that an increase in RH resulted in an increase in the PM concentration due to the hygroscopic growth of particles. This is consistent with consistent with some studies in the Pearl River Delta, China, and in Srinagar, India (Hu et al., 2021; Romshoo et al., 2021).

Previous studies have shown that persistent pollution was not only affected by local emissions and regional transport, but was also restricted by meteorological conditions. The variation of the PM2.5 and PM10 concentrations in the two types of persistent pollution were different (Zhou et al., 2019b). Song et al. (2019) showed the PM2.5 concentration significantly increased during stagnant atmospheric conditions. Jia et al. (2008) studied the variation of atmospheric aerosols in persistent pollution in Beijing, and discovered a growth mechanism characterized by an asymmetric “saw-tooth cycle” including a steady growth baseline in the mechanism, which is mainly based on an increase in secondary inorganic aerosols and an overlap with the daily cycle; this was controlled by the weather cycle. A similar phenomenon was found during Ep2 in this study. The increase in the PM2.5/PM10 ratio over time indicated an increase in fine particulates when the atmosphere was stagnant (low temperature, wind speed, and BLHmax (Fig. 6c, d); this mainly was because the increase in this ratio was promoted by the formation of secondary aerosols from high concentrations of precursors (SO2, NOx) through gas particles (Wu et al., 2019). The study of Yang et al. (2018a) also showed that increased PM2.5/PM10 ratio reflected a rapid secondary formation of particulates in persistent pollution. Because the synoptic circulation was the main factor involved in the diffusion and accumulation of pollutants, the durations of the sawtooth periods were controlled by synoptic cycles in Ep2. The explosive growth type had also obvious characteristics involving the accumulation of pollutants in stage 1 because of the stagnant atmospheric condition. However, a prominent feature was a sudden increase in PM10 and a sharp decrease in the PM2.5/PM10 ratio in stage 2. Lanzhou is located downwind of some major dust sources in China; this city frequently experiences dust storms in spring and winter. As a result, dust storms have a significant impact on the PM10 concentration (Feng & Wang, 2012). Previous studies in the Hexi Corridor area and the city of Beijing have also reported similar findings (Guan et al., 2017; Tao et al., 2012).

Many studies had pointed out that synoptic circulation and meteorological conditions are the main factors involved in heavy and persistent air pollution (Li et al., 2014; Xu et al., 2020). The typical synoptic conditions that contribute to persistent air pollution in China are weak weather systems, prevailing straight westerly airflow over North China, strong zonal circulation, and weak meridional circulation at 500 hPa (Wang et al., 2014). We also found that the 500 hPa synoptic circulation was mostly controlled by zonal circulation (such as straight westerly airflow and weak frontal systems) of the stagnant type and in stage 1 of the explosive growth type. This was consistent with the results from previous studies (Yang, 2018 b). It is well known that high-pressure systems lead to a widespread downward airflow. At the same time, taking Ep2 as an example, ground meteorological conditions featured low temperatures, wind speeds and BLHmax (Fig. 6c, d). These conditions inhibited the diffusion of air pollutants and enabled pollutants to gradually accumulate; thus, a stable synoptic circulation and adverse meteorological conditions provided a period of continuously stagnant weather with persistent pollution of the stagnant type in this study (Han et al., 2014). Moreover, the backward trajectory analysis showed that the airflow moved across the Huanghuang Valley area of Qinghai (Fig. 5c, d), which was already heavily burdened with PM2.5 pollution. Anthropogenic pollutants from this area aggravated the PM2.5 pollution are presented in Ep2. Dust events in spring and winter are usually associated with frontal and/or cyclonic systems in northern China (Yang et al., 2019b). Lanzhou lays on the southeastern edge of a low-pressure system during Ep1. Isopiestics were dense and pressure gradients were large, which guided the strong cold air system to Lanzhou. The backward trajectory also indicated that the airflow came from the northwesterly direction of Lanzhou. Moreover, the temperature and RH were remarkably lower than those during the previous stage. This typically indicated the arrival of cold air. A dust storm was also observed at the Zhongchuan Airport Meteorological station. Dust particulates are typically accompanied with the burst of severe cold air mass and lead to an abrupt increase in the PM10 concentration; thus, this was one of the main reasons why the PM10 concentration suddenly increased in the explosive growth type system.

In this study, cluster 1 from the west was the predominant airflow trajectories, followed by cluster 2 from the northwest. This may occur because Lanzhou is located on the northeast edge of the Qinghai-Tibetan Plateau, and the airflow was significantly affected by the topography of this Plateau. A few studies have pointed out the westerly airflow is the main airflow pattern in northwest China because of the obstruction of air masses by the Qinghai-Tibetan Plateau (Guan et al., 2019b; Manabe & Terpstra, 1974). Some studies have pointed out trajectories from various directions had distinct effects on the air quality at the point of reception because of the different pollution emissions in the pathway areas (Liu et al., 2016; Tang et al., 2016). The mean PM2.5 and PM10 concentrations in this study corresponding to various clusters were also significantly different. Clusters 1 and 2 originated from or passed through dust sources including the Kumtag and Qaidam desert as well as the Hexi Corridor. The patterns airflow easily allows air masses to carry large amounts of dust toward Lanzhou when passing through those areas, which gives rise to a significant increase in the PM10 concentration. Luo et al. (2020) reported that dust particles caused significant increases in the PM10 concentration in the area of the present study. Therefore, clusters 1 and 2 were the main transport pathways in Lanzhou for PM10. Clusters 1 and 2 passed through Hehuang Valley in Qinghai and Linxia Hui Autonomous Region in Gansu, respectively. Air masses in this area were already heavily burdened with PM2.5 pollution in the region surrounding Lanzhou (Fig. 5c, d). The air masses could easily take up anthropogenic pollutants from sources on the ground when passing through this region. Thus, clusters 1 and 2 were the predominant transport pathway for PM2.5. Cluster 3 was the northern airflow. The trajectories in cluster 3 may have been associated with deeper Mongolian cyclonic weather systems, resulting in strong wind. These winds facilitated the ventilation and vertical mixing of pollutants while bringing in fresh air from sparsely populated and less industrialized northern regions such as the Ningxia Hui Autonomous Region and north-central Gansu. Thus, the airflow trajectories associated with cluster 2 had less impact on PM10 and PM2.5 loadings in Lanzhou.

PM10 is emitted from natural sources such as dust storms, sea salt, volcanic eruptions, and biological particles (Guan et al., 2019a). The strongest potential sources of PM10 in this study were mainly located in the Kumtag Desert, the Qaidam Basin, and the western Hexi Corridor (Figs. 8a and 9a). These areas have arid climates, experience strong winds, and have aeolian land with low vegetation cover that are conducive to the formation of dust storms, making them important dust sources in China (Qian et al., 2004; Song et al., 2016). Lanzhou is located downwind from these areas (Fig. 1), so that the airflow passing though these areas easily picks up large amounts of dust particulates and carries them toward Lanzhou, which gives rise to significant increase in PM10 concentration. The results indicated that these regions were important potential source areas of PM10 in Lanzhou. Some studies have also found that these areas are important potential sources of PM10 for many cities in northwest China (Guan et al., 2019a; Yang et al., 2019a). Some scholars have pointed out that areas with limited plant cover including croplands, grasslands, and cropland mosaics are the main sources of anthropogenic dust (Guan et al., 2015; Chen et al., 2018). The present study found that the eastern part of the Hexi Corridor was also an important potential source area for PM10. This mainly occurs because that the region is an important agricultural area in northwest China (Figs. 8a and 9a). In winter and spring, when the farmland lays fallow or some crops are in the seedling stage, the ability of soil particles to consolidate is weak (Yan et al., 2009), and frequent windy weather and increased human activity can easily release dust particulates into the upper air stream (Goudie & Middleton, 2001; Hooper & Marx, 2018), resulting in the region becoming major potential sources of PM10 for Lanzhou (Yan et al., 2009). The PM2.5 pollution is mainly derived from anthropogenic emission sources including transportation, fuel burning, and industrial processes (Jain et al., 2020). Previous studies have reported the high PSCF and CWT values coincided with the locations of densely populated and industrial areas (Jeong et al., 2017; Saxena et al., 2017). The Hehuang Valley and Linxia Hui Autonomous Region are densely populated and economically developed areas with the areas surrounding Lanzhou (Wang, 2013). Frequent human activities usually release abundant particulate pollutants and their precursors, which causes serious PM2.5 pollution. According to the Environmental Bulletin of Qinghai Province in 2017, the mean annual concentrations of PM2.5 in the cities of Xining and Haidong, two important cities located in Hehuang valley, were 39 μg m−3 and 47 μg m−3, respectively, which exceeded the Chinese Ambient Air Quality Grade II annual standard (35 μg m−3). As a result, these make a strong contribution to PM2.5.

Conclusion

This paper analyzed the formation mechanism of persistent pollution, qualitatively identified major pollution transport pathways and potential source regions, and quantified the regionally transported contributions to PM2.5 and PM10 loadings in Lanzhou during the heating season from November 2016 to March 2017.

During the entire period, the daily mean concentrations of PM10 and PM2.5 reached 177 ± 142 g/m3 and 70 ± 32 μg/m3, respectively; the days over the Grade II standard accounted for 53% and 36% of all days, respectively, and particulate pollution was serious.

Five episodes with persistent air pollution were observed, which were classified into either the stagnant accumulation or explosive growth type based on the mechanism of pollution formation and evolution. Under the influence of stagnant weather, the concentration of PM2.5 and PM10 and PM2.5/PM10 ratio followed a growing “saw-tooth cycle” pattern during stagnant accumulation episodes. Dust storms created a prominent feature in which PM10 increased suddenly and the PM2.5/PM10 ratio decreased sharply in explosive growth type episodes.

Trajectories were grouped into five clusters. Cluster 1 came from the west and had the highest frequency (64.6%) of all of episode clusters, followed by cluster 4 from the northwest with 11.6%. Cluster 4 contributed the least (4.6%) to the total trajectories. The highest PM10 concentrations were observed in clusters 1 and 5 at 205 ± 176 μg/m3 and 148 ± 48 μg/m3, respectively. The highest PM2.5 concentrations were observed in clusters 1 and 4 at 78 ± 33 μg/m3 and 73 ± 27 μg/m3, respectively. The lowest PM10 and PM2.5 concentrations were found in cluster 2 at 105 ± 27 μg/m3 and 41 ± 33 μg/m3, respectively.

The potential sources of PM10 were mainly distributed in the Kumtag Desert in Xinjiang, the Qaidam Basin and Hehuang Valley in Qinghai, and the western and eastern Hexi Corridor in Gansu, where the contributions to PM10 were more than 120 μg/m3. The important potential sources of PM2.5 in Lanzhou were located in Hehuang Valley in Qinghai and Linxia Hui Autonomous Prefecture in Gansu; the contributions to PM2.5 from these areas were more than 60 μg/m3.