Co-benefits of reducing PM2.5 and improving visibility by COVID-19 lockdown in Wuhan

The less improvement of ambient visibility suspects the government’s efforts on alleviating PM2.5 pollution. The COVID-19 lockdown reduced PM2.5 and increased visibility in Wuhan. Compared to pre-lockdown period, the PM2.5 concentration decreased by 39.0 μg m−3, dominated by NH4NO3 mass reduction (24.8 μg m−3) during lockdown period. The PM2.5 threshold corresponding to visibility of 10 km (PTV10) varied in 54–175 μg m−3 and an hourly PM2.5 of 54 μg m−3 was recommended to prevent haze occurrence. The lockdown measures elevated PTV10 by 9–58 μg m−3 as the decreases in PM2.5 mass scattering efficiency and optical hygroscopicity. The visibility increased by 107%, resulted from NH4NO3 extinction reduction. The NH4NO3 mass reduction weakened its mutual promotion with aerosol water and increased PM2.5 deliquescence humidity. Controlling TNO3 (HNO3 + NO3−) was more effective to reduce PM2.5 and improve visibility than NHx (NH3 + NH4+) unless the NHx reduction exceeded 11.7–17.5 μg m−3.


INTRODUCTION
Atmospheric visibility provides intuitive grasp of air quality for public 1 . China has suffered substantial visibility deterioration in the past years [2][3][4][5] , which adversely impacts traffic 6 and human happiness 7 . Intensive occurrences of haze with low visibility have raised public awareness [8][9][10][11][12] . Since the promulgation of Air Pollution Prevention Control and Action Plan in 2013 13 , the national emissions of SO 2 , NO x , and primary fine particle (PM 2.5 ) declined by 59, 21 and 33%, respectively [14][15][16] . The PM 2.5 mass concentrations in Beijing-Tianjin-Hebei, Yangtze River Delta, and Pearl River Delta reduced by 28-40% during 2013-2017 17 . However, such great mitigations in air pollution are not directly visible to the general population because the ambient visibility seems less improved, especially in winter 11,12,14,16,18,19 . For example, the frequency of low visibility events only decreased by 5% despite the reduction in PM 2.5 > 30% in 2018 in Southern China when compared with 2013 19 . This depressing visibility improvement is also found in Eastern China even though PM 2.5 has lowered by 50.8 μg m −3 from 2013 to 2018 16 . The annual average visibilities for Fenwei Plain and Central China are still < 10 km, and the haze days are still > 71 days 20 . All these mask the intense and painstaking efforts that the government devoted for defending the blue sky.
Temporary emission control measures have been frequently implemented during mega-event periods, aimed at reducing the mass concentrations of PM 2.5 , SO 2 , NO x , and O 3 24,37,38 . The mass concentrations of PM 2.5 decreased obviously by 40-49% in these events 37,38 , which actuated the appearances of blue sky 37,39 . Li et al. 40 observed that the frequency of hazy days decreased to 36% during the Beijing Olympics because the b ext contributions of (NH 4 ) 2 SO 4 and NH 4 NO 3 decreased by 17.1 and 13%, respectively. Tao et al. 41 found that the "APEC blue" was mainly raised by SNA extinction reduction (30%), with 5 days holding ambient visibility > 20 km. However, the frequency of haze occurrence increased by 7% during the 16th Asian Games period though PM 2.5 decreased by 32% 42 . The unexpected haze during the former mega-event periods implies that the temporal pollution control measures for air pollutants at a regional scale may be not effective for improving the ambient visibility necessarily.
The lockdown events due to coronavirus disease 2019 (COVID-19) provide the widest, longest, and thorough "controlled experiment," to investigate the impacts of unexpected control measures on reducing air pollutant concentrations and improving visibility. Abundant studies have reported large decreases in CO 2 43-45 , NO x 46-48 , particulate matter [49][50][51] , and associated chemical components [52][53][54][55][56] . While the reduced anthropogenic emissions did not hold back the occurrences of severe haze events in China because of unfavorable meteorology 57,58 , enhanced secondary formation 18,59 , and regional transport 60,61 . In addition, the aerosol optical depth (AOD) was less affected by the reductions 47,57 . In fact, the impacts of the strictest lockdown measures on aerosol optical properties are puzzled and till now no studies have focused on this point.
Wuhan is the first locked city and the lockdown measures are the strictest. This study analyzed the online monitoring datasets including b sp , b ap , and chemical components for PM 2.5 in Wuhan for pre-lockdown period (PLP) and lockdown period (LP). The impacts of chemical compositions and hygroscopic growth on b ext and corresponding mechanisms were investigated, and the key chemical component impacting b ext was identified. Priority policies for reducing PM 2.5 and improving ambient visibility effectively were proposed. Results here can provide a reference for policy making from the view of improving ambient visibility.  Fig. 1a), since the lockdown measures actually reduced the anthropogenic emissions 45,59,61 . The decrements in the average mass concentrations of major compounds varied from 0.8 μg m −3 (elemental carbon (EC)) to 24.8 μg m −3 (NH 4 NO 3 ) except for secondary organic aerosol (SOA) (Fig. 1b). The decrease in NH 4 NO 3 made up 63.6% of PM 2.5 mass reduction. The average SOA concentration showed an increase of 1.6 μg m −3 and its mass percentage in PM 2.5 raised by 6.9%, verifying the enhanced secondary formation 18 .

RESULTS AND DISCUSSION
The mean b sp and b ap decreased by 39.0% (151.2 Mm −1 ) and 31.4% (8.9 Mm −1 ) during LP, respectively (Fig. 1a). The single scattering albedo decreased only by 1.1% during LP (0.91), implying the strong scattering ability of particle. The RH slightly descended by 8.8% from PLP (78.4 ± 13.8%) to LP (71.4 ± 15.7%). The visibility displayed a remarkable (p < 0.01) increase of 106.7% (14.4 km) during LP, demonstrating that the strict control measures were effective to improve the ambient visibility along with the decrease of PM 2.5 . While in Eastern 16 and Southern China 19 the PM 2.5 substantially decreased, the ambient visibility was less improved due to its nonlinear relationship with PM 2. 5 16,19,22,28 . During LP, severe haze events with ambient visibility < 10 km occurred on 3 and 5 February with the maximal b sp and b ap as 689.0 and 53.7 Mm −1 , respectively. The SNA contributed highest to PM 2.5 mass (66.7-67.4%) and b ext (56.0-59.8%) ( Supplementary  Fig. 1). The haze events on 3 and 5 February were related with local accumulation and regional transport of air pollutants, respectively 60 ( Supplementary Fig. 1), while the dominant chemical species of the 2 days were similar under the two different atmospheric circulations, highlighting the substantial role in improving ambient visibility by regional-joint control of the precursor gas emissions for SNA. Figure 2a shows the non-linear responses of visibility to PM 2.5 under different RH intervals, with strong negative power function relationships (R 2 > 0.71, p < 0.01). The PM 2.5 thresholds corresponding to visibility of 10 km (PTV 10 ) varied from 54 to 175 μg m −3 and decreased with RH increasing due to rapid hygroscopic growth of particles 27 . The spatiotemporal variabilities of PTV 10 , e.g., 50-63 μg m −3 in Sichuan basin in 2014 62 and 66 μg m −3 in Wuhan during 2018 winter 28 , reduced the consistency and reliability of air quality studies using a fixed visibility (10 km) to reflect the occurrence of haze without considering air pollution intensity 63 . Meanwhile, the PTV 10 for RH > 90% was 13-28% lower than the Chinese secondary PM 2.5 standard (75 μg m −3 ), implying that the air quality standard was not always able to keep the visibility > 10 km. Thus, a strict hourly PM 2.5 standard value should be emphasized to prevent haze formation. In this study, it is 54 μg m −3 .

Increases in PTV 10
Compared with those (54-126 μg m −3 ) during PLP, the PTV 10 increased by 9-58 μg m −3 (p < 0.01) for different RH ranges during LP. In other words, the visibilities were higher in LP than those for PLP under the same RH and PM 2.5 concentration. Moreover, the elevations in PTV 10 implied that the reductions in PM 2.5 and RH could not thoroughly explain the increase in visibility. Other parameters including mass scattering efficiency (MSE), mass absorption efficiency (MAE), and optical hygroscopicity (f(RH)) for PM 2.5 should be also considered. In Fig. 2b, PM 2.5 was highly correlated with b sp and b ap (R 2 ≥ 0.58, p < 0.01). The slopes of the linear regressions can be considered as the bulk MSE and MAE 38,42 . During LP, the MSE decreased by~5% (0.26 m 2 g −1 ) compared with that for PLP. While due to the aerosol aging 64 , MAE raised by 24% (0.06 m 2 g −1 ), partly counteracting the MSE reduction. However, compared with the decrease in MSE, the decrements in f(RH) were more obviously as 6-14% (p < 0.01) for various RH ranges from PLP (1.57-3.46) to LP (1.48-3.12) (Fig. 2c). Since the ambient visibility was highly sensitive to f(RH) changes under high RH conditions 19 , the decline in f(RH) was one of the key reasons for the increase of PTV 10 . It is worth noting that the hygroscopic behavior of SOA has not been considered due to its minor contributions to b sp and f(RH).

The driver for ambient visibility improvement
The MSEs and MAEs of major chemical components in PM 2.5 were estimated by multiple linear regression (MLR) (Supplementary Table 1). The MSEs for NH 4 NO 3 varied slightly from 5.9 m 2 g −1 for PLP to 4.2 m 2 g −1 for LP, while the values for (NH4) 2 SO 4 changed dramatically from 1.2 to 4.4 m 2 g −1 . These MSEs were comparable to the values of 1.7-17.4 m 2 g −1 65,66 for NH 4 NO 3 and 1.9-9.2 m 2 g −1 23,67 for (NH4) 2 SO 4 in previous studies, which are summarized in Supplementary Table 2. For (NH 4 ) 2 SO 4 and NH 4 NO 3 , the variations in MSEs are related to particle size distribution, with particles in droplet mode (600-700 nm) holding the highest values 23,68,69 . The large increase in the MSE for (NH4) 2 SO 4 was likely due to the fact that its particle size increased and approached to droplet mode during the aerosol aging processes as the COVID-19 lockdown significantly reduced primary emissions 45,59,61 . For primary organic aerosol (POA), the calculated MSEs (8.4-9.3 m 2 g −1 ) were within the range of 1.0-16.7 m 2 g −1 67,70 (Supplementary Table 2). The MSEs for SOA, varying from 0.3 m 2 g −1 for PLP to 2.6 m 2 g −1 for LP, were lower than those for POA. They were also acceptable when compared with the values (1.1-8.5 m 2 g −1 71 ) in the former studies (Supplementary Table 2 Table 2, the MSEs for POA and SOA do not show necessarily high or low relationships as they are determined by the integral effects of mass concentrations, size distributions, emission sources, and morphology 72,73 . The increment in MSE for SOA might be associated with the variation in emission sources and the enlargement in particle size during LP. The MAEs for EC increased from 11.6 m 2 g −1 for PLP to 12.1 m 2 g −1 for LP due to the aerosol aging and were consistent with 7-14 m 2 g −1 66,74 . MAE of EC depends on its size distribution and coating 64 . It is noteworthy that there are no studies reporting MSEs of POA and SOA in PM 2.5 for comparison. From Fig. 1c, NH 4 NO 3 contributed the highest fractions to b ext by 61.8% and 31.8% for the PLP and LP, respectively. Then POA (22.1 and 28.1%, respectively) and (NH 4 ) 2 SO 4 (6.0 and 25.6%, respectively) followed. Other chemical species contributed < 7%. The estimated b ext contributed from all chemical species The differences in the mass concentrations and fractions of major chemical components in PM 2.5 for PLP and LP. c The differences in the contributions of major chemical components in PM 2.5 to aerosol extinction coefficient (b ext ) for PLP and LP. EC elemental carbon, POA primary organic aerosol, SOA secondary organic aerosol, FS fine soil. decreased by 8.7-179.5 Mm −1 during LP compared with those for PLP, except for (NH 4 ) 2 SO 4 and SOA. As the decreases in the mass concentration (56.1%) and MSE (28.8%), NH 4 NO 3 held the highest b ext reduction and its contribution to b ext displayed a decrease of 30%, which actuated the visibility improvement during LP. Liu et al. 16 found that from 2013 to 2017 the increased contributions of nitrate to particle mass and b ext elevated the f(RH) and mass extinction efficiency of PM 2.5 in Eastern China, which hindered the visibility improvement. From PLP to LP, the estimated b ext of SOA and (NH 4 ) 2 SO 4 increased by 15.0 and 40.3 Mm −1 , although the mass concentration of (NH 4 ) 2 SO 4 decreased by 29.4%. It indicated that the reduction strategies of PM 2.5 and associated chemical components currently in China would not reduce b ext and improve the ambient visibility necessarily. A significant cutting down of NH 4 NO 3 and its precursor (NO x or NH 3 ) can serve as the most effective way to improve ambient visibility in the future.

Weakened mutual promotion between AWC and NH 4 NO 3
Previous studies revealed the vital role of aerosol composition alterations on hygroscopicity and b ext 16,19,75,76 . The b ext induced by aerosol hygroscopicity (Δb ext ) was estimated, which was the difference between the ambient and measured b ext . In Fig. 3a, the Δb ext and aerosol water content (AWC) displayed a similar temporal pattern and decreased by~60% from PLP to LP averagely. The AWC was significantly (p < 0.01) correlated with Δb ext (R 2 = 0.86) and ambient visibility (R 2 = 0.72) (Fig. 3b), indicating that the reduction in AWC would be another reason for improving ambient visibility. Besides the decreases in ambient RH and PM 2.5 , 5.0-22.4 and 4.1-37.1% of the reductions in Δb ext and AWC could be ascribed to the aerosol composition variations, respectively (Fig. 4). NH 4 NO 3 and (NH 4 ) 2 SO 4 are the main hygroscopic compounds in aerosols and their abilities of water uptake are comparable with the same particle size and RH 33,77,78 . However, compared to (NH 4 ) 2 SO 4 (80% at 298 K), NH 4 NO 3 has a lower deliquescence RH (62% at 298 K) 77 and is more easily liquefied 79 . Following the method in Liu et al. 16 and Wexler and Seinfeld 80 , the average PM 2.5 deliquescence humidity for LP (71.3%) was significantly (p < 0.01) higher than that for PLP (70.0%) as the decrease in NH 4 NO 3 mass percentage (Supplementary Fig. 2). It means higher ambient RH requirement for hygroscopic growth 16 . In addition, the aerosol water facilitates NH 4 NO 3 formation 81-83 and the enhanced NH 4 NO 3 fraction will promote water uptake correspondingly 33,84,85 . Such mutual promotion between the aerosol water and NH 4 NO 3 degraded ambient visibility effectively (Fig. 3c), while the decreases in NH 4 NO 3 and RH weakened the promotion and then reduced AWC and Δb ext , which further improved the ambient visibility during LP.
Priority policies for co-regulating PM 2.5 and visibility Reducing NH 4 NO 3 can substantially reduce PM 2.5 and improve ambient visibility. This can be realized by reducing NO x to lower HNO 3 , which further transfers to particle phase, or reducing NH 3 to lower aerosol pH and keep HNO 3 in the gas phase 86 . In Fig. 5, all variables responded to NH x (NH 3 + NH 4 + ) reduction nonlinearly.
They flattened out until 36% (9.2 μg m −3 ) and 43% (6.9 μg m −3 ) NH x reductions achieved for PLP and LP, respectively, at which points they started to decrease rapidly. The sweet spots for NH x reduction (36 and 43%) were determined by a critical pH of 3, which balanced the partition between HNO 3 and NO 3 − 86 . It increased by 7% (Fig. 5a) due to the reductions of TNO 3 (NO 3 − + HNO 3 ) and SO 4 2− converting more NH x into gas phase during LP 87 , reflecting that the ambient visibility improvement would become more difficult via NH x control. Fu et al. 88 also reported that increases in free NH 3 concentration could decrease the sensitivity of PM 2.5 reduction to NH x emission control.
The impacts of reducing TNO 3 showed different responses. Decreasing TNO 3 did not obviously change pH due to the buffering by NH 3 -NH 4 + partitioning 86,89 . While the pH decreased clearly when NH x reduction exceeded its sweet spots due to the increase of TNO 3 partitioning to HNO 3 , the decrease of AWC, and the increase of hydrogen ion concentration in aerosol water 87   elevate ambient visibility than NH x . However, if the NH x reduction surpassed 69% (17.5 μg m −3 ) and 73% (11.7 μg m −3 ) for PLP and LP, respectively (Fig. 5b), it would become more effective in increasing ambient visibility than TNO 3 reduction. Wu et al. 90 suggested that the measures to reduce NH x pollution should be focused on non-agricultural emission sources in both local and surrounding areas of urban regions as the NH x emitted from agricultural sources has been highly overrated 90,91 .
For guaranteeing blue sky in the future, the responses of average PM 2.5 concentration for PLP to NH x or TNO 3 reduction were roughly simulated given that the anthropogenic emissions have rebounded to pre-pandemic levels after Wuhan reopened 45,48 . A reduction of 51% in TNO 3 (17.9 μg m −3 ) or 59% in NH x (15.0 μg m −3 ) could make the PM 2.5 concentration < 54 μg m −3 and ensure the ambient visibility > 10 km (Fig. 5d). When the NH x reduction exceeded 64% (16.3 μg m −3 ), it might be more effective in improving ambient visibility than TNO 3 reduction. The simultaneous reductions of NH x and TNO 3 with different ratios did not decrease their reduction threshold percentages corresponding to PM 2.5 of 54 μg m −3 (Fig. 6), which meant that just control TNO 3 was enough for improving ambient visibility. However, there are other welfares to control NH x emissions, for instance, reducing  nitrogen deposition and minimizing eutrophication in aqueous system 92 . Thus, multi-pollutant control but more priority given to TNO 3 reduction is proposed from the view of improving ambient visibility in China.
Since the secondary transformation has not been considered in the thermodynamic model, how to reduce TNO 3 and NH x to certain concentrations by controlling their corresponding precursors (NO x and NH 3 ) needs more in-depth studies. Cutting down the TNO 3 only via abating NO x emissions should be treated with cautions as decreasing NO x emissions may increase ozone and hydroxyl radical concentrations, which can enhance the conversion efficiency of NO x to HNO 3 and then subdue the response of TNO 3 to NO x emission reductions in the volatile organic compound (VOC)-limited ozone formation regime 18,93 . The increased photochemical oxidants were the major drivers for persistent heavy nitrate pollution in winter in North China Plain 93 . It should be noted that Wuhan is also in a VOC-limited ozone formation regime 94 .

Implications
To tackle the haze pollution, the Chinese government has implemented toughest ever emission control measures since 2013 13 . Consequently, the anthropogenic emissions of NH 3 , NO x , PM 2.5 , and SO 2 in Hubei province decreased by 7.8-70.0% from 2013 to 2017 (Supplementary Fig. 3a). The SO 2 and NO 2 concentrations in Wuhan reduced by 84.0 and 27.9% from 2014 to 2019, respectively ( Supplementary Fig. 3b). The PM 2.5 mass concentrations continually dropped by half from 141.2 to 73.6 μg m −3 during the 2014-2019 wintertime (Fig. 7a). However, the air quality improvement might not be sensed by the public since the average ambient visibility was less improved and still remained < 10 km (Fig. 7a), which obscured the efforts government paid to alleviate the air pollution. Though RH could also diminish the sky 16,19 , it was not the main reason curbing ambient visibility elevation as it displayed small fluctuations and no obvious variation was observed during 2014-2018 wintertime (Fig. 7a). The sharp decrease of ambient visibility in 2019 wintertime could be partly explained by the moderate increase in RH. The nonlinear responses of visibility to PM 2.5 could also explain its unsatisfactory improvement. In Fig. 2a, the ambient visibility showed decreasing sensitive to PM 2.5 decrement with the aggravation of air pollution especially when the PM 2.5 concentrations were higher than the PTV 10 . Thus, the large abatements in PM 2.5 mass concentrations during 2014-2019 wintertime did not bring about the huge improvement in ambient visibility as they were still > 54 μg m −3 , which pointed out the importance of establishing a strict hourly PM 2.5 standard. The great ambient visibility improvement appearing in LP can be expected in the future with the decrease of PM 2.5 when it is below the standard.
Such frustrating visibility improvement in Wuhan is not a particular case, which has been also found in Eastern China 16 and Southern China 19 . Even worse, it is likely to be widespread in China as Liu et al. 16 has demonstrated that nearly 73.2% stations across the country exhibited increasing slopes of AOD/PM 2.5 from 2013 to 2018. That is to say, though the PM 2.5 mass concentration has been substantially reduced, the increase of AOD per unit PM 2.5 indicated the less improved ambient visibility. Emission controls successfully brought down the loads of primary PM 2. 5 15 and inevitably reduced its b ext 19 , while most of the visibility  improvement benefits raised by PM 2.5 reduction were balanced out by the elevation in aerosol optical hygroscopicity 16,19 . This increment was associated with the elevated proportions of NH 4 NO 3 in PM 2.5 mass and b ext 16 . Indeed, (NH 4 ) 2 SO 4 typically dominated b ext (~40%) in the past decade 36 , while the distinct emission controls of SO 2 and NO x resulted in a larger reduction in sulfate than in nitrate in China from 2013 to 2017 17,93,95 . Instead, nitrate is more important than sulfate as a driver for ambient visibility impairment 96,97 . Similarly, the nitrate mass concentration and its contribution to SNA have gradually increased during 2015-2019 wintertime in Wuhan despite the tremendous mitigation of PM 2.5 pollution (Fig. 7b). The evidently increased nitrate proportions would directly trigger the decrease in PM 2.5 deliquescence humidity 16 . It meant that a lower ambient RH was required for aerosol hygroscopic growth, which hindered the visibility improvement in Wuhan during wintertime.
Based on a positive example induced by the unexpected COVID-19 pandemic, this study reveals the co-benefits of reducing PM 2.5 and improving ambient visibility by cutting down NH 4 NO 3 . Reducing NH 4 NO 3 will increase deliquescence humidity and decrease optical hygroscopicity, which can maximize the efficiency of decreasing PM 2.5 on improving ambient visibility under current air pollution condition. The recommendations for reducing PM 2.5 and improving visibility in a short term by reducing more TNO 3 than NH x are proposed. To resolve haze once and for all, the joint control of the two pollutants will gain other more welfares. It must be noted that wiping out the haze is not the terminus. The average PM 2.5 concentrations during LP still remained four times higher than the World Health Organization recommendations. Secondary inorganic aerosol (45.6%) and biomass burning (26.8%) were still the largest contributors to PM 2.5 ( Supplementary Fig. 4) though the masses they contributed both decreased during LP, which needed further reductions. As shown in Supplementary Fig.  5, these contributions were both enhanced by the air masses transported from Eastern China 56 , suggesting the necessity of regional-joint control.
Meteorological parameters, including atmospheric pressure, ambient temperature, RH, wind speed, and wind direction, were obtained by an automatic meteorological observation instrument (WS6000-UMB, Luff, Germany) with 1-h resolution. Hourly precipitation was provided by local meteorological administration. Ambient visibility was measured with a visibility monitor (Belfort Model 6000, USA) with ±10% of uncertainty. The mixing layer height was derived from the HYSPLIT model 56 . The time series of meteorological parameters are displayed in Supplementary Fig. 8.
The dry b sp at 525 nm was measured using an integrating Nephelometer (Aurora-1000, Ecotech, Australia) with RH of inflow air heated to < 40%. The b ap at 532 nm was converted by the concentration of black carbon measured at 880 nm with an Aethalometer (Magee Scientific Company, Berkeley, CA, USA, Model AE-31). The converted coefficient was 8.28 m 2 g −1 100 . To match the b sp at 525 nm, the b ap at 532 nm was converted as referred to Nessler et al. 101 . The verification and calibration of the Nephelometer and Aethalometer can be found in Cao et al 22 . and Tao et al 65 . The single scattering albedo was calculated as the ratio of b sp to b ext .

Data analysis
Fog, rain, and dust can reduce the atmospheric visibility. The datasets with RH > 97% was excluded to eliminate the effects of fog 19 . Data collected in the occurrence of precipitation were removed 63 . When the PM 2.5 /PM 10 ratio was < 30%, the data were excluded as they might be impacted by long-range transport dust 63,102,103 . The hourly data were deleted if PM 2.5 concentrations were higher the PM 10 concentrations. Totally, the eliminated data accounted for 20% of the whole data.
The measured PM 2.5 was reconstructed by the sum of (NH 4 ) 2 SO 4 , NH 4 NO 3 , OM, EC, and fine soil 104 . The minimum R-Squared method 105 and a constant converting factor were used to divide OM into POA and SOA 24,59 . The MSEs for above chemical components expect for EC were estimated by MLR 41,66,106,107 . The MAEs for EC were estimated based on the scatter plots of EC against b ap 24,41 . Statistics of MSEs and MAEs are presented in Supplementary Table 1. The bulk f(RH) for PM 2.5 was the ratio of estimated ambient b sp to corresponding measurements 19,35,108 . The thermodynamic model ISOROPPIA-II 109,110 was run with "forward mode" to calculate the AWC and to conduct sensitivity tests 86,87,111 . Positive matrix factorization (PMF 5.0) was employed to identify the sources of PM 2. 5 24,38,56 . More details about the data processing are listed in Supplementary Methods. The glossaries of abbreviations are provided in Supplementary Table 3.

DATA AVAILABILITY
Data are available on reasonable request from the corresponding author (kongshaofei@cug.edu.cn).