Spatiotemporal evolutionary characteristics of PM2.5 . During 2000–2017, the overall PM2.5 level in the YREB showed a trend of first increasing and then decreasing (Fig. 4a). During 2005–2010, the annual PM2.5 concentration in the YREB was higher than the second-level standard of China of 35 µg/m3. During 2000–2007, the PM2.5 concentration increased from 23.49 µg/m3 to 37.37 µg/m3, an increase of 59.07%. Afterwards, it reduced to 31.79 µg/m3 in 2017, a decrease of 14.93%. It is evident that 2007 was a turning point towards improvement in terms of PM2.5. This improvement may be due to the effects of the national tenth five-year plan on controlling the total emission of major pollutants, adjusting industrial structure, and establishing a monitoring, statistics, and assessment system for energy conservation and pollution emission reduction. This was further reinforced by the implementation of the Action Plan for Air Pollution Prevention and Control in 2013 (http://www.mee.gov.cn).
Figure 4a shows that the sliding interval of the global Moran’s I value over the years was [0.825, 0.894], and all the values were significant at the 99% level. This indicates that the PM2.5 pollution distribution was not random, but it is a significant spatial agglomeration. We pursued the association of the evolution rules of spatial correlation with distance and found that the distance threshold to maintain spatial correlation was about 870 km (Fig. 4b). Within this spatial range, PM2.5 had significant positive interaction effects, which increased with the shortening of distance. Beyond 870 km, the global correlation changed from positive to negative, and the PM2.5 pollution among cities shifted from high–high clustering to low–high or high–low clustering.
Fig. 5 displays the spatial patterns and evolution of PM2.5 pollution in the YREB from 2000 to 2017. Its main features are as follows: (1) Cities with annual PM2.5 level of less than 15 μg/m3 were mainly concentrated in ethnic minority areas, where the ecological and environmental conditions were relatively good. However, air quality was continuously deteriorating albeit at low levels, which needs attention. (2) In the Yangtze Basin the PM2.5 level was higher in the lower reaches than the upper reaches and higher in the north bank than the south bank. The PM2.5 level shows a diagonal spatial distribution pattern with an obvious lowland plain directivity. (3) Urban economic activities and population density were closely related to PM2.5 pollution levels in three centres, namely, the Cheng-Yu economic zone of the upper reaches, the Wuhan metropolitan area of the middle reaches, and the northern Anhui–Jiangsu region of the lower reaches41.
Spatiotemporal variation characteristics of PM2.5 . We further explored the spatial heterogeneity evolution of PM2.5 pollution in the YREB by using the variogram geostatistical method (Table 2). The value of the variogram increased with the increase in separation distance indicating that the spatial autocorrelation of PM2.5 pollution changed from strong to weak with the increase of distance. During 2000–2017, the variation range of PM2.5 pollution was 625–738 km, and it showed an overall upward trend implying that the spatial correlation of PM2.5 pollution was partly expanded in scope. The higher the sill parameter is, the higher is the spatial heterogeneity. Therefore, the spatial heterogeneity of PM2.5 pollution was the lowest in 2017. In addition, the nugget effect indicated that regional scale influencing factors are more important for the distribution of PM2.5 pollution.
In terms of the fractal dimension (Table 3), the isotropic dimension continuously decreased from 1.536 in 2000 to 1.453 in 2017 indicating that the spatial difference of PM2.5 pollution was continuously expanding. The northeast–southwest direction had the greatest goodness of fit, the smallest fractal dimension, and showed a downward trend. This denoted that the spatial variation of PM2.5 pollution in this direction was continuously strengthened making it the main direction in terms of spatial difference. The southeast–northwest fractal dimension was the largest and its decisive coefficient kept decreasing showing that the spatial difference of PM2.5 pollution in this direction continued to weaken and remained relatively evenly balanced. The south–north fractal dimension first rose and then fell suggesting that the variability of the spatial pattern of PM2.5 pollution in this direction was enhanced. The change in the east–west fractal dimension indicated spatial heterogeneity of PM2.5 pollution in this direction and the degree of differentiation kept increasing with time. We conducted 3D-kriging interpolation, which further depicted the spatial distribution and evolution morphology of PM2.5 pollution (Fig. 6). It was evident that the overall PM2.5 level showed a distribution pattern with a higher value in the east and lower value in the west. Furthermore, PM2.5 pollution pattern was steadily transitioned from a gradient differentiation to a relatively balanced structure that formed a trend indicating that the middle–lower reaches of the Yangtze drove the whole basin PM2.5 level to increase.
Influence factor analysis models
GWR model results. GWR model fitting results are shown in Table 4, in which the adjusted R2 values were above 0.9 indicating a good fitting performance.
The regression coefficients of socioeconomic factors such as per capita GDP, population density, urbanization rate, and industrial structure were mainly positive. Among them, population density had the largest impact on PM2.5 and the most obvious spatial difference, followed by per capita GDP, while the coefficients of urbanization and industrial structure were relatively small. The regression coefficients of natural environmental factors such as wind, precipitation, vegetation, and topography were distributed in positive and negative intervals, and the instability was striking in different years. Among them, the coefficient of distribution interval for topographic relief was the longest indicating that spatial heterogeneity was the largest. While the coefficient of distribution interval for NDVI was shorter showing that the spatial heterogeneity was small.
Spatial heterogeneity of influencing factors. During 2000–2007, overall, the coefficient of socioeconomic factors increased signifying that extensive economic development and intensive human activities aggravated PM2.5 pollution. During this period, the coefficient declined but did not turn negative implying that the effects of economic development on environmental improvement did not appear yet. Existing studies have argued that economic development strongly correlates with regional haze pollution, in particular, the relationship between per capita GDP and PM2.5 level was significantly different in various regions19. Per capita GDP had positive impacts on PM2.5 in the middle–upper reaches of the Yangtze Basin (Fig. 7a, b, c). This implied that PM2.5 pollution in economically backward areas was more sensitive to economic development and economic growth of these regions came at the cost of the environment. Some cities in the YRD show negative correlation effects, which indicate that the development planning in the above areas was relatively good. With technological progress and industrial upgradation, economic development of these areas are essentially in coordination with the surroundings.
The coefficient of population density showed approximately the same spatial distribution at the three time nodes (Fig. 7d, e, f), all of which increased from the coastal areas to the inland areas, and among them the east–west difference was obvious in 2017. This may be because of the increase in urban traffic flow and production with a higher population density contributing to an increase in the local PM2.5 level. The control of air pollution in the middle–upper reaches of the Yangtze Basin was still weak, thereby, making the impact of population density more significant. Some studies have held that the increase of population density may have an agglomeration effect in promoting regional technical progress, and thus, facilitating the reduction of the local PM2.5 level14. The technical advances brought by the agglomeration effect for the population size in the YREB was not significant. This may be related to the migration of people to big cities in the recent years and the disordered nature of population mobility.
The coefficient of urbanization rate was positive at the three time nodes. The proportion of positive values in 2007 was relatively large, and the positive values in 2017 has marginally declined (Fig. 7g, h, i). During 2000–2007, areas with low urbanization rate in the Yangtze Basin were experiencing rapid urbanization and the urban infrastructure industry developed rapidly9. A large quantity of building dust entered the atmosphere aggravating urban PM2.5 pollution. However, areas with high urbanization rate, such as the YRD, tend to mature and a stagnant infrastructure industry is conducive for reduction in emission of fine particles.
The industrial structure had a negative impact on PM2.5 level in the middle–lower Yangtze plains and the Sichuan–Chongqing areas (Fig. 7j, k, l) aggravating local air pollution. This is consistent with findings of existing studies confirming that industrial activities were the main drivers of PM2.5 pollution in most areas19. The impact of the industrial structure in Sichuan–Chongqing areas was strong, which may be because of heavy industries such as energy, chemical, and machinery with relatively high direct energy consumption and pollutant emission. The optimization and adjustment of the industrial structure will significantly affect the local PM2.5 level in the air. The coefficient had a weak impact on PM2.5 in the YRD area, because the local industrial structure was dominated by the service industry and it was relatively stable, and thus, there was limited scope for further optimization.
The impact of intensity of wind speed on regional PM2.5 showed obvious east–west differences. The coefficients in the central and western areas were mainly positive and decreased from the central region to the west (Fig. 7m, n, o). The negative impact was dominant in the eastern areas, and as the distance to the coast reduced, the negative impact increased. This may be related to the impacts of topography and monsoon. Coastal areas were mostly alluvial plains with flat terrains, and they were affected by the local circulation caused by monsoon and temperature differences27. Clean air from the ocean had important dilution effects on pollutants, and thus, the coefficient was mainly negative. In the central and western areas, especially in the Sichuan Basin, the impact of closed topography restricted the diffusion of airflow and the transport of wind caused pollutants in the region to interact with each other, and thus, the coefficient was mainly positive. Similar findings were made in the Fenwei Plain, where a basin topography exists30.
The impact of precipitation on regional PM2.5 concentration presented a negative correlation at the three time nodes (Fig. 7p, q, r). A positive influence was mainly distributed in the Sichuan Basin and some areas of the Yunnan–Guizhou Plateau in western China, while the negative effect decreased from the coastal areas to inland areas. In terms of spatial heterogeneity, the regions with a high regression coefficient were mainly located in the upper and middle reaches of the Yangtze Basin, while the eastern coastal areas with abundant rainfall had a small regression coefficient indicating that abundant precipitation had a positive impact on PM2.5 pollution in most cities, and this effect was more distinct in areas with relatively deficient precipitation.
The impact of NDVI on regional PM2.5 concentration had a negative correlation that was mainly distributed in the middle–lower Yangtze plains (Fig. 7s, t, u). Research has shown that vegetation growth is correlated with climate and is affected by topography and human activities resulting in a complex correlation between vegetation and PM2.521. The spatial distribution of coefficients shows that the high-value areas were mainly in Yunnan,and Hunan provinces, while the low-value areas were concentrated in the middle–lower Yangtze plain areas. In 2017, the NDVI showed a negative impact indicating vegetation’s inhibitory effect on PM2.5 pollution was distinctly enhanced. This may be related to the rapid restoration and development of vegetation in the Yangtze Basin whose forest coverage rate has risen from 24% in 2000 to 39% in 20179. Thus, vegetation is an important inhibiting factor for PM2.5.
Topographic relief was negatively correlated with regional PM2.5 pollution in the Yangtze Basin, which was conducive to the improvement in air quality. From the spatial distribution of the coefficient (Fig. 7v, w, x), the high-value regions were located in Sichuan, Chongqing, and Zhejiang. The lower values were concentrated in the middle–lower reaches of the Yangtze. Local topography affects the diffusion and dilution of atmospheric pollutants in an area by influencing meteorological conditions20.
Interactions of influencing factors. Table 5 shows the results of the interactions of the influencing factors. The impact of the interactions between factors was greater than that of individual factors, and the interaction types included nonlinear enhancement and bi-factor enhancement. In terms of socioeconomic factors, when industrial structure interacted with per capita GDP and urbanization level on urban PM2.5 pollution, a nonlinear enhancement effect was exerted at all three time nodes, and the explanatory power was continuously improved. When annual average wind speed, annual precipitation, and vegetation coverage (natural factors) interacted with urban PM2.5 level in pairs, the nonlinear enhancement effect was generated at all three time nodes, and the explanatory power was notably varied in different periods. Nonlinear enhancement means that the interactive impact of two factors is greater than the sum of the impacts when they act alone. The interactive types of pgdp ∩ popd, pgdp ∩ urba, wind ∩ topo, prec ∩ topo, and ndvi ∩ topo were dominated by nonlinear enhancement though they were varied in different years. The interactive types of popd ∩ urba and popd ∩ indu exerted a bi-factor enhancement effect, which was not as significant as that of the nonlinear enhancement.