This study revealed the spatial migration process of PM2.5 by the concept and calculation method of the gravity center in physics. Moreover, the study also characterized the geographical center of PM2.5 pollution by the gravity model. The X and Y coordinates of the PM2.5 pollution center in China are defined as:

where X¯ is the longitude of the gravity of the PM2.5 pollution center; the latitude of the gravity of the PM2.5 pollution center is expressed by Y¯; the number of grids in the research scope is represented by n; the grid number is represented by i; X_i and Y_i represent the longitude and latitude of the geometric center of the grid whose number is i, respectively; S_i represents the area of the grid i; and W_i represents the AM of the PM2.5 of the grid i (μg/m³).

Spatial correlation refers to the correlation of the same variable at different locations (40). A variable in a certain position increases or decreases simultaneously with the same variable in its adjacent position, which is called spatial positive correlation. When one increases and one decreases, it is a spatial negative correlation. Spatial autocorrelation analysis describes the potential interdependence of some variables in adjacent spatial units and explores the spatial aggregation mode of factors, which can reveal the potential trend of environmental pollution in recent years (41, 42). According to the size of the analysis space, spatial autocorrelation can be divided into global spatial autocorrelation and local spatial autocorrelation (43), which are listed in the Supporting Information.

The change in PM2.5 may be related to industrial activities and human life. Considering the potential multicollinearity between these socio-economic factors, the variance inflation factor (VIF) is usually used for screening. Independent variables with VIF values of more than 10 should be removed before using GWR. The p-value is considered an indicator to determine the reliability of Moran's I. When the p-value of Moran's I is lower than 0.05, spatial econometrics and GWR can be applied. Otherwise, the application of spatial econometrics or GWR is unreasonable. Therefore, the GWR model is used to calculate the coefficients (R) and local regression coefficients for each city in the study area, and these coefficients are then mapped to show spatial variability.

The GWR model is shown as follows:

where y_i represents the concentration of air pollutants in city i, μg/m³; (u_i, v_i) represents the geographical location of city i; β₀(u_i, v_i) represents the concept of city i; β_k (u_i, v_i) represents the local regression coefficient of meteorological factor k; x_ik represents the value of the corresponding meteorological factor in city i; and ε_i represents the residual of city i. The local regression coefficient is calculated as follows:

where β(u_i, v_i) represents the local regression coefficients in cities; x represents the meteorological factor matrix; x^T is the transposition matrix of X; w(u_i, v_i) is the spatial weight matrix; y means the matrix of air pollutants; k means the number of meteorological factors; and n represents the number of samples.

According to the provisions of GB3095-2012 (Supplementary Table 1), limiting values for the daily average of PM2.5 concentrations are 15 for Level I and 35 for Level II, and limiting values for the hourly average are 35 for Level I and 75 for Level II. To analyze the pollution level clearly, the PM2.5 content was further divided into five intervals: 15–25, 25–35, 35–55, 55–75, and >75 μg/m³. The number of Chinese cities in different PM2.5 content intervals was counted, as shown in Supplementary Figure 1 and Supplementary Table 2. In 2018, the number of cities with an annual PM2.5 concentration in the range of 15–25 μg/m³ was the smallest, which was zero. As the years progressed, the number of cities in the range of the annual PM2.5 concentration gradually increased. The number of cities with a PM2.5 concentration in the range of 25–35 μg/m³ increased from 34 in 2018 to 86 in 2019 and 99 in 2020. The number of cities with a PM2.5 concentration in the range of 35–55 μg/m³ also showed an increasing trend. The average annual concentration of PM2.5 in the range of 55–75 μg/m³ showed a significant decreasing trend, from 103 in 2018 to 61 in 2019 and 14 in 2020. This indicated that the air quality was getting better year by year. In the range of the annual PM2.5 concentration above 75 μg/m³, the number of cities was seven in 2018, one in 2019, and two in 2020, indicating that the overall trend of pollution was decreasing. Through the above analysis, PM2.5 pollution has improved year by year from 2018 to 2020, thanks to policy support; policies have effectively promoted the improvement of air quality.

Figures 2A–D shows that PM2.5 pollution was more serious in the first and fourth quarters of 2018, followed by the second and third quarters. Areas with a PM2.5 concentration of more than 75 μg/m³ in the first quarter were mainly distributed in Xinjiang, such as Kashgar, Hotan, Turpan, and Urumqi, as well as Baoding, Weifang, Suihua, Shijiazhuang, and Zaozhuang. There are two main sources of PM2.5. In terms of climate, the rainfall in the first quarter was less than in other quarters. Thanks to warmer weather and increased rain, fine particles can be washed away, and the PM2.5 concentration then goes down in the second quarter. In the third quarter, only the PM2.5 concentration near Hotan in Xinjiang was >75 μg/m³, and the pollution in the Beijing–Tianjin–Hebei region was serious. In the fourth quarter, the weather turned cold, the northern region began to heat up, and the pollution in northern China was serious.

The geographical location of the PM2.5 pollution center in the four quarters of 2018 was statistically analyzed. It can be seen from Figure 2E that the pollution center in the first quarter was located at the boundary between Huanggang and Anqing. In the second quarter, it moved to the northeast and was located in Anqing. In the third quarter, it moved westward and was located in Huanggang. In the fourth quarter, it moved southeastward at a large distance and was located at the boundary between Jiujiang and Shangrao. From a macro point of view, the shift of the pollution gravity center from the first quarter to the fourth quarter was small, so it can be seen that the PM2.5 pollution in different quarters across the country did not show significant regional differences in terms of increases and decreases.

Human activity is one of the important factors affecting PM2.5 (45). Frequent human activities will increase the concentration of PM2.5 in the atmospheric environment. Population density refers to the number of people living in a unit area, which can accurately represent the population density of a region. Based on the data of the 2010 national population census, the distribution of national population density is shown in Figure 4A. From Figure 4A, it can be seen that the population density in China showed the law of “high in the east and low in the west.” The population density was taken as one of the influencing factors in this study. Compared with the hotspots in Figure 4B, different levels of regions were divided more scientifically.

The population density level could be divided into gathering area, transition area, less sparse area, and severely sparse area, and the corresponding population density of each area is 400, 200–400, 50–200, and <50 people/km². The hotspots with a population density of more than 400 people/km² were considered Area A, 200–400 people/km² were considered Area B, 50–200 people/km² were considered Area C, and the hotspots with a population density of fewer than 50 people/km² were considered Area D. The results are shown in Figure 4B. According to the distribution of monitoring sites, there were few monitoring sites in the western regions of China, and it is particularly necessary to set the classification of areas A, B, C, and D. Due to the small number of sites and the low density of the potentially exposed population in Area D, no further analysis was carried out. Due to the natural environmental conditions, such as wind and sand in Area D, the air conditions were poor. More afforestation should be carried out to increase the coverage of green plants and improve the state of pollution. For Areas A, B, and C, further driving factor analyses were carried out to develop targeted policies.

A total of 18 indicators were chosen as independent variables in this study. The average annual concentration of PM2.5 (μg/m³) was analyzed as a dependent variable. Before the GWR analysis, the VIF test should be preprocessed first to remove the indicators with VIF values of more than 10 to avoid collinearity affecting the experimental results. Then, the remaining usable indicators were the annual average population (10,000 people), urban construction land area (square kilometers), per capita GDP (Chinese yuan), GDP growth rate (%), the proportion of the primary industry in the GDP, the proportion of the secondary industry in the GDP, the employees (people) of the primary industry (agriculture, forestry, animal husbandry, and fishery), the employees (people) of the secondary industry, the employees (people) of the tertiary industry, the proportion of the employees of the primary industry, the proportion of the employees of the secondary industry, the proportion of the employees of the tertiary industry, the green space area (hectare), the ownership of civilian vehicles (10,000 vehicles), and the annual grain planting area (10,000 mu).

After excluding the local and global collinearity, the remaining seven factors were effective, which were the following: the annual average population (10,000 people), urban construction land area (square kilometers), the per capita GDP (Chinese yuan), the proportion of the secondary industry in the GDP, the green space area (hectares), the ownership of civilian vehicles (10,000 vehicles), and the annual grain planting area (10,000 mu). The spatial distribution of the regression coefficients of these factors is shown in Figure 5.

From Figure 5A, it can be seen that the annual grain planting area was negatively correlated with the PM2.5 concentration in northern Shanxi, the northern Beijing–Tianjin–Hebei region, eastern Shandong, and most of Jiangsu and Shanghai, and the rest was positively correlated. The correlation coefficients in northern Shanxi, Zhangjiakou, and Beijing were the smallest, and the correlation coefficients in southern Shaanxi and western Hubei were higher, showing an overall increase from the east to the west. The grain planting area in economically developed regions is relatively small. For example, as the political and economic center of the country, Beijing has developed rapidly. The planting area reserved for grain is relatively small, and the correlation coefficient was relatively small. The annual grain planting area was negatively correlated with the PM2.5 concentration, and the air pollution situation was still severe. In southern Shaanxi, for example, the annual grain planting area in Hanzhong in 2018 was 3.8097 million mu, and the grain planting area was relatively large. Straw burning occurred from time to time, which would cause serious air pollution. The annual grain planting area was positively correlated with PM2.5 concentration.

Figure 5B shows that in cities except for Jiangsu, southeast Anhui, and Shanghai, the number of civilian vehicles was positively correlated with PM2.5 concentration, and increased from the southeast to the northwest. Traffic is one of the major factors affecting air pollution. The larger the number of civilian vehicles, the more serious the pollution is. There was a relatively strong correlation between the ownership of civilian vehicles and PM2.5 concentration in most areas of Shaanxi Province. For example, Xi'an has carried out special rectification actions for motor vehicle exhaust pollution in recent years. Since 2016, the concentration of PM2.5 in Xi'an has decreased significantly in winter, and the effect of pollution control and haze reduction was obvious. However, heavy haze weather still occurs. The contribution rate of heavy trucks to PM2.5 pollution in Xi'an was 47.0%, and that of diesel vehicles was 80.2% (46). Therefore, Xi'an should focus on controlling heavy trucks and diesel vehicles to form a targeted management pattern.

As can be seen from Figure 5C, the urban construction land area was positively correlated in the west of the hotspot area and was negatively correlated in the rest of the area. Shaanxi and western Shanxi had the strongest positive correlation, indicating that the larger the urban construction land area, the higher the annual average PM2.5 concentration was. For example, Yulin is located in the Loess Plateau and is carrying out urban construction in a large area. The dust generated by a large number of construction projects will cause serious air pollution. The soil of the Loess Plateau is thick and loose, which aggravates the increase in PM2.5 concentration.

According to Figure 5D, the proportion of the secondary industry in the GDP was significantly positively correlated with PM2.5 concentration; it decreased from the west to the east, and the correlation coefficient was the highest in eastern Shandong. The second industry is a major force in environmental pollution. The higher the proportion of the second industry in the GDP, the more serious the air pollution is. Yantai is a traditional, strong industrial city, and its industrial output value ranks first in Shandong Province. Yantai vigorously develops the secondary industry, so that the pollution emissions are serious. Yantai should change its urban economic structure, reduce the proportion of the secondary industry, avoid the excessive agglomeration of the same type of secondary industry in the region, and improve the current situation of environmental pollution.

It can be seen from Figure 5E that the green area was negatively correlated with PM2.5 pollution. The high-value area was located in Shanxi and northwestern Shaanxi, and the low-value area was located in Enshi, Chengde, and Qinhuangdao. The larger the green area was, the more conducive it was to reducing PM2.5 pollution. Yan'an belongs to northern Shaanxi, and the green space area had a relatively obvious negative correlation with PM2.5 concentration. Green space helps to improve air pollution; therefore, green space can be further expanded in cities.

It can be seen from Figure 5F that the annual average population was positively correlated with PM2.5 concentration. The high-value area was located in Anhui, southeast Jiangsu, and Shanghai, and the low-value area was located in southern Shaanxi. The larger the annual population is, the more frequent human activities are, and this leads to more serious pollution and higher PM2.5 concentrations. In Shanghai, with a large population, the annual average population was positively correlated with PM2.5 concentration. It is necessary to promote a green low-carbon lifestyle and advocate for the participation of all people in haze prevention and control. In Shandong, Henan, and other populous provinces, the population should be reasonably controlled, and the population structure should be optimized.

It can be seen from Figure 5G that per capita GDP was negatively correlated with PM2.5 concentration. A region with a high per capita GDP has a higher level of economic development, a reasonable industrial structure, and advanced technology. Therefore, the region is more environmentally friendly and produces less pollution. The high-value area was located in the north-central Shaanxi, and the low-value area was located to the south of the hotspot. The per capita GDP of Suzhou was 173,765 Chinese yuan, and the level of economic development was relatively high. The average annual PM2.5 concentration in 2018 was 42 μg/m³. The per capita GDP was negatively correlated with PM2.5 concentration. With the rapid economic development of the city, the government should pay attention to the sustainable development of the environment and reduce air pollution.

It can be seen from the above analysis that the seven factors selected in this study had different effects on the average annual PM2.5 concentration. Comparing the average absolute value of the correlation coefficient, the influence degree of each factor on the average annual PM2.5 concentration decreased in the following order: the proportion of the secondary industry in the GDP, the ownership of civilian vehicles, the annual grain planting area, the annual average population, the urban construction land area, the green space area, and the per capita GDP. The proportion of the secondary industry in the GDP has a great influence on PM2.5 concentration. In some areas, the proportion of the secondary industry in the GDP is high, so pollution emissions are serious. It is necessary to accelerate the transformation of the urban economic structure and increase the proportion of the tertiary industry in the economic structure. At the same time, it is necessary to change the urban economic structure, reduce the proportion of the secondary industry, avoid the excessive aggregation of the same type of secondary industry in the region, and improve the current situation of environmental pollution. In Shaanxi Province, attention should be paid to air pollution in urban construction; strict regulations should be made on the air standards for construction, and a supervision system should be implemented to ensure that air pollution caused by construction is controlled within a reasonable range. At the same time, attention should be paid to air pollution caused by automobile exhaust emission pollution and crop planting. The number of cars should be further controlled, green travel should be promoted, penalties should be strengthened for vehicles with unqualified emissions, supervision should be strengthened in agriculture, and straw burning should be strictly prevented. Good economic conditions are the basis for effective environmental protection, so it is urgent to accelerate the high-quality growth of regional GDP. At the same time, expanding the green space in various regions helps reduce air pollution. Because of the differences in atmospheric resource endowments and the functional positioning of each region and its cities, it is recommended to establish a regional classification management system and ecological compensation strategy.