Variation of size-segregated particle number concentrations in winter 1 Beijing

17 The spatial and temporal variability of the number size distribution of aerosol particles 18 is an indicator of the dynamic behavior of Beijing’s atmospheric cocktail. This variation 19 reflects the strength of different primary and secondary sources, such as traffic and new 20 particle formation, as well as the main processes affecting the particle population. In 21 this paper, we report size-segregated particle number concentrations observed at a 22 newly-developed Beijing station during the winter of 2018. Our measurements covered 23 particle number size distributions over the diameter range of 1.5 nm-1 μm (cluster mode, 24 nucleation mode, Aitken mode and accumulation mode), thus being descriptive of a 25 major fraction of the processes taking place in the atmosphere of Beijing. Here we focus 26 on explaining the concentration variations in the observed particle modes by relating 27 them to the potential aerosol sources and sinks, and on understanding the connections 28 between these modes. We considered haze days and new particle formation event days 29 separately. Our results show that during the new particle formation (NPF) event days 30 increases in cluster mode particle number concentration were observed, whereas during 31 the haze days high concentrations of accumulation mode particles were present. There 32 was a tight connection between the cluster mode and nucleation mode on both NPF 33 event and haze days. In addition, we correlated the particle number concentrations in 34 different modes with concentrations of trace gases and other parameters measured at 35 our station. Our results show that the particle number concentration in all the modes 36 correlated with NOx, which reflects the contribution of traffic to the whole sub-micron 37 size range. We also estimated the contribution of ion-induced nucleation in Beijing, and 38 found this contribution to be negligible. 39

allows them to deposit into the brain (Oberdörster et al., 2004). Indeed, studies have 48 pointed out that ultra-fine particles, which contribute to a negligible fraction of the mass 49 concentration, dominate the total number concentration in urban areas (von Bismarck-50 Osten et al., 2013;Wehner et al., 2004;Wu et al., 2008). Due to their high concentrations, 51 ultrafine particles' toxicological effects are enhanced by their large total surface area 52 (Kreyling et al., 2004). 53 Apart from their health effects, the temporal and spatial variation of particle number 54 concentrations of different sizes is a good indicator of the strength of their emission 55 sources. Aerosols are emitted either directly as primary particles, such as sea salt or dust 56 particles as a result of natural phenomena (Solomos et al., 2011), or they can be formed 57 through new particle formation (Kulmala, 2003;Kulmala et al., 2004;Kulmala et al., 58 2013; Kerminen et al., 2018;Chu et al., 2019). Newly formed particles can grow up 59 diameters of 20-100 nm within a day (Kulmala et al., 2004), and they have been found 60 to contribute to a major fraction of the global cloud condensation nuclei population 61 (CCN), thus indirectly affecting the climate (Kerminen et al., 2012). For all 62 aforementioned reasons, and in order to form a collective and complete picture about 63 atmospheric aerosol particles to understand their origin and potential impacts at a 64 specific location, the whole size distribution of these particles needs to be studied.
For instance, two-years of observations of particle number size distributions at a site in 73 northern Beijing reported that traffic emissions were the major source of nucleation (3-74 20 nm) and Aitken (20-100 nm) mode particles in urban Beijing (Wang et al., 2013). 75 On the other hand, research conducted in western downtown of Nanjing reported that 76 local new particle formation events were the main contributors of both nucleation (5-77 20 nm) mode and CCN particle populations (Dai et al., 2017). Measurements of 78 nucleation mode particle concentrations in urban Hong Kong reported the dominant 79 contribution of combustion sources to the nucleation mode (5.5-10 nm) (Wang et al.,80 2014a), whereas observations in urban Guangzhou found that accumulation and 81 secondary transformation of particles were the main reasons for high concentrations of 82 accumulation mode particles (100-660 nm) (Yue et al., 2010). However, only a few 83 studies in China have reported measurements of cluster mode (sub-3 nm) particles and 84 related them to new particle formation events (Cai et al., 2017;Xiao et al., 2015;Yao et 85 al., 2018;Yu et al., 2016). 86 The observation of sub-3 nm particles and ions has been made possible by recent major 87 developments in instrumentations, such as the particle size magnifier (PSM) (Vanhanen 88 et al., 2011), diethylene glycol-based scanning mobility particle sizer (DEG-SMPS) 89 (Jiang et al., 2011) and Neutral Cluster and Air Ion Spectrometers (NAIS) (Manninen 90 et al., 2016;Mirme et al., 2007). 91 In complicated environments like Beijing, it is very hard to relate each particle mode to 92 a specific source. Indeed, several sources could contribute to aerosol particles in the 93 same size range. For instance, cluster mode particles mainly originate from secondary 94 gas-to-particle transformation processes (Kulmala et al. 2013), although recently also 95 traffic has been identified as a source for these particles (Rönkkö et al., 2017). While 96 cluster mode particles can grow into the Aitken mode , also other sources like traffic 97 contribute to this mode, making the source identification of the Aitken mode 98 complicated (Pirjola et al., 2012). Various anthropogenic activities and biogenic 99 processes contribute to accumulation mode particle sizes. Thus, correlating trace gases 100 and aerosol concentrations of different sizes during different time periods help 101 narrowing down these aerosol sources.

102
In this study, we analyzed the number concentration of four sub-micron aerosol modes: 103 cluster mode (sub-3 nm), nucleation mode (3-25 nm), Aitken mode (25-100 nm), and 104 accumulation mode (100-1000 nm). Our aims were i) to investigate the number 105 concentration variations of size-segregated aerosol number concentrations for each 106 mode, ii) to explore the relationships between the different modes under different 107 atmospheric conditions, iii) to connect the number size distribution modes with multiple 108 trace gases (NOx, SO2, CO and O3) and PM2.5 (particulate matter with aerodynamic 109 diameter less than 2.5 μm), and iv) to quantify the contribution of NPF and haze  The station represents a typical area in urban Beijing subject to pollution sources, such 129 as traffic, cooking and long-range transport of pollution. The campus is surrounded by 130 highways and main roads from the east (3 rd ring main road), north (Zizhu road) and 131 south-east (Zizhu Bridge). From the east, west and south, the campus is surrounded by 132 residential and commercial areas.

133
Measurements at SMEAR Beijing started on 16 January, 2018(Lu et al., 2018 within 240 s. We averaged the data over 3 scans to make it smoother, and therefore the 153 time resolution of PSM data was 12 minutes. The data were inverted with a kernel 154 function method. When comparing the particle number concentrations obtained with 155 the expectation-maximization method, the cluster mode particle number concentration 156 was, on average, twice higher on the NPF event days and eleven times higher on the 157 haze days (Cai et al., 2018). Therefore, there is some uncertainty in the reported cluster 158 mode particle concentrations.

159
A particle size distribution (PSD) system measured the particle number size distribution resolution of SO2 data was 1 hour before 22 January, 2018, and 5 minutes after that.

181
The PM2.5 data were obtained from the nearest national monitor station, Wanliu station, 182 around 3 km north of our station. The PM2.5 data from Wanliu station compared nicely 183 with the PM2.5 data from three other adjacent national stations. The time resolution of 184 the PM2.5 data was 1 hour, and these data were recorded every hour. Detailed 185 information is reported in Cao et al. (2014). 186 We measured the relative humidity (RH, %), visibility (km), wind speed (m/s) and wind 187 direction ( 。 ) from a weather station on the roof of our station.

188
When data sets having different time resolutions were used, we chose the smallest time  193 We classified days into "NPF event days" and "haze days". The days that did not fit 194 either of these two categories were marked as "Other days", and they were excluded 195 from our future analysis unless otherwise specified. We observed 28 NPF event days 196 and 24 haze days in total.  198 We identified the NPF event days following the method introduced in (Dal Maso et al.,199 2005), which requires an appearance of a new mode below 25 nm and that the new were classified as haze days when the haze event lasted for at least 12 consecutive hours.

204
During our study periods, there was no overlap between the NPF events and haze days, 205 as these two phenomena never occurred simultaneously. While the NPF events 206 appeared right after sunrise and lasted for several hours, the haze events did not have 207 any specific time of appearance but lasted from a few hours up to several days.

208
The particle number size distribution was divided into 4 modes according to their  Figure S1, total particle number concentrations from the NAIS and PSD system 218 correlated well with each other on both NPF event days (R 2 was 0.92) and haze days 219 (R 2 was 0.90) in the overlapping size range. The slopes between the total particle 220 number concentration from the PSD system and that from the NAIS were 0.90 and 0.85 221 on the NPF event days and haze days, respectively. The particle number size 222 distribution in the overlapping size range of the NAIS and PSD system matched well 223 on both NPF event days and haze days as shown in Figure S2. The growth rates of cluster and nucleation mode particles were calculated from positive 230 ion data and particle data from Neutral Cluster and Air Ion Spectrometer (NAIS), 231 respectively, by using the appearance time method introduced by Lehtipalo et al. (2014).

232
In this method, the particle number concentration of particles of size dp is recorded as 233 a function of time, and the appearance time of particles of size dp is determined as the 234 time when their number concentration reaches 50% of its maximum value during new 235 particle formation (NPF) events.

236
The growth rates (GR) were calculated according to: where 2 and 1 are the appearance times of particles with sizes of 2 and 1 239 respectively. Figure S3 shows an example of how this method was used.

Calculation of the coagulation sink 241
The coagulation sink (CoagS) was calculated according to the equation (2)

Calculation of the formation rate 249
The formation rate of 1.5-nm particles ( 1.5 ) was calculated using particle number  (3) and equation (4), respectively: where is the coagulation sink in the size range of [ , + ∆ ] and GR 257 is the growth rate.
The fourth and fifth terms on the right hand side of equation (4)   severe haze episode that took place in Beijing in January 2013 (Wang et al., 2014b).

301
The median concentration of O3 was 10 ppb on the haze days during our observations, 302 a little bit higher than the severe haze episode in 2013 (<7 ppb; Wang et al., 2014b).

303
The median levels of SO2, CO, NOx and O3 were 230%, 50%, 100% and 50% higher, 304 respectively, on the haze days than on the NPF days. SO2, CO and NOx are usually

Diurnal behavior 309
In order to draw a clear picture of the evolution of size-segregated particle number 310 concentrations, we analyzed the diurnal concentration behavior of the different trace 311 gases ( Figure 5) and particle modes ( Figure 6).

312
Since trace gases have more definitive sources than particles, we can get some insight  for O3 via photochemical reactions . In our observations, the diurnal 322 pattern of O3 was opposite to that of NOx, which is consistent with O3 loss by large 323 amounts of freshly emitted NO during rush hours and O3 production by photochemical 324 reactions involving NO2 after the rush hours in the morning.

325
In Figure 7, we show the median diurnal pattern of particle number size distribution on

343
As shown in Figure 6, on the NPF event days, the cluster mode particle number shoulder was only 20% of the maximum nucleation mode particle number 361 concentration. These results suggest that, compared with atmospheric NPF, traffic 362 contributed much less to the nucleation mode particle number concentration.

363
During the haze days, the diurnal pattern of the nucleation mode particle number

394
The concentration of accumulation mode particles was an order of magnitude higher 395 during the haze days compared with the NPF days, causing a higher condensation sink 396 (on average 0.015 s -1 for the NPF event days and 0.10 s -1 for the haze days, as shown 397 in Figure S4), and thus introducing a reason why NPF did not take place on the haze 398 days (Kulmala et al., 2017). The concentration, on the other hand, did not experience 399 much diurnal variation. There was a slight increase in the accumulation mode particle

Correlation between the particle modes and trace gas and PM2.5 concentrations 407
Beijing's atmosphere is a very complicated environment (Kulmala, 2015). Aerosol  (Table 2a and Table 2b), we can get further insights

421
SO2 is a key precursor for H2SO4 through photochemical reactions in Beijing, which is 422 in turn a requirement for new particle formation in megacity environments (Wang et al.,423 2013; Yao et al., 2018). Although being a very important precursor of NPF, SO2 had 424 lower concentrations on the NPF event days than on the haze days (Figure 8). High 425 concentrations of SO2 have been ascribed to regional pollution and anthropogenic 426 condensation sink even in semi-pristine environments (Dada et al., 2017). Earlier 427 observations report that the main sources of SO2 are power plants, traffic and industry, 428 so SO2 can be used as a tracer for regional pollution (Yang et al., 2018;Lu et al., 2010).

429
Generally, as shown in Figure 8, the SO2 concentration correlated negatively with both 430 cluster and nucleation mode particle number concentrations. Higher SO2 concentrations 431 were encountered on more polluted days when NPF events were suppressed due to the 432 high particle loadings, explaining the overall negative correlation. However, if we look 433 at the NPF event days and haze days separately, we cannot see any clear correlation 434 between the SO2 concentration and cluster mode or nucleation mode particle number 435 concentration, as shown also in Table 2a and Table 2b. This result indicates that during 436 our observations, NPF occurred in relatively clean conditions, but the strength of a NPF 437 event was not sensitive to the regional pollution level as long as NPF was able to occur.

449
NOx is usually considered as the pollution tracer mainly from traffic (Beevers et al.,450 2012). As shown in Table 2a and Figure 9, the NOx concentration correlated negatively 451 with both cluster and nucleation mode particle number concentrations on the NPF event 452 days. Compared with the correlation between SO2 and cluster and nucleation mode 453 particle number concentrations, this result indicates that local traffic emissions affected 454 cluster and nucleation mode particles more than regional pollution on the NPF event 455 days.

456
On the haze days, we did not see any correlation between the cluster mode particle 457 number concentration and NOx concentration (Table 2b), although according to our . On the haze days, the accumulation mode particle number 474 concentration correlated less with NOx than with SO2, suggesting that regional and 475 transported pollution was a more important contributor to accumulation mode particles 476 than traffic emissions.

492
Ozone is a secondary pollution trace gas and its concentration represents the oxidization 493 capacity of atmosphere. Earlier observations found that high O3 concentrations favor 494 NPF by enhancing photochemical reactions (Qi et al., 2015). However, we did not see 495 any correlation between the O3 concentration and cluster mode particle number 496 concentration, suggesting that O3 was not the limiting factor for cluster mode particle 497 number concentration.

498
The O3 concentration correlated positively with both nucleation and Aitken mode 499 particle number concentration on the NPF event days during the NPF time window, 500 whereas on the haze days O3 concentration correlated only with the Aitken mode 501 particle number concentration.

502
The above results suggest that O3 influences heterogeneous reactions and particle 503 growth rather than the formation of new aerosol particles.

Connection to PM2.5 505
As shown in Figure 10, the PM2.5 concentration correlated negatively with the cluster 506 and nucleation mode particle number concentrations, and positively with the 507 accumulation mode particle number concentration. High PM2.5 concentrations tend to 508 suppress NPF by increasing the sinks of vapors responsible for nucleation and growth 509 of cluster and nucleation mode particles. The particles causing high PM2.5 510 concentrations also serve as sinks of cluster and nucleation mode particles by 511 coagulation.

512
As shown in Table 2a and Figure 12, the Aitken mode particle number concentration  (Table 2b). A possible reason for this is that pre-existing large 519 particles acted as a sink for Aitken mode particles by coagulation as well as a sink for 520 vapors responsible for the growth of smaller particles into the Aitken mode. In addition, 521 while PM2.5 is dominated by regional and transported secondary aerosols, Aitken mode 522 particles mainly originate from local emissions such as traffic and cooking in Beijing 523 (Wu et al., 2007;Wang et al., 2013;Du et al., 2017;de Jesus et al., 2019). 525   Table 3a and Table 3b as well as Figure 11 show the correlation between particle 526 number concentrations in different modes. On the NPF event days, cluster and 527 nucleation mode particle number concentrations correlated positively with each other 528 due to their common dominant source, NPF. Both cluster and nucleation mode particle 529 number concentrations correlated negatively with the Aitken and accumulation mode 530 particle number concentrations because, as discussed earlier, high concentrations of 531 large particles tend to suppress NPF and subsequent growth of newly-formed particles.

532
On the NPF event days, Aitken and accumulation mode particle number concentrations 533 correlated positively with each other, as well as with the SO2 and NOx concentration.

534
This suggests that on the NPF event days, Aitken and accumulation mode particles both 535 formed during regional transportation as secondary particles and were emitted by traffic 536 as primary particles.

537
On the haze days, cluster and nucleation mode particle number concentrations 538 correlated positively with each other, and with the Aitken mode particle number 539 concentration. This is suggestive of a similar dominating sources for these particle, 540 most likely traffic emissions. Similar to the NPF event days, cluster and nucleation 541 mode particle number concentrations correlated negatively with the accumulation mode 542 particle number concentration, even though this correlation was rather weak (Table 3b).

543
As expected based on the discussion in section 3.3.5, the Aitken mode particle number 544 concentration had a negative correlation with the accumulation mode particle number 545 concentration on the haze days.

547
In order to estimate the contribution of ions to the total cluster mode particle number 548 concentration and the importance of ion induced nucleation in Beijing, we studied ion 549 number concentrations in the size range of 0.8-7 nm by dividing them into 3 sub-size 550 bins: constant pool (0.8-1.5 nm), charged clusters (1.5-3 nm) and larger ions (3-7 nm).

551
As shown in Figure 12, number concentrations of positive ions were higher than those 552 negative ions in all the size bins on both NPF event days and haze days. We will only 553 discuss positive ions here.

554
The median number concentration of positive ions in the constant pool on NPF event 555 days was only 100 cm -3 in Beijing, much less than that in the boreal forest (600 cm -3 ; 556 Mazon et al., 2016). Also, the median number concentration of positive charged clusters 557 was 20 cm -3 on the NPF event days, and the ratio to the total cluster mode particle 558 number concentration was 0.001 to 0.004 during the NPF time window (Figure 13).

559
This ratio is comparable to that observed in San Pietro Capofiume (0.004), in which the 560 anthropogenic pollution level was also high, but clearly lower than that observed in 561 another megacity in China, Nanjing (0.02; Kontkanen et al., 2017). Considerably higher 562 ratios were observed in clean environments, for example during winter in the boreal 563 forest at Hyytiälä, Finland (0.7; Kontkanen et al., 2017). The median number 564 concentration of larger ions (3-7 nm) on the NPF event days was 30 cm -3 , a little bit 565 higher than the charged cluster mode particle number concentration, indicating that not 566 all of the larger ions originate from the growth of charged clusters, but rather from 567 charging of neutral particles by smaller ions. On the haze days, charged ion number 568 concentrations were much lower than those on the NPF days, which could be attributed 569 to the higher condensation sink.

570
The diurnal pattern of the ratio of number concentration between charged and total 571 cluster mode particles was the highest during the night with a maximum of 0.008, and Hyytiälä (Kontkanen et al., 2017). This ratio reached its minimum around noon, 575 because the total cluster mode particle number concentration reached its maximum 576 around that time due to NPF. The ratio had a small peak at around 9:00, similar to earlier 577 observations in Centreville and Po Valley (Kontkanen et al., 2016;Kontkanen et al., 578 2017). The possible reason is that charged clusters were activated earlier in the morning 579 than neutral clusters. The ratio increased from the midnight until about 4:00, similar to 580 the number concentration of charged clusters.

581
As shown in Figure 14, the diurnal median of the ratio between the formation rate of 582 positive ions of 1.5 nm ( 1.5 + ) and the total formation rate clusters of 1. The growth rates of particles generated from NPF events were examined in three size 588 ranges: <3 nm, 3-7 nm and 7-25 nm ( Figure 15). The median growth rates of particles 589 in these size ranges were 1.0 nm/h, 2.7 nm/h and 5.5 nm/h, respectively. The growth 590 rate of cluster mode particles was comparable with that observed in Shanghai (1.5 nm/h; 591 Yao et al., 2018). The notable increase of the particle growth rate with an increasing 592 particle size is a very typical feature in the sub-20 nm size range (Kerminen et al., 2018), 593 and it may also extend to larger particle sizes (Paasonen et al., 2018).  4 Summary and conclusions 599 We measured particle number concentrations over a wide range of particle diameters sources of Aitken mode particles were local emissions, while transported and regional 612 pollution as well as growth from the nucleation mode also contributed to the Aitken 613 mode. The main source of accumulation mode particles was regional and transported 614 pollution. PM2.5 affected the number concentration of sub-100 nm particles by 615 competing for vapors responsible for particle growth and by acting as sinks for particles 616 by coagulation. The main contributors to the PM2.5 mass concentration were 617 accumulation mode particles on the haze days.

618
As demonstrated here and in many other studies (e.g. Brines et al., 2015), ultrafine 619 particles (< 100 nm in diameter) tend to dominate the total aerosol particle number