Impacts of urbanization on air quality and the related health risks in a city with complex terrain

. Urbanization affects air pollutants via urban expansion and emission growth, thereby inevitably changing the health risks involved with air pollutants. However, the health risks related to urbanization are rarely estimated, especially for cities with complex terrain. In this study, a highly urbanized city with severe air pollution and complex terrain (Chengdu) is selected to explore this issue. The effects of urban expansion are further compared with emission growth because air quality management is usually achieved by regulating anthropogenic emissions. Air pollution in Chengdu was mainly caused by PM 2 . 5 and O 3 from 2015 to 2021. PM 2 . 5 pollution tended to appear in cold months (November to February) owing to the blocking of air and the stable atmospheric layer, whereas O 3 pollution was likely to occur in warm months (April to August) that experience high-temperature and strong-sunlight conditions and are dominated by high-pressure systems. From 2015 to 2021, the 7-year annual


Introduction
Air pollutants are substances that damage humans, plants and animals drastically when present in the atmosphere at sufficient concentrations (Baklanov et al., 2016;Kinney, 2018;Pautasso et al., 2010).The most common air pollutants are ozone (O 3 ), fine particulate matter (PM 2.5 , particulate matter with an aerodynamic diameter of 2.5 µm or less), sulfur dioxide (SO 2 ) and nitrogen oxides (NO x , which is comprised of NO and NO 2 ).These air pollutants threaten human health in many parts of the world, evoking a series of health risks including cardiovascular diseases, respiratory diseases and chronic obstructive pulmonary disease (Brauer et al., 2016;Lelieveld et al., 2013;Manisalidis et al., 2020).According to the World Health Organization (WHO), exposure to ambient air pollutants is associated with 4.2 million premature deaths worldwide annually (https://www.who.int/health-topics/air-pollution#tab=tab_2, last access: 5 January 2023).
Most of those premature deaths occur in urban areas, as urban areas currently host more than 50 % of the population (over 3.5 billion people).This proportion is projected to increase to 70 % by 2050 due to ongoing urbanization (United Nations, 2018).Urbanization since the industrial revolution in the 19th century has led to a profound modification of land use via urban expansion (Seto et al., 2012).Natural surfaces are replaced by impervious surfaces, and the surface physical properties (e.g., albedo, thermal inertia and roughness) and processes (e.g., the exchange of water, momentum and energy) are then modified.These changes in the surface physical properties and processes exert an important influence on urban meteorology and air quality, which has been widely acknowledged in previous studies.Wang et al. (2009) explored the impacts of urban expansion on weather conditions as well as their implications for the O 3 concentration in the Pearl River Delta, and they pointed out that urban land use changes can cause a 1.0 %-3.7 % increase in the 2 m temperature, a 5.9 %-6.3 % increase in the planetary boundary layer height and a 4.2 %-8.5 % increase in the surface O 3 concentration.Liao et al. (2015) conducted a similar study in the Yangtze River Delta, and they found that urbanization increased the 2 m temperature, planetary boundary layer and surface O 3 concentration but decreased the surface PM 10 (particulate matter with an aerodynamic diameter of 10 µm or less) concentration.Similar conclusions about the impacts of urbanization on meteorology and air quality have also been reported in the Beijing-Tianjin-Hebei region (Yu et al., 2012) and the Sichuan Basin (H.Wang et al., 2021Wang et al., , 2022)).
Urban areas are centers of resource utilization and are a major contributor to air pollutant and greenhouse gas emissions (Karl et al., 2019;Qian et al., 2022).According to the UN-Habitat (https://unhabitat.org/topic/energy,last access: 5 January 2023), cities consume about 75 % of global primary energy and emit 50 %-60 % of the world's total greenhouse gases.Air pollutants that originate from anthropogenic sources can accumulate and degrade urban air quality under unfavorable meteorological conditions, characterized by weak winds, which leaves urban dwellers vulnerable to air pollution (Holman et al., 2015;Lin and Zhu, 2018).Excessive emissions are the root cause of poor air quality in urban areas; thus, efforts have been made to reduce anthropogenic emissions to achieve the goal of urban air pollution control.Urbanization can increase urban land use and anthropogenic emissions, which will affect the concentrations as well as health risks of air pollutants.However, the health risks related to urbanization are rarely estimated, especially for cities with complex terrain.This is of great concern to policymakers, and estimation of these heath risks could help inform future air quality control strategies.
Although building in the mountains is not as easy as building on plains, about 12 % of the global population (over 720 million people) resides in mountainous areas for historical, political, strategic and economic reasons.Thus, it is important to understand the fate of air pollutants in mountainous cities where air pollution is usually more severe than in flat locations, as atmospheric dispersion is limited (Zardi and Whiteman, 2013).The mountain-plain wind, resulting from horizontal temperature differences between air over mountain massifs and air over the surrounding plains, is a key feature of the climatology of mountainous regions (Whiteman, 2000) and is important for determining the transport and dispersion of air pollutants.During daytime, the plain-to-mountain wind (plain wind) brings low-level air into the mountain massifs, whereas the mountain-to-plain wind (mountain wind) brings air out of the mountain massifs during nighttime.This wind system can often recirculate urban air pollutants and worsen air quality.Examples of this process can be found in Mexico City (Molina et al., 2010), Hong Kong (Guo et al., 2013), Seoul (Ryu et al., 2013), the Salt Lake Valley (Baasandorj et al., 2017), the Colorado Front Range (Bahreini et al., 2018), the Alps (Karl et al., 2019) and Taiwan (Lee et al., 2019).
Chengdu (30.70 • N, 104.01 • E) is the largest city in western China, occupying an area of 12 390 km 2 with a population of more than 20 million people.Located in the west of the Sichuan Basin, this city is surrounded by the Tibetan Plateau to the west, the Wu Mountains to the east, the Yunnan-Guizhou Plateau to the south and the Daba Mountains to the north (Fig. 1a).Chengdu has experienced rapid urbanization over the past few decades that has been accompanied by a surge in urban construction and a loss of cropland (Fig. 1b; Dai et al., 2021).Luo et al. (2021) reported that Chengdu's urban area has increased 4-fold from 1996 to 2016.Because of the substantial anthropogenic emissions from human activities and the poor atmospheric diffusion ca- pacity associated with terrain, Chengdu is one of the most polluted cities in China and has suffered from severe PM 2.5 and O 3 pollution in recent years (Shu et al., 2021;Yang et al., 2020;Zhan et al., 2019).Complex terrain, rapid urbanization and severe air pollution make Chengdu an ideal place to study the impact of urbanization on the health risks related to air pollutants in mountainous areas.The results could also provide valuable insight for other cities with complex terrain in the world.
In this study, we investigate the impacts of urbanization on air pollutant concentrations and the corresponding health risks in Chengdu.We also compare the impacts of urban expansion with emission growth.First, the basic characteristics of air pollutants in Chengdu from 2015 to 2021 are analyzed.The impacts of urbanization on air pollutant concentrations are then investigated using the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem).Finally, premature mortality attributable to changes in air pollutant concentrations is estimated using the standard damage function.The rest of this paper is organized as follows: Sect. 2 introduces the data, the model configurations and the experimental design; Sect. 3 shows the main results and discussions; and the conclusions are given in Sect. 4.

Air quality and meteorological data
Air pollutants, including PM 2.5 , PM 10 , O 3 , NO 2 , SO 2 and CO, are monitored by the National Environmental Monitoring Center (NEMC) of China.These data are issued hourly on the national urban air quality real-time publishing platform (https://air.cnemc.cn:18007/,last access: 5 January 2023).The monitoring data are strictly in accordance with the national monitoring regulations (http://www.cnemc.cn/jcgf/dqhj/, last access: 5 January 2023).It should be noted that the O 3 measurements are reported in micrograms per cubic meter (µg m −3 ) under standard atmospheric conditions (273.15 K, 1 atm) before September 2018 and at 298.15 K conditions after this time.There are eight air quality stations throughout Chengdu (Fig. 1c), and the urban hourly pollutant concentrations reported in this paper are the average results of measurements at all monitoring sites.The daily PM 2.5 concentrations are obtained by averaging observations over 24 h of the day.The daily maximum 8 h average (MDA8) O 3 concentrations are calculated only on days with more than 18 h of O 3 measurements.
As shown in Fig. 1a, three nested domains are used with a grid spacing of 27, 9 and 3 km, respectively.A total of 32 σ levels extend from the surface to 100 hPa in the vertical direction, with 12 levels located below 2 km to resolve the boundary layer processes.The height of the lowest model level is about 25 m.The MODIS-based land use data, set as default in WRF, are selected.The domains and main options for physical and chemical parameterization schemes are listed in Table 1.The National Centers for Environmental Prediction (NCEP) Final (FNL) reanalysis data with a resolution of 1 • × 1 • at 6 h time intervals are adopted as the initial and boundary conditions for meteorological fields.Anthropogenic emissions are provided by the Multi-resolution Emission Inventory for China (MEIC) with a grid resolution of 0.25 • × 0.25 • .It should be noted that we empirically cut the PM 2.5 emissions by about 20 % to avoid overestimation of PM 2.5 in the model.Biogenic emissions are calculated online using the Guenther scheme (Guenther et al., 2006).
To estimate the impacts of urbanization, six numerical simulations are designed (Table 2).The year of the numerical simulations is 2017, as the MEIC emission inventory is currently updated to 2017.Considering the computational cost, January is deemed to be representative of cold months with frequent PM 2.5 pollution, whereas July is deemed to be representative of warm months with frequent O 3 pollution (Sect.3.1).Jan_Base is a baseline simulation using the MODIS land use data and the MEIC emission inventory over all three domains.The land cover maps in domain 3 are shown in Fig. 1b.Jan_noCD is a sensitivity simulation in which the urban land use of Chengdu is replaced by cropland to examine the impacts of urban expansion.Jan_noEmi is another sensitivity simulation in which the anthropogenic emissions in Chengdu are shut down to identify the impacts of emission growth.The abovementioned three numerical experiments use the same configurations (Table 1), running from 00:00 UTC on 28 December 2016 to 00:00 UTC on 1 February 2017 with the first 96 h as spin-up time.July_Base, July_noCD and July_noEmi are the same as Jan_Base, Jan_noCD and Jan_noEmi, respectively, but they run from 00:00 UTC on 27 June to 00:00 UTC on 1 August 2017 with the first 96 h as spin-up time.

Estimation of health risks
Daily premature mortality attributable to PM 2.5 and O 3 exposure from all non-accidental causes (ANACs), cardiovascular diseases (CVDs), respiratory diseases (RDs) and chronic obstructive pulmonary diseases (COPDs) is estimated using the standard damage function (Anenberg et al., 2010;Zhan et al., 2021): where M is the daily premature mortality, y 0 is the daily baseline mortality rate, RR is the relative risk, (RR − 1)/RR is the attributable fraction, and Pop is the exposed population.RR is calculated as follows: Mellor-Yamada-Janjić TKE scheme (Janjić, 1994) Cumulus parameterization Grell 3D ensemble scheme (Grell and Devenyi, 2002) Gas-phase chemistry RADM2 (Stockwell et al., 1990) Photolysis scheme Fast-J photolysis (Fast et al., 2006) Aerosol module MADE/SORGAM (Schell et al., 2001) Table 2. Six numerical simulations are conducted in this study.

Jan_Base Baseline simulation in January Jan_noCD
Replacing urban land use of Chengdu with cropland in January Jan_noEmi Shutting down anthropogenic emissions in Chengdu in January July_Base Baseline simulation in July July_noCD Replacing urban land use of Chengdu with cropland in July July_noEmi Shutting down anthropogenic emissions in Chengdu in July where β is the concentration-response function that relates a unit change in air pollutant concentrations to a change in the health endpoint incidence.In practice, β usually represents the percentage increase in daily mortality associated with a 10 µg m −3 increase in the daily PM 2.5 /MDA8 O 3 concentrations.C is the exposure concentration, which is the daily average concentration for PM 2.5 and the MDA8 O 3 concentration for O 3 .C 0 is the threshold concentration.When C is not greater than C 0 , the value of C − C 0 is 0.
In this study, C 0 is 10 µg m −3 for daily PM 2.5 (Song et al., 2015) and 75.2 µg m −3 for MDA8 O 3 (Liu et al., 2018).β and y 0 for ANACs, CVDs, RDs and COPDs are summarized in Table 3 (Chen et al., 2017;Yin et al., 2017).The population of Chengdu provided by the National Bureau of Statistics of China for the years from 2015 to 2021 are 16.853 million, 18.582 million, 19.188 million, 19.183 million, 20.409 million, 20.947 million and 20.938 million people, respectively.We first calculate the PM 2.5 -and O 3 -induced daily premature mortality using the methods mentioned above; we then sum the daily premature mortality for the whole year/month to get the total premature mortality.As the largest uncertainty among the factors that determine premature mortality usually comes from β, premature mortality is presented as the mean and 95 % confidence intervals (CIs) based on β at 95 % CI in this study.In addition, it should be noted that we use the average air pollutant concentration at all monitoring sites to represent the air pollutant concentration in Chengdu.Correspondingly, the total population of Chengdu is used as the exposed population.Thus, our results are for Chengdu as a whole and do not address the spatial distribution of premature mortality.

PM 2.5 and O 3 pollution in Chengdu
According to Chinese ambient air quality standards, PM 2.5 pollution occurs when daily PM 2.5 concentrations are greater than 75 µg m −3 , and O 3 pollution occurs when MDA8 O 3 concentrations are greater than 160 µg m −3 .As shown in Fig. 2, Chengdu has suffered from severe PM 2.5 and O 3 pollution in recent years: there were 97, 101, 68, 53, 33, 43 and 37 PM 2.5 pollution episodes and 61, 48, 42, 40, 42, 71 and 48 O 3 pollution episodes in Chengdu for the years from 2015 to 2021, respectively.In China, the annual evaluation criterion for PM 2.5 is the annual average concentration, whereas it is the 90th percentile of the MDA8 O 3 concentration for O 3 .The annual average concentrations of PM 2.5 were 60.7, 59.9, 52.6, 47.2, 40.6, 40.171.5, 188.9 and 167.1 µg m −3 , respectively.This suggests that PM 2.5 pollution improved significantly, whereas O 3 pollution did not.Thus, O 3 pollution control in Chengdu should be taken seriously in the future.In addition, PM 2.5 and O 3 pollution had clear seasonal trends: PM 2.5 pollution tended to appear in cold months (November to February), whereas O 3 pollution tended to appear in warm months (April to August).High PM 2.5 concentrations in cold months may be associated with the consumption of fossil fuels for heating as well as frequent temperature inversions, whereas high-temperature and strong-sunlight conditions contribute to the elevated O 3 concentrations in warm months.
3.3 Impacts of urbanization on PM 2.5 and O 3

Meteorological conditions in January and July
In this study, January and July 2017, when respective PM 2.5 and O 3 pollution episodes are likely to occur (Fig. 2), are selected to study the role of urbanization.In January 2017, Chengdu experienced PM 2.5 pollution for 23 out of 31 d, with  a monthly average concentration of 128.8 µg m −3 .From the perspective of atmospheric circulations, westerly winds prevailed over Chengdu due to the large north-south geopotential height gradient at 500 hPa (Fig. 4a).However, the westerly winds were blocked by the Tibetan Plateau and, thus, the dispersion of PM 2.5 was limited.At 700 hPa, the southwestern airflow originating from the Bay of Bengal could reach Chengdu (Fig. 4b).This warm advection was conducive to the formation of a stable layer near 700 hPa (Fig. 4c, d), which made the vertical diffusion of PM 2.5 difficult.The blocking of air and the stable layer were two important reasons for frequent PM 2.5 pollution episodes during this period (Hu and Wang, 2021;Ning et al., 2018).
In July 2017, there were 19 d of O 3 pollution in Chengdu, and the monthly average MDA8 O 3 concentration was 172.9 µg m −3 .At 500 hPa, Chengdu was dominated by strong high-pressure systems; thus, the air temperature was high and the wind speed was low (Fig. 5a).The monthly average T 2 was as high as 28.6 • C, while the monthly average WS 10 was only 1.6 m s −1 during this period (Fig. 6b).High temperature favored photochemical reactions of O 3 , while weak winds trapped O 3 .Furthermore, the thickness of the stable layer in July was far less than that in January (Figs. 4c,d;5c,d).A well-developed boundary layer facilitated the vertical mixing of O 3 within the boundary layer, which is an important way to maintain high surface O 3 concentrations during the daytime (Aneja et al., 2000;Tang et al., 2017).

Evaluation of model performance
We first compare vertical profiles in the model with the sounding data to determine whether the model captures the vertical structure of the troposphere.As shown in Fig. 4c and  d and in Fig. 5c and d, the WRF-Chem model can successfully simulate the changes in temperature and dew point temperature in the vertical direction (in both January and July as well as during both daytime and nighttime).Therefore, the vertical results from the model are reliable.Furthermore, simulated variables are compared to observed variables, and the results are presented in Fig. 6.The mean bias (MB) of the simulated and observed concentrations of PM 2.5 and O 3 are 12.7 and 11.6 µg m −3 , respectively, with normalized mean bias (NMB) values of 9.9 % and 12.0 %, respectively, which are within the acceptable standards (NMB < ± 15 %).The correlation coefficients (COR) of PM 2.5 and O 3 are 0.44 and 0.77, respectively.The statistical metrics for PM 2.5 and O 3 are similar to those from previous studies (Y.Wang et al., 2022;Wu et al., 2022), indicating that our model results for PM 2.5 and O 3 are reasonable and acceptable.With regard to the meteorological variables, T 2 is well simulated with low MB (0.2 and 0.1 • C, respectively) and high COR (0.76 and 0.70, respectively) values in both January and July.The simulations underestimate TD 2 to some extent, with MB values of −1.5 and −2.6 • C in January and July, respectively.As for the 10 m wind speed and direction, poor simulation results are predictable in the case of low wind and complex terrain.The observed calm wind frequency was particularly high due to the starting speed of the anemometer (typically 0.5-1 m s −1 ), resulting in an overestimation of simulated WS 10 , as in the studies of other scholars (Shu et al., 2021;Wu et al., 2022).With respect to this overestimation, it could also be argued that unresolved topographic features produce an additional drag to that generated by vegetation, but their effects are not considered in WRF (Jimenez and Dudhia, 2012).On the other hand, the model captures the shift in wind direction except for the case of calm wind.In summary, the WRF-Chem model using our configuration has a good capability with respect to simulating PM 2.5 , O 3 and meteorological variables in Chengdu; thus, the simulations can be used for subsequent analysis.

Spatiotemporal variations in PM 2.5 and O 3
The spatiotemporal characteristics of PM 2.5 were first investigated based on the Jan_Base simulation.PM 2.5 had a diurnal variation with a high concentration at night and a low concentration at noon, which was contrary to the boundary layer height (Fig. 7a).The nocturnal atmospheric boundary layer was often characterized by a stable boundary layer, and the boundary layer height was only ∼ 320 m above ground.As a consequence, PM 2.5 was trapped and maintained on the ground.The daytime atmospheric boundary layer, also known as the convective boundary layer, could develop to ∼ 1300 m above ground.Turbulence in the convective boundary layer could dilute PM 2.5 concentrations, resulting in low PM 2.5 concentrations at surface.Chengdu is on the eastern side of the Tibetan Plateau, with a large elevation drop exceeding 3000 m over a short horizontal distance (Fig. 1a).In this case, the mountain-plain wind can easily form.During nighttime, the mountain wind was characterized by westerly and downslope flow at lower levels along the eastern slope of the Tibetan Plateau (Fig. 7b, d  was likely to form and could spread hundreds of kilometers downstream.The daytime plain wind was nearly a reversal of the nighttime circulation, with easterly and upslope flow over the Sichuan Basin (Fig. 7c, e).The upslope flow could draw PM 2.5 to a higher elevation, which could also facilitate vertical dispersion of PM 2.5 during the day.
O 3 exhibited strong diurnal variation, with an afternoon maximum and an early-morning minimum (Fig. 8a).After sunrise, the nocturnal residual layer was destroyed while the convective boundary layer developed as the surface heated up on account of the incoming radiation.The highconcentration O 3 in the residual layer was then transported downward (Hu et al., 2018).Meanwhile, O 3 could be gen- erated by photochemical reactions between volatile organic compounds (VOCs) and NO x in the presence of sunlight.Through these two pathways, the surface O 3 concentration increased rapidly in the morning (Zhan and Xie, 2022).By noon, O 3 was mixed within the convective boundary layer via strong turbulence.Strong photochemical production and vertical mixing could maintain high surface O 3 concentrations until late afternoon.The daytime plain wind drove the westward transport of O 3 and aggravated O 3 pollution along the eastern slope of the Tibetan Plateau (Fig. 8c, e).After sunset, O 3 production ceased as the intensity of sunlight diminished.O 3 concentrations decreased substantially owing to surface deposition and nitrogen oxide titration (O 3 + NO → O 2 + NO 2 ) and then gradually reached their minimum in the early morning (Fig. 8b).However, O 3 in the nocturnal residual layer was still at a high level, with values of more than 160 µg m −3 .The nighttime mountain wind could carry O 3 -rich air eastward, and it enhanced O 3 concentrations aloft over the eastern slope of the Tibetan Plateau (Fig. 8d).Compared with the Jan_Base simulation, O 3 with a concentration of ∼ 100 µg m −3 had always existed over the Tibetan Plateau where PM 2.5 concentrations were quite low, indicating that the background concentration of O 3 was much higher than that of PM 2.5 .This can pose a huge challenge to O 3 pollution control in Chengdu.

Impacts of urban land use on PM 2.5 and O 3
Modification of urban land use changes surface dynamic and thermal characteristics, affecting the exchange of energy, moisture and momentum and hence altering urban meteorology and air quality.As illustrated in Fig. 9, surface PM 2.5 concentrations in the Jan_Base simulation were lower than those in the Jan_noCD simulation, with the monthly average concentrations reduced by 10.8 µg m −3 (7.6 %).Moreover, the decrease in PM 2.5 concentrations was larger during nighttime than during daytime.The monthly average PM 2.5 concentrations decreased by 13.9 µg m −3 (8.6 %) at 02:00 LST (LST is UTC+8h) but only by 3.0 µg m −3 (2.6 %) at 14:00 LST (Fig. 9a, b).The decrease in surface PM 2.5 concentrations was mainly attributed to the modification of the boundary layer height.Urban land use can enhance surface heating and then increases air temperature.The vertical air movement is then enhanced by the warming up of https://doi.org/10.5194/acp-23-771-2023Atmos.Chem.Phys., 23, 771-788, 2023 the air temperature, increasing the boundary layer height (Fig. S1), which facilitates the vertical diffusion of surface PM 2.5 .PM 2.5 concentrations increased by 2-6 µg m −3 in the upper boundary layer (∼ 1 km above ground) (Fig. 9c, d), further confirming this point.O 3 is a secondary air pollutant that is not only related to meteorological conditions but also to its precursors (VOCs and NO x ).Due to the increase in upward air movement and the boundary layer height induced by urban land use compared with cropland (Fig. S2), like PM 2.5 , NO x concentrations also decreased near the surface (Liao et al., 2015;Zhu et al., 2017).The decrease in NO x near the surface resulted in an increase in surface O 3 at night, as NO x titration was weakened (Fig. 10a, c).Although the elevated boundary layer diluted O 3 concentrations to some extent, the nighttime O 3 concentrations were mainly dominated by chemical effects and increased by 15.6 µg m −3 (16.0 %) at 02:00 LST (Fig. 10a).During daytime, the increased air temperature was conducive to the photochemical production of O 3 , and the well-developed convective boundary layer favored the vertical mixing of O 3 .O 3 concentrations also increased (Fig. 10b, d), with the monthly average value increas-ing by 5.4 µg m −3 (4.5 %) at 14:00 LST.As high O 3 concentrations were mainly concentrated in the afternoon, the monthly average MDA8 O 3 concentrations finally increased by 10.6 µg m −3 (6.0 %) due to the effects of urban expansion.

Impacts of anthropogenic emissions on PM 2.5 and O 3
Rising anthropogenic emissions of air pollutants and their precursors can significantly increase ambient air pollution.Therefore, the impacts of anthropogenic emissions are more intuitive than urban land use. Figure 11 shows the differences in PM 2.5 between the monthly average of the Jan_Base and Jan_noEmi simulations (Jan_Base minus Jan_noEmi).PM 2.5 concentrations in the Jan_Base simulation were significantly higher than those in the Jan_noEmi simulation, with the monthly average concentration enhanced by 23.9 µg m −3 (16.8 %), more than twice the difference between the Jan_Base and Jan_noCD simulations.Furthermore, increases in the PM 2.5 concentrations appeared throughout the boundary layer (Fig. 11c, d) and could extend downstream for hundreds of kilometers (Fig. 11a, b), indi- cating that reducing anthropogenic emissions is an effective way to reduce PM 2.5 concentrations.
As for O 3 , the monthly average O 3 concentrations in the July_Base simulation were only 1.6 µg m −3 (1.4 %) higher than those in the July_noEmis simulation at 14:00 LST (Fig. 12b, d), which was much smaller than the change in PM 2.5 .This phenomenon may be related to the nonlinear sensitivity of O 3 to VOCs and NO x precursor emissions.O 3 formation regimes can be classified into VOC-limited, NO x -limited and transition regimes depending on the ratio of VOCs to NO x (Jin et al., 2020;Lu et al., 2019).At low VOC/NO x ratios (VOC-limited regime), reducing the concentration of NO x would even lead to an increase in O 3 formation.Considering that Chengdu remained in a VOClimited regime during the period from 2013 to 2020 (Tan et al., 2018;Y. Wang et al., 2022), the effects of reducing NO x emissions may be partially offset by changes in VOCs; thus, a reasonable regulation framework that involves joint control of NO x and VOC emissions is necessary to alleviate O 3 pollution.Although the presence of anthropogenic emissions reduced the monthly average O 3 concentrations by 3.0 µg m −3 (3.1 %) at 2:00 LST, the monthly average MDA8 O 3 concentrations in the July_Base simulation were 4.8 µg m −3 (2.7 %) higher than those in the July_noEmis simulation.

Health risks caused by urbanization
According to the above results, urban land use decreased the monthly average PM 2.5 concentrations by 10.8 µg m −3 (7.6 %) but increased the monthly average MDA8 O 3 concentrations by 10.6 µg m −3 (6.0 %).On the other hand, anthropogenic emissions increased both PM 2.5 and MDA8 O 3 concentrations, with monthly average values of 23.9 µg m −3 (16.8 %) and 4.8 µg m −3 (2.7 %), respectively.We then calculate the changes in premature mortality under different simulation scenarios to assess the health risks related to changes in PM 2.5 and O 3 concentrations.As shown in Fig. 13, the premature mortality from ANACs, CVDs, RDs and COPDs due to PM 2.5 decreased by 171 (95 % CI of 129-200, or about 6.9 %), 45 (95 % CI of 34-53, or about 6.7 %), 22 (95 % CI of 16-27, or about 6.5 %) and 23 (95 % CI of 17-26, or about 6.2 %), respectively, in January 2017 when the Chengdu area was classified as urban land use rather than cropland.On the other hand, anthropogenic emissions in Chengdu increased premature mortality from ANACs, CVDs, RDs and COPDs due to  In summary, the total premature mortality due to PM 2.5 and O 3 changed by about −6.9 % and 9.5 % due to urban expansion, and these values changed by about 15.7 % and 4.1 % due to emissions growth.

Conclusions
With the development of urbanization, urban land use and anthropogenic emissions increase, thereby affecting urban air quality and, in turn, the health risks associated with air pollutants.In this study, the impacts of urban land use and anthropogenic emissions on air pollutant concentrations and the related health risks in Chengdu, a highly urbanized city with severe air pollution and complex terrain, are quantified.Management of urban air pollution is usually achieved by reducing anthropogenic emissions.Thus, the effects of urban expansion are further compared with those of emissions growth.
Chengdu has been suffering from severe PM 2.5 and O 3 pollution in recent years.During the years from 2015 to 2021, there were 97,101,68,53,33,43  pollution showed different seasonal trends: owing to the blocking of air and the stable atmospheric layer, PM 2.5 pollution tended to appear in cold months (November to February), whereas O 3 pollution was likely to occur in warm months (April to August) due to high-temperature and strong-sunlight conditions as well as the fact that these months are dominated by high-pressure systems.PM 2.5 concentrations were high at night and low at noon, which was contrary to the boundary layer height.O 3 exhibited strong diurnal variation, with an afternoon maximum and an earlymorning minimum, which was related to photochemical reactions during daytime and nitrogen oxide titration at night.
The urban land use of Chengdu was replaced by cropland in the WRF-Chem model to examine the impacts of urban expansion.Urban land use led to an increase in air temperature and the boundary layer height compared with cropland, and it decreased monthly averaged surface PM 2.5 concentrations by 10.8 µg m −3 (7.6 %).A higher temperature and boundary layer height increased O 3 concentrations via stronger photochemical reactions and better vertical mixing during daytime.During nighttime, dominated by the weakened chemical NO x titration, O 3 concentrations also increased.Finally, the monthly averaged MDA8 O 3 concentrations increased by 10.6 µg m −3 (6.0 %).In this case, when the Chengdu area was classified as urban land use rather than cropland, the premature mortality from ANACs due to PM 2.5 exposure decreased by 171 (95 % CI of 129-200, or about 6.9 %) but those due to O 3 increased by 203 (95 % CI of 122-268, or about 9.5 %).Anthropogenic emissions increased the surface PM 2.5 significantly, with the monthly average concentration increasing by 23.9 µg m −3 (16.8 %), more than twice the difference caused by urban land use.Owing to the nonlinear sensitivity of O 3 to its precursors, O 3 concentrations increased at noon but decreased at night.In particular, the monthly average O 3 concentrations increased by 1.6 µg m −3 (1.4 %) at 14:00 LST but decreased by 3.0 µg m −3 (3.1 %) at 2:00 LST.As O 3 concentrations during the daytime were much higher than those at night, the monthly average MDA8 O 3 concentrations still increased by 4.8 µg m −3 (2.7 %).As a consequence, the premature mortality from ANACs due to PM 2.5 increased by 388 (95 % CI of 291-456, or about 15.7 %) whereas that due to O 3 increased by 87 (95 % CI of 54-112, or about 4.1 %) with anthropogenic emissions in Chengdu.
Our results show that the impacts of urban expansion (about −6.9 % for PM 2.5 and about 9.5 % for O 3 ) are of the same order as those induced by emissions growth (about 15.7 % for PM 2.5 and about 4.1 % for O 3 ) on air pollutants.This suggests that, although the focus of air quality management is traditionally to regulate emissions, urban planning is an ancillary option and should also be considered in future air pollution strategies.
Data availability.Air quality monitoring data were acquired from the official NEMC real-time publishing platform (https://air.cnemc.cn:18007/, last access: 5 January 2023; National Environmental Monitoring Centre of China, Zhan and Xie, 2022).Meteorological data were obtained from the University of Wyoming website (http://weather.uwyo.edu/,last access: 5 January 2023; Department of Atmospheric Science, Zhan et al., 2020).The NCEP FNL data were taken from the NCEP (https://doi.org/10.5065/D6M043C6,National Centers for Environmental Prediction, 2000).The MEIC data can be accessed at http://meicmodel.org/ (Li et al., 2017;Zheng et al., 2018).These data can be downloaded for free as long as one agrees to the official instructions.
Author contributions.CZ and MX had the original idea for the study, designed the research, collected the data and prepared the original draft of the paper.CZ undertook the numerical simulations and carried out the data analysis.MX acquired financial support for the project leading to this publication.HL, BL and ZW collected the data.TW, BZ, ML and SL reviewed the initial draft and checked the language of the original draft.
Competing interests.The contact author has declared that none of the authors has any competing interests.
Disclaimer.Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
for the MEIC inventories.We acknowledge the High-Performance Computing Center of Nanjing University of Information Science and Technology for their support of this work.The authors also thank the anonymous reviewers for their constructive comments and suggestions.
Financial support.This work was supported by the National Natural Science Foundation of China (grant nos. 42275102, 42222503 and 42175098), the open research fund of Chongqing Meteorological Bureau (grant no.KFJJ-201607) and the Natural Science Foundation of Jiangsu Province (grant no.BK20211158).
Review statement.This paper was edited by Thomas Karl and reviewed by two anonymous referees.

Figure 1 .
Figure 1.(a) Map of three nested Weather Research and Forecasting (WRF) model coupled with Chemistry (WRF-Chem) domains with terrain heights, (b) domain 3 with land cover maps, and (c) the locations of air quality stations, meteorological stations and soundings in Chengdu.The red dot in panel (a) shows the location of Chengdu.

Figure 2 .
Figure 2. Heat maps of (a) daily PM 2.5 and (b) MDA8 O 3 concentrations in Chengdu from 2015 to 2021.

Figure 3 .
Figure 3. Premature mortality from ANACs, CVDs, RDs and COPDs attributable to (a) PM 2.5 and (b) O 3 in Chengdu from 2015 to 2021.The dots represent the mean estimate, and the whiskers represent 95 % CIs.

Figure 4 .
Figure 4.The weather charts at (a) 500 hPa and (b) 700 hPa for January 2017 are based on the NCEP FNL reanalysis data.The purple stars show the location of Chengdu.Panels (c) and (d) show the skew-T diagrams at 00:00 UTC and 12:00 UTC, respectively, in January 2017.The red and blue solid lines are the respective simulated air temperature and dew point temperature in the Jan_Base simulation, and the red and blue dashed lines are the respective sounding temperature and dew point temperature.These results are monthly averages.

Figure 5 .
Figure 5.The weather charts at (a) 500 hPa and (b) 700 hPa for July 2017 are based on the NCEP FNL reanalysis data.The purple stars show the location of Chengdu.Panels (c) and (d) show the skew-T diagrams at 00:00 UTC and 12:00 UTC, respectively, in July 2017.The red and blue solid lines are the respective simulated air temperature and dew point temperature in the July_Base simulation, and the red and blue dashed lines are the respective sounding temperature and dew point temperature.These results are monthly averages.

Figure 6 .
Figure 6.(a) Times series of PM 2.5 , T 2 , TD 2 , WS 10 and WD 10 for January 2017.(b) Times series of O 3 , T 2 , TD 2 and WS 10 and WD 10 for July 2017.The black dots are observations.The colored lines and cyan dots are simulated values from the baseline simulations.

Figure 7 .
Figure 7. (a) Time-altitude cross sections of PM 2.5 (colored shading), potential temperature (purple contour lines) and boundary layer height (thick black contour line) at Chengdu.Panels (b) and (c) present the horizontal distributions of PM 2.5 with wind vectors at the lowest model level at 02:00 and 14:00 LST, respectively.Panels (d) and (e) present east-west vertical cross sections of PM 2.5 with wind vectors at 02:00 and 14:00 LST, respectively.Purple stars show the locations of Chengdu.Brown shaded areas represent the terrain.These results are the monthly average based on the Jan_Base simulation.

Figure 8 .
Figure 8.(a) Time-altitude cross sections of O 3 (colored shading), potential temperature (purple contour lines) and boundary layer height (thick black contour lines) at Chengdu.Panels (b) and (c) present the horizontal distributions of O 3 with wind vectors at the lowest model level at 02:00 and 14:00 LST, respectively.Panels (d) and (e) present east-west vertical cross sections of O 3 with wind vectors at 02:00 and 14:00 LST, respectively.Purple stars show the locations of Chengdu.Brown shaded areas represent the terrain.These results are the monthly average based on the July_Base simulation.

Figure 9 .
Figure 9. Panels (a) and (b) present the horizontal distributions of the differences in PM 2.5 at the lowest model level at 02:00 and 14:00 LST, respectively.Panels (c) and (d) present the east-west vertical cross sections of the difference in PM 2.5 at 02:00 and 14:00 LST, respectively.Purple stars show the location of Chengdu.Brown shaded areas represent the terrain.These results are the difference between the monthly average of the Jan_Base and Jan_noCD simulations (Jan_Base minus Jan_noCD).

Figure 10 .
Figure 10.Panels (a) and (b) present the horizontal distributions of the differences in O 3 at the lowest model level at 02:00 and 14:00 LST, respectively.Panels (c) and (d) present the east-west vertical cross sections of the difference in O 3 at 02:00 and 14:00 LST, respectively.Purple stars show the location of Chengdu.Brown shaded areas represent the terrain.These results are the difference between the monthly average of the July_Base and July_noCD simulations (July_Base minus July_noCD).

Figure 11 .
Figure11.Same as Fig.9but for the difference between the monthly average of the Jan_Base and Jan_noEmi simulations (Jan_Base minus Jan_noEmi).

Figure 12 .
Figure 12.Same as Fig. 10 but for the difference between the monthly average of the July_Base and July_noEmi simulations (July_Base minus July_noEmi).

Figure 13 .
Figure 13.Differences in premature mortality from ANACs, CVDs, RDs and COPDs due to PM 2.5 (left of the dotted line) and O 3 (right of the dotted line) between the baseline and sensitivity simulations.The dots represent the mean estimate, and the whiskers represent 95 % CIs.
To verify upper-air fields, the sounding observa-These sounding data contain information, such as temperature, dew point temperature and wind speed, at different pressure layers with a time resolution of 12 h (00:00 and 12:00 UTC), and they are often plotted on a Skew-T diagram (https://www.ncl.ucar.edu/Applications/skewt.shtml# ex2, last access: 5 January 2023).

Table 1 .
The domains and main options for WRF-Chem.