Circulation-regulated impacts of aerosol pollution on urban heat island in Beijing

. Unprecedented urbanization in China has led to serious urban heat island (UHI) issues, exerting intense heat stress 20 on urban residents. Based on observed temperature and PM 2.5 concentrations in Beijing over 2016-2020, we find diverse influences of aerosol pollution on urban heat island intensity (UHII) under different circulations. When northerly winds are prevalent in urban Beijing, UHII tends to be much higher in both daytime and nighttime and it is less affected by aerosol concentration. However, when southerly and westerly winds are dominant in rural Beijing, UHII is significantly reduced by aerosol pollution. Using coupled aerosol-radiation-weather simulations, we demonstrate the underlying physical mechanism, 25 which is associated with local circulation and resulting spatial distribution of aerosols. Our results also highlight the role of black carbon in aggravating UHI, especially during nighttime. It could thus be targeted for cooperative management of heat islands and aerosol pollution

and Grell 3D Ensemble Scheme (Grell, 1993) were employed to simulate aerosol-cloud interactions and precipitation. Sub-grid long-wave and short-wave radiation were calculated by RRTM (Mlawer et al., 1997) and Goddard 100 (Chou et al., 1998) schemes, respectively. Boundary layer processes were simulated with the Yonsei University planetary boundary layer parameterization (Noh et al., 2006). Noah land surface model (Tewari et al., 2004) was used to simulate landatmosphere exchange. We also included the Urban Canopy Model (UCM, Chen et al., 2011), which considers threedimensional structure of city and calculates energy balance on the surface of roof, wall and street.
The difference in heat storage is one important factor that affects the diurnal variation of UHII. In WRF-Chem model, 105 heat storage is calculated with land surface model, and we applied Noah land surface scheme for non-urban grids and Urban Canopy model for urban grids. In Noah land surface scheme, heat storage is calculated using the following equations: where denotes fractional vegetated area, and are heat storage for bare ground and vegetated ground, respectively, and +1 represents thermal conductivity of the surface layer of snow or soil; +1 is layer thickness of the surface layer of snow or soil, +1 represents temperature of the surface layer of snow (when +1<0) or soil (when = 0), and , and , stand for ground surface temperature at bare ground fraction and vegetated fraction, respectively. In Urban Canopy model, heat storage is calculated using 115 where G0 is the surface heat flux into the ground per unit area, including roof and road, and , , and are density, specific heat, and temperature of buildings.

Experimental design
We designed two groups of simulations of a severe haze event in the winter of 2010 (Case_2010) and a light pollution 120 event in the spring of 2018 (Case_2018). Each of them had four sets of simulations, namely AF, NAF, NBC and Ndust, to explore the impacts of aerosols on UHII, including roles of scattering and absorbing aerosols. AF cases were performed with actual conditions, while aerosol-radiation feedbacks were turned off in NAF. NBC was designed as the simulation that ignored the absorption of black carbon (BC) and absorption of dust was turned off in Ndust. In Case_2010, the simulation period covered from 11 to 20 January 2010 with first five days set up as spin-up time. The study period included three days 125 and nights from 8:00 LST on 16 January to 8:00 LST on 19 January 2010. It covered an entire severe haze pollution event in winter, during which UHI was formed and wind direction changed over days, offering conditions to analyze the impacts of ARF on UHII under different circulation conditions. In Case_2018, the simulation period covered from 19 to 28 April 2018 with first five days set up as spin-up time. The study period was from 7:00 LST on 24 April to 7:00 LST on 27 January 2010.
It covered a light aerosol pollution in spring and was used to evaluate if the impacts of aerosols on UHII are consistent under 130 different seasons and aerosol pollution conditions. As changes from turning off absorption of dust were negligible, we did not show results from Ndust in figures.

Calculation of UHII
We defined UHIIobs as observed differences in average 2m air temperature (T2m) between all urban stations and all rural stations. Following , we also calculated UHIImax and UHIImin as differences in daily maximum temperature 135 (Tmax) and daily minimum temperature (Tmin). As Tmax often appears in the afternoon and Tmin usually happens at late night or early morning before sunrise, we used UHIImax and UHIImin to refer to daytime and nighttime UHII. For simulated UHII, we defined UHIIsim as the difference in average T2m between urban areas and a buffer zone around the urban area that has the same size as the urban area, which is similar to that adopted with satellite products in Zhou et al. (2014). We chose these two different definitions of UHII for observation and simulation to evaluate uncertainty induced by the spatial limitation of 140 monitoring stations. K. It decreases to 1.8 K on pollution days (daily average PM2.5 concentration above 75 μg m -3 ) (Fig. 1a, d). UHII exhibits higher values at nighttime than those in daytime. In both daytime and nighttime, PM2.5 pollution is associated with decreased UHII in Beijing. In this analysis, we calculated mean PM2.5 concentration of all stations (Table S2) within Beijing and used it to determine if Beijing is polluted (daily mean PM2.5 concentration ≥ 75 μg m -3 ) or clean (daily mean PM2.5 concentration < 75 μg m -3 ). We also examined the distribution of daily mean urban and rural PM2.5 concentrations under clean and polluted 150 conditions (Table S3), and we found that 17.07 % of clean days of rural stations were classified as polluted due to the pollution gradient between urban and rural areas. However, these misclassified days were mostly light polluted with PM2.5 concentrations over 60 μg m -3 (Table S3).

Observational evidence of circulation-regulated impacts of aerosol pollution on UHII
We also evaluated how different standards of polluted or clean would affect results, and we included results based on the standard that PM2.5 concentrations of all stations in Beijing meet the thresholds of clean or polluted (Fig. S2) and results 155 based on the standard that both average PM2.5 concentrations of all urban stations and rural stations meet the criterion (Fig.   S3). Compared with Fig. 1 using mean PM2.5 concentration of all stations, we found similar distributions and negligible differences in mean values. When PM2.5 concentrations of all stations meet the criterion, we found the mean values increased by 0.03-0.04 K for clean conditions but decreased by 0.14 K during daytime and 0.06 K during nighttime. When we used average PM2.5 concentrations of all urban stations and rural stations to determine clean or polluted, mean values decreased by 6 0.01 K for clean conditions and increased by 0.01 K and 0.06 K during daytime and nighttime, respectively. We thus believe that using the daily mean PM2.5 concentrations averaged over all stations can properly represent the regional feature of aerosol pollution and would not affect our findings.
It was found previously that aerosol pollution led to decreased UHIImax (daytime) but increased UHIImin (nighttime) (Yang et al., 2021;. This discrepancy is associated with the differences in regions that considered as rural 165 in the calculation. We used rural stations located in the west and north of Beijing as rural in the calculation of UHII, and PM2.5 concentrations are usually much lower there. As a result, temperature at these rural stations is less affected by aerosol pollution. We designed a simplified flow chart to show how UHII is changed in daytime and nighttime, assuming that rural areas are not influenced by ARE (Fig. 2). ARE reduces near surface temperature in urban areas, leading to a weakened UHII and heat storage throughout the day. Although the strengthened longwave radiation process in nighttime that due to 170 absorption of aerosols in daytime alleviates the reduction of temperature in urban areas, decreased daytime temperature and heat storage release contributes more to near surface temperature and results in weakened UHII. The increase of UHII due to strengthened longwave radiation process is smaller than that by the process during daytime (see difference between Fig. 1b, c, e and f). in urban sites and rural sites. We observe elevated UHII when northerly winds are prevalent in urban areas on polluted days ( Fig. 3a, c). The mean UHIIs are 2.0 and 1.8 K in daytime and 2.9 and 2.8 K in nighttime on clean and polluted days, respectively. This is associated with reduced aerosol concentrations in urban regions by northerly winds in urban areas (Table 1). From clean to polluted conditions under northerly, lower reduction in UHII by aerosols is accordingly found (Fig.  180 3). Larger decreases in UHII in daytime can be found from clean to polluted conditions under easterly, southerly and westerly winds conditions, and these decreases are weakened at nighttime. The weakening may be caused by the longwave radiation process as absorptive aerosols release heat during night to alleviate decreases in surface temperature, especially in urban areas (Cao et al., 2016;. This process has also been confirmed with our simulation that AREinduced enhanced longwave radiation weakens UHII in nighttime (Fig. S4). 185 When sorted by wind directions in rural areas, we still find strongest UHII under northerly wind conditions (Fig. 3b, d).
However, UHII is relatively weak and the probability of "cold islands" in daytime increases when westerly or southerly winds are prevalent. The weak UHII under westerly wind condition is associated with foehn wind that northwesterly or westerly travel through the Yan Mountains, as foehn wind is able to heat rural areas and reduce the urban-rural thermal gradients (Ma et al., 2013). When southerly winds are prevalent, warmer southerly winds from lower latitude tend to heat 190 southern rural areas faster than urban due to blocking of air by buildings and larger heat capacities of urban impervious surface and buildings. We also detect larger reductions of UHII by aerosols when westerly or southerly winds are dominant ( Fig. 3b, d), suggesting that foehn wind and warm southerly winds are likely to amplify the weakening effect of aerosols on UHII.

Diurnal variations of the impacts of ARE on UHII 195
Although we identified above consistent weakening of UHII by aerosols during both daytime and nighttime, the influences vary with wind directions, which are regulated by background circulation patterns. To understand the underlying mechanism of the varying influences and to reduce uncertainty induced by selection of monitoring stations, we conducted model simulations of a typical haze event that occurred in winter in Beijing (Gao et al., 2016b) since aerosol concentrations are usually higher in winter in Beijing (Gao et al., 2018). We also designed simulations of a light pollution event in spring to 200 evaluate if the results are robust under different seasons and aerosol pollution conditions. As the aim of this section is to explore the underlying mechanism of interactions between aerosol pollution and urban heat island, although the period differs from observations shown above, the selected cases are sufficient to represent the observed varying wind conditions.
Model configurations in this study follow Gao et al. (2016b), and extensive model evaluations using multi-source observations indicated reliable reproduction of the wintertime haze event (Case_2010) by WRF-Chem. We additionally 205 evaluated the performance of WRF-Chem in simulating Case_2018 ( Fig. S6 and Table S4), and similar results were yielded.
Further validation of the ability of the model to simulate UHII is shown in Fig. S7 and Fig. S8. The model successfully reproduces the temporal variation of UHII in Beijing, and differences in values are generally within the trusted range, compared with previous simulations (Li and Bou-Zeid, 2013;Miao et al., 2009). To better clarify the influence induced by selection of rural areas, we added Fig. S9 to show the simulated UHII calculated based on site locations and area average. 210 Apparent difference can be found that site-based UHII decreases more than area-based UHII especially in nighttime because of lower PM2.5 concentrations in the rural sites than selected rural area. Fig. 4 shows the temporal variation of UHII of three cases, namely AF, NAF and NBC in Case_2010. Given the negligible contribution of absorption of dust to UHII, the results from the Ndust case are not shown. The impacts of ARE on UHII exhibit a bimodal distribution during daytime (Fig. 4a). The first peak and valley appear after sunrise, and the second 215 peak and valley occur before sunset. These variations are associated with the fact that changes in T2m occur earlier in rural areas. Aerosol pollution cuts down SWD in both urban and rural areas (Fig. S10a, b) after sunrise. Near surface temperature in rural areas usually increases faster than that of urban areas (Oke, 1982). As a result, temperature in rural areas exhibits earlier declines in response to ARE, as indicated by ARE induced changes in T2m in Fig. 4b, d. The second peak is caused by the similar reason that ARE results in the earlier decrease in T2m of rural areas (Fig. 4b), but the second peak is also 220 contributed by release of heat storage. Heat storage of rural areas is usually lower than that of urban areas, yet heat is released more slowly in rural areas, as suggested in Fig. S10 that heat storage is smaller in daytime but reaches zero earlier than it in urban areas (Fig. S10c, d). As a result, a faster declining of T2m in rural areas is found than that in urban areas (Fig.   4b, d). ARE reduces heat storage in both rural and urban areas, and the smaller heat storage and slower release of heat in rural areas is not sufficient to reverse the change rate of T2m, leading to the second peak and valley. Fig. 5 shows related 225 results for the Case_2018 case, and we find that ARE generally reduces UHII except on 26 April. The bimodal distribution during daytime still exists but is inconspicuous. This is because the lower ambient PM2.5 concentrations in the spring of 2018 reduce the gradient between urban and rural areas, and weakens the impacts of ARE on shortwave radiation and near-ground air temperature.

Diverse influences of ARE on UHII and the role of local circulation 230
We label days and nights of the study period as D1, N1, D2, N2, D3, N3 in Case_2010 and D4, N4, D5, N5, D6, N6 in Case_2018 in order, and find diverse influences under different wind patterns. On D1 and N1, we observe that ARE weakens UHII by 0 -0.4 K if the absorption of BC is not considered, due to larger amount of scattering aerosols in urban areas (Cao et al., 2016;. The weakening is larger at daytime and UHII is enhanced at nighttime when absorption of BC is considered (Fig. 4a). BC is potent in absorbing radiation, and it causes larger decrease in SWD at daytime. BC also 235 warms the atmosphere which increases downward longwave radiation ( Fig. S11 and Fig. S12) in nighttime (Cao et al., 2016;Zheng et al., 2018). On D2, a cold island with an intensity of ~-0.8 K is formed in Beijing, and ARE enhances the intensity of cold island. Due to the large reduction of UHII by aerosols at daytime (Fig. 4a), we still find negative effect of ARE on UHII on N2. Yet the negative effects weaken and become positive before sunrise. Different from previous two days, ARE enhances UHII with a maximum value of 1 K on D3. This is associated with reduced differences in PM2.5 concentrations 240 between urban and rural areas on D3 ( Fig. 4c and Fig. S13c). The conditions on N3 are similar with those on N2. The impacts of aerosols on UHII are mainly generated by modified downward longwave radiation in nighttime (Yang et al., 2021;Zheng et al., 2018), which influences the thermal difference of the atmosphere maintained after sunset. BC is the main light-absorbing aerosol (Gao et al., 2021;Ramanathan and Carmichael, 2008), and higher concentrations of BC (Fig. S11) lead to enhanced UHII in nighttime (Fig. S12). This explains the larger intensified UHII (~2 K) on N2. On D4 and D5, due 245 to much lower PM2.5 concentration, ARE reduces UHII by less than 0.2 K (Fig. 5). The lower concentration also diminishes absorption of shortwave radiation during daytime, which further reduces downward longwave radiation and causes weakened UHII on N4 and N6. On N5, we find a sudden ARE-induced increase in UHII (Fig. 5), and it is associated with the elevated PM2.5 concentration on N5 (Fig. 5c).
The above-mentioned diverse influences on different days of the study period are mainly controlled by local circulation. 250 Fig. 6 presents spatial distributions of daytime 2m air temperature and 10m wind fields over the study period. On D1 (Fig.   6a, d, g), southerly winds dominate the NCP, bringing warmer air to Beijing. However, due to relatively higher PM2.5 concentrations in the south of Beijing (Fig. S13), ARE decreases T2m as well as wind speeds. As a result, the warmer air transported from the southern regions to the south of Beijing is weakened, and only southern rural areas can be significantly heated, reducing the UHII of Beijing. This explains why UHII tends to be relatively weaker and larger reductions of UHII by 255 aerosols when southerly winds are prevalent in NCP (Fig. 3). On D2 (Fig. 6b, e, h), strong northwesterly winds (foehn wind) influence Beijing, and the entire western suburbs of Beijing heat up rapidly, forming a cold island. Meanwhile, mountains block strong northwesterly winds, and wind speeds on NCP are relatively weak, favoring accumulation of aerosols in urban areas (Fig. S13). Accordingly, ARE significantly reduces T2m in urban areas and further inhibits the UHII in the west of the city, consistent with the results shown in Fig. 3b that largest reductions in UHII was caused by aerosol pollution. On D3 9 (Fig. 6c, f, i), we detect a southeasterly sea breeze coming from the Bohai Gulf. Under the influence of the Yan Mountains, wind directions change to northeasterly when they reach Beijing. Consequently, more aerosols accumulate in the southern rural areas of Beijing (Fig. S13), ARE contributes to larger decrease in T2m in rural areas than that in urban areas. We thus observe an enhanced UHII caused by ARE on that day (Fig. 4a). This situation is consistent with observations that strongest UHII and alleviated reduction of UHII by aerosol pollution occur when urban areas are under northerly winds (Fig. 3a, c). 265 When light pollution event happens, similar responses (except results on N5) but smaller values are found (Fig. S14). The identified sudden ARE-induced increase in UHII on N5 (see Fig. 5) is caused by southerly winds. Southerly wind transports warm air masses with high PM2.5 concentrations from lower latitude to the north, and this process enhances the downward longwave radiation to heat the surface of urban and southern rural regions, resulting in enhanced UHII (Fig. S15). This also explains why UHII tends to decrease less when southerly winds are prevalent in nighttime (Fig. 3c). 270

Summary
Observed temperature and PM2.5 concentrations in Beijing over 2016-2020 suggest that aerosol pollution is associated with decreased UHII in Beijing at both daytime and nighttime, yet the influences of aerosol pollution on UHII are diverse under different circulation patterns. When northerly winds are prevalent in urban Beijing, UHII tends to be much higher at both daytime and nighttime and it is less affected by aerosol concentration. The mean values are 2.0 (1.8) and 2.9 (2.8) K in 275 clean (polluted) conditions in daytime and nighttime, respectively. However, when southerly and westerly winds are dominant in rural Beijing, UHII is significantly reduced by aerosol pollution by over 0.5 K. Using coupled aerosol-radiationweather simulations, we demonstrate the underlying physical mechanism, which is associated with local circulation and resulting spatial distribution of aerosols.
Previous studies documented opposite effects of aerosol pollution on UHII in Beijing (Cao et al., 2016;Yang et al., 280 2021;Yu et al., 2020;Zheng et al., 2018), and other cities Li et al., 2020a;Wu et al., 2017;Wu et al., 2019a). Our study highlights that the influences of aerosol pollution on UHII vary with local circulation, which is particularly important for Beijing due to the complex topography. Besides, heat can be modulated by local circulation to influence the impacts of aerosol pollution on UHII. Therefore, investigating the dominant synoptic pattens in certain areas may contribute to a better understanding of the aerosol-UHII interactions and provide guidance for mitigation strategies 285 Yu et al., 2020). Aerosol pollution in China has been alleviated significantly since the implementation of strict clean air policies after 2013 (Gao et al., 2020;Wang et al., 2020b). Yet there is still no evidence showing that it has cobenefits of reducing UHI (Li et al., 2007;Cao et al., 2016). It was found that decreasing aerosols led to intensification of urban warming and UHI, which further contributed to aggravation of ozone pollution (Wang et al., 2020b;Yu et al., 2020).
Thus, controlling aerosol pollution might even pose greater challenges for urban climate and environment management. In 290 this study, our model experiments emphasize the role of BC in aggravating UHI, especially during nighttime (Fig. 4). It could thus be targeted for cooperative management of heat islands and pollution. Some climate and environment friendly