Size segregated particle number and mass emissions in urban Beijing

Jing Cai, Biwu Chu, Lei Yao, Chao Yan, Liine M. Heikkinen, Feixue Zheng, Chang Li, Xiaolong Fan, Shaojun Zhang, Daoyuan Yang, Yonghong Wang, Tom V. Kokkonen, Tommy Chan, Ying Zhou, Lubna Dada, Yongchun Liu, Hong He, Pauli Paasonen, Joni T. Kujansuu, Tuukka Petäjä, Claudia Mohr, Juha Kangasluoma, Federico Bianchi, Yele Sun, Philip L. Croteau, Douglas R. Worsnop, Veli-Matti Kerminen, Wei Du*, Markku Kulmala*, Kaspar R. Daellenbach* 1 Aerosol and Haze Laboratory, Beijing Advanced Innovation Center for Soft Matter Science and Engineering, Beijing University of Chemical Technology, Beijing, 100029, China

Ionization efficiency (IE) in this study was obtained from ammonium nitrate calibration, which was 167.6 ions/pg under air beam of 3.1×10 5 ions/s.
Recently, it was discovered that NO3 induces a positive bias on organic CO2 + concentrations in the AMS/ACSM systems by Pieber et al. (2016), which can be described as a function of ambient NO3 (μg/m 3 ) in combination with the CO2 + /NO3 ratio from pure NH4NO3 measurements (CO2 + /NO3)AN: For pure NH4NO3 aerosol from calibrations, we determined the magnitude of the CO2 + /NO3 artifact (Pieber et al., 2016) and parametrized it as a function of the fragmentation pattern of NO3 (NO + /NO2 + ) to account for changes in the vaporizer in the ACSM: (CO2 + /NO3)NH4NO3 = 0.025 ± 0.002 × (NO + /NO2 + )NH4NO3 Then we determined the CO2 concentration from OA using a two week moving average (NO + /NO2 + ) from ambient observations: (CO2 + )OA,meas = (CO2 + )meas -(CO2 + /NO3)NH4NO3 × (NO3)meas Further, we propagated the uncertainty of the subtraction when computing the PMF input matrices.   And to compare our sampling period with long-term measurement, we also present the NR-PM2.5 and BC measured from Feb 2018 to Jun 2019, which is shown in Figure S4.   Table S1. NPF days during the sampling period (April 6, 2018 to July 2, 2018)
Therefore, we employ five factors in this analysis.

OA-PMF validations
Prior to OA-PMF analysis, we tested constraining cooking, gasoline and diesel profiles in order to test whether vehicular emissions can be separated according to the fuel type. We performed 300 PMF runs times using a random a-value in the range of 0 to 0.1 independently for HOA and COA (increment of 0.05). A relatively small a-value is applied due to the high similarity of the profiles from gasoline and diesel exhausts. The results show clear indications that gasoline and cooking emissions are mixed (similar diurnals, Fig. S10). Based on this test, it seems not possible to separate vehicular emissions based on the fuel type used in our dataset. In OA-PMF analysis, organic mass spectra are imported and analyzed by the multi-linear engine (ME-2) algorithm (ME-2) implemented in the toolkit SoFi, Source Finder. Unconstrained runs exhibit that a mixed POA (primary OA) component and a SOA (secondary OA) component. According to numerous previous researches in Beijing, during our sampling period, BBOA (biomass burning OA) and CCOA (coal combustion OA) are usually the lowest throughout the year (Sun et al., 2018;Hu et al., 2017). To separate the primary sources, the mass spectral fingerprint of HOA and COA from Crippa (Crippa et al., 2013a;Crippa et al., 2013b) are applied to constrain the PMF runs. Since adding factors cannot further decrease Q/Qexp ratio, four-factor result is chosen (Fig. S11). In the next step, we perform sensitivity tests by performing 2500 PMF runs times using a random a-value in the range of 0 to 1 independently for HOA and COA(increment of 0.1).

Fig. S11 Q/Qexp as a function of the number of factors in OA-PMF analysis
We evaluate the environment interpretability of all the PMF runs by the following criteria. Only PMF runs that meet all the criteria below are accepted.  Fig. S12 (c) and (d). The residual profiles exhibit no sign of significant unexplained primary fragments, suggesting the primary sources are well resolved. No significant trend of m/z 115 is observed. The daily average m/z 60 shows a slightly higher level for the first period with very large variations. This may due to the uncertainties in PMF analysis or a bit more biomass burning activities, which is similar with the trend of large particles in the Size-PMF (Regional-related 1 factor). However, considering the low m/z 60 concentrations and residuals as well as the large size of biomass burning particles, it will not significantly affect the primary source estimations in this study.    -Villegas et al., 2018). In this plot, the concentrations and error bars from cooking source tests are multiplied with a factor of 0.5. The green and orange regions showed the RIEs for OA typically applied  and cooking source test (Reyes-Villegas et al., 2018). The yellow and green dash line represent the average RIECOA of 2.29 from cooking sources tests and typical default RIEOA of 1.4.

Fig. S18
Daily comparison between COA mass concentration and normalized signal intensity of (a) linoleic acid (C18H32O2) and (b) pyrogluamic acid (C5H7NO3). The normalized signals intensities are defined as the ratio of the raw signal intensities of those compounds to the total reagent ions (sum of NO3 -, HNO3·NO3and (HNO3)2·NO3 -). Daily averaged concentrations are applied due to it is less affected by diurnal variations of temperature and photochemistry. To compare with gas phase procurers in a long period, here COA concentration was extracted from all sampling days, which has a good correlation with the COA extracted from non-NPF days in this study (Slope=0.99, r=0.89).

Fig. S19
Average particle size distribution of NPF days, Haze days, no Haze nor NPF event days and days used in PMF.