Analysis of the urban turbidity island effect: a case study in Beijing City, China

The urban turbidity island (UTI) effect is an important research topic in urban climate studies. It is closely related to urban visibility and the health of urban residents; however, it has received little attention in previous research. This study analyzes the temporal and spatial distribution characteristics of the UTI effect through the combined use of satellite remote sensing and ground observation data. Specifically, absolute and relative urban turbidity island intensity (UTII_A and UTII_R) indices are proposed and calculated for 2000–2020 by using aerosol data products and atmospheric fine particle mass concentration inversion products, which are represented by aerosol optical depth (AOD), PM1, PM2.5, and PM10. The results show the following: (a) there has been a clear footprint of the UTI effect in Beijing since 2000, generally consistent with trends of urban sprawl; (b) there are great differences in the interannual distribution of AOD, normalized AOD and PM values in urban and suburban areas; and (c) there are seasonal differences in the UTI distribution and air pollutant concentrations. The differences among indices between urban and suburban areas are mainly caused by heat island-induced air convection, complex structures in urban areas and regional weather conditions. Importantly, the interannual distribution of AOD and UTII_A of PM values decreased from 2014 to 2020, indicating that the government’s air pollution control policy has significantly improved air quality. Analysis from this study could support the formulation of urban planning and control policies to guide human activities.


Introduction
Approximately 45% to 60% of the world's population lives in urban areas (UNFP 1999, Arnfield 2003. The urbanization process has changed the characteristics of the original underlying surface of cities, and anthropogenic warming and pollutants have considerably increased due to the consumption of a large amount of energy resources (Gao et al 2004, Roger and Pielke 2005, Han et al 2019. These changes have impacted the near-stratum atmospheric structure, formed a dominant local climate effect in the city, and affected the surrounding climate (Arnfield 2003, Ulrickson 2009). Therefore, research on urban climate effects has practical application value for urban layout planning, human settlements, energy utilization and distribution and residents' health and is also of great significance for large-scale climate change planning and regulation (Kukla and Gavin 1986, Balling and Brazel 1987, Baik et al 2001, Seaman et al 2010. Urban economic activities and special underlying surface structures change the characteristics of the atmospheric boundary layer and have an impact on climate factors such as precipitation, temperature and radiation (Karl et al 1988, Cardelino andChameides 1990). The local urban climate can be roughly summarized by the urban 'five islands' effect, namely including heat island, dry island, wet island, rain island and turbidity island effects (Zhou 1988).
For example, the urban heat island effect describes the significantly higher near-ground temperature in the center of an urban area compared to that in its suburbs and surrounding villages (Zhou 1988, Liu et al 2006. In addition, the daily amplitude of absolute and relative humidity in urban areas is higher than that in suburban areas, forming a dry island effect (Oke et al 2017). Absolute and relative humidity in urban areas at night is higher than in the suburbs, creating a wet island effect (Ulpiani 2021). Rainstorms and waterlogging are more significant in urban areas than in suburban areas, resulting in the rain island effect. The emission of air pollutants in urban areas is higher than in surrounding areas, creating an urban pollution/haze or turbidity island (UTI) effect (Zhou et al 1991, Crutzen 2004, Ulpiani 2021. Urban heat island, dry island and wet island effects are typical results of urbanization and have been the focus of previous studies (Ulpiani 2021). For example, the urban heat island research center uses the directly related surface temperature to classify the heat island intensity (Shi et al 2009). Commonly used methods include the mean standard deviation division method, temperature normalization, temperature difference and temperature difference divided by the mean (Pan et al 2007, Zhu et al 2019. Zhou et al (1991) used the ratio of solar direct radiation (S) to diffuse radiation (D) as a measure of the turbidity factor (D/S) to study the turbidity island effect in the city of Shanghai. The authors found that the D/S factor in urban areas is obviously greater than that in suburban areas. Among urban climate studies, there have been few studies on UTI and haze effects (Zhu et al 2020). However, previous studies of the five-island effect provide important methodological reference and background information for the present study.
In most urban climate effect studies, researchers mainly used two methods to obtain research data: traditional weather stations or remote sensing methods. The amount of data obtained through ground meteorological stations is limited and discrete and cannot reflect the temporal and spatial variation characteristics of the climate effect in the whole city. Compared with the single-source data obtained by traditional meteorological stations, multisource remote sensing observations obtain a sufficiently large amount of data, such as advanced very highresolution radiometer, moderate resolution imaging spectrometer (MODIS) (King et al 1992), and other medium-resolution remote sensing data. These data feature long time series and wide coverage and are convenient for time series studies of large-scale urban climate effects . However, several shortcomings are inevitable, including but not limited to inconsistent spatial and temporal resolutions, contamination by clouds, and incomparable aerosol retrievals due to different kinds of biases.
A notable feature of the UTI effect is that air pollutant emissions are higher in urban areas than in surrounding areas. In recent years, urban air pollution has exhibited an increasing trend in some typical cities in China. Atmospheric monitoring has focused on extremely fine particulate matter (PM 1 ), fine particulate matter (PM 2.5 ) and inhalable particulate matter (PM 10 ) (Chang et al 2019, Ding 2019, Jiang et al 2020, Wei et al 2021a, 2021b. Excluding the impact of humidity on atmospheric visibility, the content of atmospheric pollutants determines the degree of atmospheric turbidity to a certain extent. Aerosol optical depth (AOD), i.e. the extinction coefficient integral of the atmosphere in the vertical direction (Kaufman and Joseph 1982), is a physical quantification of the degree of atmospheric turbidity, and it has been a popular research topic for satellite remote sensing inversion in recent years. Currently, there are many satellite AOD products (Koelemeijer et al 2006, Emili et al 2010, Zaman et al 2017, among which MODIS AOD products have a high resolution and degree of accuracy and are widely used in the inversion of ground PM 2.5 , PM 10 and other air pollutant concentration parameters (Wei et al 2021a(Wei et al , 2021b. The different representations of air pollutants and AOD in suburban areas can be used to analyze the UTI effect and explore the theoretical problem of how human activities affect the climate (Cheng et al 2017. With the rapid process of urbanization in China, many cities, including large, medium and small cities all over the country, have carried out urban climate research (Zhang et al 1991). Therefore, the objective of this study was to investigate the spatial distribution of the UTI effect in Beijing, one of the largest cities in China, using ground observation and remote sensing inversion data over the last 20 years. A schematic diagram of this study is shown in figure 1. Specifically, the urban-suburban boundaries of Beijing were derived from an impervious area dataset. Then, the urban turbidity island intensity (UTII) index was calculated from satellite-retrieved aerosol products. Then, several annual and monthly indices were calculated to analyze the temporal and spatial distribution characteristics of the UTI effect. Finally, the correlation between the UTII and its influencing factors was analyzed. The results from this study can provide guidance for planning and controlling human activities that affect the environment and climate in metropolitan areas.

Study area
The city of Beijing, located at 115.7 • E-117.4 • E, 39.4 • N-41.6 • N, is the political and cultural center of China (figure 2). This region has a large, highdensity population, a large proportion of its population lives in urban areas, and it has one of the highest immigration rates in China. By 2020, the city had 16 districts with a total area of 16 410.54 km 2 (BMBS 2021b). According to the data of the seventh census, the permanent resident population of the city was 21 893 095 (BMBS 2021a).
Located on the northwest edge of the North China Plain, this area has high terrain in the northwest (Yanshan and Taishan Mountains) and low terrain in the southeast. It has a typical warm temperate, semihumid continental monsoon climate, with an annual average temperature of 11 • C-13 • C. Spring is dry and windy, with a large temperature difference between day and night. Summer is hot and rainy. The average temperature in July in midsummer is close to 26 • C, and 70% of the annual precipitation is concentrated in summer. Autumn is sunny and pleasant with plenty of light, while winter is cold, dry and sunny and lasts up to five months (BMBS 2021a).
The process of urbanization in Beijing inevitably creates serious air pollution. Facing severe challenges from this air pollution, the government issued a new air quality standard ((MEEPRC 2016), hereafter the new standard) in 2012. Specifically, the monitored air pollutants include PM 2.5 , PM 10 , and an air quality index. As has been widely reported, in the winter of 2013, severe haze events frequently occurred in Beijing. The annual average PM 2.5 concentration reached 90 µm m −3 , which is nearly three times the threshold specified in the national standard (35 µm m −3 ) (MEEPRC 2014). Consequently, the municipal government formulated the Beijing Clean Air Action Plan 2013-2017 (hereinafter referred to as the plan) (PGBM 2017). The plan focused on the prevention and control of PM 2.5 pollution by adopting a series of treatment measures in key areas, such as coal reduction, mobile source emissions control, industrial pollution reduction and dust reduction.

Remote sensing datasets
In this study, MODIS Level 2 aerosol products (Optical_Depth_Land_and_Ocean) were selected (Levy et al 2013(Levy et al , 2015 based on the dark target algorithm and a specific look-up table strategy. Then, a pixel that has a minimum of 5 pixels over the ocean and 6 pixels over land in a retrieval box (6 × 6 retrieval boxes at a 0.5 km resolution) is deemed a 'good' pixel. Specifically, MYD04_3K and MOD04_3K from Terra and Aqua were used to derive AOD. These data were available from 24 February 2000 to 14 April 2021 at a spatial resolution of 3 km. Among the five bands in these data, the 0.55 band data were selected to characterize the UTI effect. In order to narrow the difference between MOD (data acquired from Terra) and MYD (data accquired from Aqua) products, we chose averaged value to conduct analysis in this study. To analyze temporal and spatial variation characteristics of the UTI effect, the averaged values of daily data were composited as monthly data, and then the averaged values of monthly data were composited as annual data.
Fine particle datasets of PM 1 , PM 2.5 and PM 10 retrieved by long-term MODIS AOD datasets (Wei et al , 2021a(Wei et al , 2021b were used to analyze the relationship between the UTI effect and its influencing factors. The inversion datasets utilized space-time extratrees models to reconstruct 1 km, hourly PM 1 , PM 2.5 and PM 10 concentration distributions in China for 2014-2018 (R 2 = 0.77), 2000-2018 (R 2 = 0.80-0.82), and 2015-2018 (R 2 = 0.86), respectively. The data coverage periods varied from 2014 to 2020 for PM 1 , from 2000 to 2018 for PM 2.5 , and from 2013 to 2020 for PM 10 . The three concentration distribution datasets of Beijing were extracted with the administrative boundary map. Specifically, the observations at 10-11 a.m. and 1-2 p.m. were averaged as the daily mean, and then the monthly mean data were obtained from the daily mean statistics.

Urban-suburban boundary extraction
Urban-suburban boundaries were prepared for subsequent analysis. The boundaries were extracted from a 30 m global artificial impermeable areas (GAIA) dataset from Landsat images via the Google Earth Engine platform (Gong et al 2019). Evaluations of the GAIA dataset across multiple years (e.g. 1985, 1990, 1995, 2000, 2005, 2010 and 2015) showed an overall average accuracy of greater than 90%. Specifically, the artificial impervious layers of the GAIA dataset in four typical years (2000/2005/2010/2015) were first extracted from the GAIA dataset. Then, the largest block of artificial impervious layers was considered the center of the city, from which the urban-suburban boundary was generated for each year. Finally, the boundary was overlaid with the urban boundary to obtain the suburban boundary. These boundary data were used for subsequent division of ground observations, extraction of remote sensing data and correlation analysis.

Definition of UTII
To analyze the UTI effect in Beijing, we defined the UTII in terms of absolute quantity and relative quantity.
The UTII of absolute quantity was calculated using the following formula: where UTII_A is the absolute value of UTII; PV Urban is the value of AOD from the MODIS; and fine particle datasets for measures such as PM 1 , PM 2.5 , and PM 10 were derived from Wei et al (2019), Wei et al (2021aWei et al ( , 2021b for urban areas. PV Suburban is the value in the suburban area. Therefore, UTII_A represents the absolute difference value of some pollutant between urban and suburban areas. Here, monthly and annual pixel value (PV) data were compiled from daily and monthly values. The UTII of relative quantity was calculated using the following formula: where UTII_R i represents the UTII_R value of the ith pixel of the image, and PV i is the ith pixel value. PV max and PV min are the 95th percentile maximum and 5th percentile minimum values among all values in PV. PV i values beyond the 95th and below the 5th percentiles are set to boundary values.
Then, the AOD value was normalized as the UTII_R index. This idea was inspired by urban heat island research (Zhou 1988, Cheng et al 2017, Zhu et al 2019. As AOD is a physical quantity representing the degree of atmospheric turbidity, it can be converted to UTII_R by abnormally removed normalization. Specifically, the UTII_R index was calculated using the following formula: where UTII_R AODi represents the UTII_R AOD value of the ith pixel of the image, and PV AODi is the ith pixel value in the previously described MOD04 AOD product (band 0.55). PV AODmax and PV AODmin are the 95th percentile maximum and 5th percentile minimum values among all values in PV AOD . These percentile metrics were used to avoid abnormal values. Similarly, PM 1, PM 2.5 and PM 10 data were normalized using the following formulas: where PM 1i , PM 2.5i , and PM 10i are the air pollution concentration values of the ith pixel of the image. PM 1min , PM 2.5min , and PM 10min are the 5th percentile minimum values of the year, while PM 1max , PM 2.5max , and are the 95th percentile maximum values of the year.
Then, the UTII_R AOD values were further divided into five classes to characterize the turbidity and cleanliness of the urban area (table 1). An equalspacing method was used to grade UTII_R AOD values, where pixels within the ranges of 0.8-1 and 0.6-0.8 were classified as high-and regular-turbidity island areas, pixels within a range of 0.4-0.6 were classified as nonturbid areas, and pixels within ranges of 0.2-0.4 and 0-0.2 were classified as relatively clean and clean island areas.

Statistical and spatiotemporal analyses
The spatiotemporal distribution characteristics of the UTI effect and its associated factors were analyzed. By drawing the monthly change curve and annual spatial distribution map of AOD and major pollutant concentrations such as PM 1 , PM 2.5 and PM 10 , the spatiotemporal distribution differences in the air pollutants in the urban and rural areas were analyzed. A thermodynamic diagram was then prepared to assist in the spatiotemporal analysis. According to the statistics of the UTII_R index calculated above, the monthly average pixel values of urban and suburban areas and the difference between the urban and suburban areas from 2000 to 2020 were illustrated as a thermal map. The spatiotemporal distribution of UTII_R and the differences between urban and rural areas could be analyzed through the distribution map. Moreover, the spatiotemporal distribution differences in the UTI effect and air pollutants in the urban and suburban areas were further analyzed.
Monthly variation curves and annual spatial distribution maps were prepared for the AOD and three types of fine particle matter (PM 1 , PM 2.5 , and PM 10 ).

Satellite remote sensing assists in the estimation of urban island turbidity intensity under rapid urban expansion
Based on the GAIA dataset, the urban-suburban boundaries of Beijing from 2000 to 2020 were extracted ( figure 3). Beijing is a typical rapidly expanding city, and accurate extraction of its urban-suburban boundaries is the basis of UTI effect analysis. In 2000, the urban area of Beijing was only 1008 km 2 . The urban area expanded by more than 70% in 2005 (1785 km 2 ), by almost 60% in 2010 (2821 km 2 ), and by 15% in 2015 (3238 km 2 ).
The annual average value of UTII_R AOD per pixel from 2000 to 2020 showed the footprint of the UTI effect ( figure 4). Similarly, annual average pollution concentration values were normalized for PM 1 from 2014 to 2020 (figure S1(a)), PM 2.5 from 2000 to 2018 (figure S1(b)) and PM 10 from 2013 to 2020 (figure S1(c)). There were spatial differences in the suburbs. The UTII_R AOD in red was greater than 0.6 and that in blue was less than 0.4. At the junctions of urban and suburban areas, there were obvious accumulations of red pixels. This showed that the UTI effect has existed for a long time since early 2000. However, the UTI effect enhancement area was not always consistent with the expansion direction of urban construction. Higher urban UTII_R AOD values were maintained as a belt surrounding the city center. This can be partially explained by heat island-induced air convection. This air convection provides sufficient moisture for smoke and dust in the air to condense and be transported to the urban area, resulting in increased UTII_R AOD values in the belt surrounding the city center . Additionally, a previous study found that topography may predominate the spatial pattern of AOD . Meanwhile, simulation results from the Weather Research and Forecasting Model coupled to Chemistry (WRF-Chem) indicate that topography exerts an important influence on urban haze pollution in downtown Beijing . Specifically, the resultant force of topography and valley wind circulation provides conditions for the accumulation of pollutants in urban areas (Oke et al 2017). This shows that the construction of a ventilation corridor system from the suburbs to the central urban area can blow away smog in areas with great ventilation potential. In addition, the difference in emission sources explains the UTI effect between urban and suburban areas. Nevertheless, the data period may have a partial impact on the analysis results. Only mid-day AOD data (10:30 a.m. and 1:30 p.m.) were obtained through MODIS. Thus, the derived UTII_R AOD made it difficult to consider the change in the daily trend.

Urban-suburban differences in UTII, AOD and PM trends
Annual statistics summarizing the AOD and normalized AOD and PM values in Beijing showed that there were great differences in their interannual distribution in urban and suburban areas (figure 5). From 2000 to 2020, both AOD and UTII_R AOD decreased in the suburbs but increased slightly in the urban area (figures 5(a) and (b)). However, the differences in both AOD and UTII_R AOD exhibited a significant upward trend (figures 5(f) and (g)), which indicated that UTII was aggravated in urban and suburban areas. Specifically, from 2015 to 2020, although the annual UTII_A AOD decreased significantly (figure 5(f)) in both urban and suburban areas, the annual UTII_R AOD difference, in contrast, exhibited a clear upward trend ( figure 5(g)).
Many studies have demonstrated that Beijing's air quality has improved because of powerful efforts against air pollution since 2015 (Dong et al 2020, Li et al 2021, which is consistent with our results. The trend over the urban-suburban difference in UTII_R AOD (figure 5(g)) was similar to that of AOD (figure 5(f)), but the trends over urban areas (red dashed line) in AOD (figure 5(a)) were the opposite to those of UTII_R AOD ( figure 5(b)). This result indicated that UTII_R AOD improved as a result of the treatment of air pollution and changes in absolute value pollutant concentration but was exacerbated in the urban−rural gap. Interestingly, changes in normalized PM 1 , PM 2.5 and PM 10 showed different trends (figures 5(c)-(e) and (h)-(j)). UTII_R PM1 exhibited a downward trend in urban areas, but the curve for suburban areas was flat (figure 5(c)). UTII_R PM10 increased in both urban and suburban areas (figure 5(e)). The difference between UTII_R PM1 and UTII_R PM10 in the urban and suburban areas decreased (figures 5(h) and (j)). In contrast, the difference in UTII_R PM2.5 between urban and suburban areas increased (figure 5(j)). The above findings are reasonable and reflect general trends in air quality and the complexity of AOD and UTII_R AOD . AOD is an absolute indicator of atmospheric cleanliness, reflecting the comprehensive degree of pollution. UTII_R AOD is a normalized index reflecting spatial heterogeneity within a city. However, as the components of PM include different pollutants, the pollution process may be influenced by different climate conditions, such as wind, sand and dust. Thus, there are contradictory trends between urban-suburban UTII_R PM2.5 , UTII_R PM1 and UTII_R PM10 .
Additional annual statistics for AOD, PM 1 , PM 2.5 and PM 10 pollutant concentrations show differences in their interannual distributions ( figure 6). UTII_A AOD exhibited a fluctuating upward trend  Concentrations of PM 2.5 fluctuated from 2000 to 2015 and decreased from 2015 to 2020 (figures 6(c) and (g)). The downward trend of PM 2.5 in urban areas was more obvious than that in suburban areas Figure 5. Annual variation trend of mean AOD and UTII_RAOD (left column, (a), (b)) and differences between urban and suburban areas (right column, (f), (g)) in Beijing from 2000 to 2020. UTII_RPM1, UTII_RPM2.5 and UTII_RPM10 (left column, (c)-(e)) and theirs differences between urban and suburban (right column, (h)-(j)).
The trends in AOD, UTII_A index and PM were related to the overall national policies and Beijing's environmental control measures. Since 2008, Beijing has promoted environmental governance on a large scale and in multiple ways based on the 'Green Olympics' concept. A 'Five-Year Plan' for ambient air control was formulated in Beijing after the well-known 'haze incident' in the winter of 2013. In Beijing, a series of measures to improve its air quality were taken, such as the optimization of the city's energy infrastructure, implementation of motor vehicle emissions policies, and upgrading of industrial structures (Li et al 2021). These policies and measures had substantial impacts on Beijing's environment and air quality. Therefore, the difference between urban and suburban areas in terms of the turbidity island effect intensified from 2000 to 2014 with the continuous expansion of the city and then decreased after 2015 due to stricter air quality policies.

Seasonal changes in turbidity island intensity and air pollutant concentrations
There were seasonal differences in the monthly means of AOD and UTII_R AOD in Beijing from 2000 to 2020 (figures 7 and 8). Overall, there were smaller changes in both AOD and UTII_R AOD over suburban areas than over urban areas (figures 7(a), (b), (d) and (e)), reflecting the influence of compact structures on air pollutant concentrations over urban areas (Oke et al 2017). The highest monthly values and differences between the urban and suburban areas (colored dark red in figures 7(c) and (f)) were in June, July and August. In urban areas, the changes in AOD and UTII_R AOD were seasonal (figures 7(a) and (d)). The turbidity degree in the suburban area changed synchronously, and the value in summer was higher than that in winter (colored blue in figures 7(b) and (e)). There were also significant differences in turbidity degree over 12 months in the suburban area; in June, the mean AOD and UTII_R AOD were noticeably higher. These seasonal patterns in the summer could be partially explained by the fact that cloud optical depth increased under high temperature and humidity conditions, which occurred frequently from June to September around the Beijing-Tianjin-Hebei region (Sun et al 2017).
In addition, there were different seasonal trends in AOD and three concentrations of air pollutants (PM 1 , PM 2.5 and PM 10 ) (figure 8). The changes in the monthly averaged AOD values in the urban and suburban areas were similar. The monthly average difference in UTII_R AOD in urban and suburban areas peaked from July to September (figures 8(a) and (f)). The most critical month in terms of turbidity was June, when the maximum UTII_R AOD values were observed in both the urban and suburban areas (0.62 and 0.45, respectively). The cleanest month was March, when the values of UTII_R AOD in both the urban and suburban areas reached their minimum (0.33 and 0.21, respectively). The maximum and minimum monthly average differences between the urban and suburban areas occurred in September (0.23) and February (0.12), respectively. In every month, the urban turbidity degree was higher than that in the suburbs. Moreover, there were different seasonal trends in concentrations of air pollutants such as PM 1 , PM 2.5 and PM 10 (figures 8(c)-(e) and (h)-(j)). Overall, the changes in the monthly averaged PM values in the urban and suburban areas were similar. PM 1 and PM 2.5 concentrations in urban and suburban areas were higher during the winter heating period and lower during the nonheating period (figures 8(c) and (d) On the other hand, the monthly averaged PM 10 concentration values had two peaks and two valleys (figures 8(e) and (j)). The two peaks were in December to January and in March to May, while the two valleys were in February and in June to September. The seasonal cycle of PM10 generally follows the winter heating period from late October to mid-April. However, there is a valley in February, possibly caused by the homecoming tide during the Spring Festival (normally from roughly late January to early February).
Further analysis of the monthly average proportion of turbid urban and clean suburban pixels showed that the UTI effect was stronger in urban areas than in suburban areas (figure 9). The annual monthly average proportion of turbid pixels in urban areas fluctuated from 20% to 70% from April to October ( figure 9(a)). The proportion of clean pixels in suburban areas in three-quarter months was greater than 60%, reaching the highest values (100%) in February, March, November and December ( figure 9(b)). This showed that in the period of weak air pollution, the air was cleaner in urban and suburban areas, and the UTI effect was weaker. These patterns were consistent with the urban UTII_R AOD distribution during summer and certain months in autumn (figure S2).

Conclusion
This study drew lessons from research on the urban heat island effect to monitor the turbidity island effect using the physical quantity of AOD, which reflects air turbidity, and analyzed the UTI effect using the typical metropolis of Beijing as an example. The UTI index was calculated by using remote sensing inversion data for more than 20 years. Then, the monthly average and annual average values of AOD, UTI and PM in suburban areas were statistically analyzed, and the temporal and spatial distribution characteristics of the turbidity island effect were explored.
The results showed that there was a strong UTI effect in Beijing, and its distribution varied in time and space. From 2000 to 2020, the temporal changes in AOD in the urban and suburban areas of Beijing tended to be consistent, but during those synchronous changes, the change in the urban areas was greater than that in the suburban areas. Second, the annual mean difference in UTI values between urban and suburban areas increased from 2000 to 2014, indicating that the UTI effect worsened during this period. The interannual distribution of AOD and UTII_A of PM values decreased from 2015 to 2020, reflecting the impact of the government's air pollution control policy since 2013. Third, the strongest UTI effect in terms of AOD occurred in the summer and peaked in July, while fine particle matter (i.e. PM 1 , PM 2.5 , and PM 10 ) levels were highest in the winter and peaked in December and January.
In conclusion, the UTI index reflected the comprehensive change in turbidity island intensity and pollutant components, while the change trends of metrics such as PM 2.5 and PM 1 /PM 10 were not completely consistent. The differences among indices between urban and suburban areas are mainly caused by heat island-induced air convection, complex structures in urban areas and regional weather conditions. Specifically, the annual variation trend of PM 2.5 differed from those of other pollutants (PM 1 /PM 10 ) and AOD. These findings indicated that the control of key pollutant components (such as PM 2.5 ) should be strengthened in the urban−rural synergy of air pollution control. Nevertheless, there is still uncertainty in the analysis above. Due to the limitation of the spatial resolution of MODIS data, the current study lacks detailed information to support the analysis of urban internal structure.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.