Source apportionment of PM 2 . 5 in Shanghai based on hourly 1 molecular organic markers and other source tracers 2

14 Identification of various sources and quantification of their contributions are a necessary step to formulating scientifically 15 sound pollution control strategies. Receptor model is widely used in source apportionment of fine particles. However, most 16 of the previous studies are based on traditional filter collection and lab analysis of aerosol chemical species (usually ions, 17 elemental carbon (EC), organic carbon (OC) and elements) as inputs. In this study, we conducted robust online 18 measurements of a range of organic molecular makers and trace elements, in addition to the major aerosol components 19 (ions, OC and EC), in urban Shanghai in the Yangtze River Delta region, China. The large suite of molecular and elemental 20 tracers, together with water-soluble ions, OC and EC, provide data for establishing measurement-based source 21 apportionment methodology for PM2.5. We conducted source apportionment using positive matrix factorization (PMF) and 22 compared PMF solutions with molecular makers added (i.e. MM-PMF) and those without organic markers. MM-PMF 23 identified 11 types of pollution sources, with biomass burning, cooking and secondary organic aerosol (SOA) as the 24 additional sources identified. The three sources accounted for 4.9%, 2.6% and 14.7% of the total PM2.5 mass, respectively. 25 During the whole campaign, the secondary source is an important source of atmospheric pollution, the average contribution 26 of secondary pollution sources is as high as 63.8% of the total PM2.5 mass. Grouping different sources to secondary and 27 primary, we note that SOC and POC contributed 45.1% and 54.9%, respectively. It is worth noting that the contribution of 28 cooking to PM2.5 mass only account for 2.6%, but it contributed to 10.7% of OC. Episodic analysis indicated that secondary 29 nitrate was always the main cause of PM2.5 pollution, while during non-episodic hours, vehicle exhaust made a significant 30 https://doi.org/10.5194/acp-2019-951 Preprint. Discussion started: 31 January 2020 c © Author(s) 2020. CC BY 4.0 License.

contribution. Through the application of the above-mentioned techniques to the Yangtze River Delta, more insights are 31 gained on the sources, formation mechanism and pollution characteristics of PM2.5 in this region. 32

33
In recent years, with the increasingly prominent problem of air quality, more and more attention has been paid to the 34 research of air pollution, which focuses on the study of atmospheric particulate matter (PM), especially fine particulate 35 matter (PM2.5) (Chen et al., 2007;Zhang et al., 2009a). The study of chemical composition of atmospheric PM2.5 is to help 36 understand the source, formation mechanism, and environmental effects. PM2.5 pollution reduces atmospheric visibility 37 (Chow et al., 2004) and exposure to PM2.5 is positively correlated with adverse health effects (Nel, 2005

66
The past receptor modeling studies relied on chemical composition data derived from 24-h time-integrated filter samples 67 followed by off-line laboratory analysis. The off-line nature severely limits its utility in addressing episodic pollution events 68 and in providing data to assess emission-based model evaluation of pollution sources and regional transport. Some 69 researchers have conducted online PM2.5 source apportionment, however, previous studies were mainly using the traditional 70 aerosol species as inputs . Organic matter constitutes a considerable share of PM2.5, while the online 71 analytical techniques used in the past were not suitable for describing this fraction in full.

72
In this study, online monitoring of atmospheric PM2.5 compositions, including inorganic ions, organic carbon (OC), 73 elemental carbon (EC), trace elements and organic molecular markers, was conducted in Shanghai from November 9 to 74 December 3, 2018. The purpose is to use the detailed high-time resolution speciation data (especially organic molecular 75 markers) to identify the sources of PM2.5 based on molecular-marker based PMF. This study gives insights into more 76 detailed source contributions, changes of sources and effects of the air pollution control strategies.  Figure S1 shows the comparison of reconstructed and measured PM2.5 mass for samples collected for 97 this study (Wang et al., 2016;Huang et al., 2014 contribution of source sectors to the concentration of ambient air pollutants at receptor sites, with an assumption of mass 102 conservation and a chemical mass balance between emission source and receptors. In this study, the Environmental 103 Protection Agency (EPA) PMF version 5.0 (Norris et al., 2014) was applied to perform the analysis. PMF decomposes the 104 measured data matrix, Xij, into a factor profile matrix, f kj, and a factor contribution matrix, gik, (Eq 1): (2)

107
In eq 1, Xij is the measured ambient concentration of target pollutants; gik is the source contribution of the k th factor 108 to the ith sample, and fkj is the factor profile of the jth specie in the kth factor; eij is the residual concentration for each data 109 point. PMF seeks a solution that minimizes an object function Q (Eq 2), based on the uncertainties of each observation uij.

110
The user provides the uij for each data point. The selection of the best factor in this study and the error estimation diagnostics 111 for each model result are described in the supplement information ( Figure S2, Figure S3, Table S1, Table S2).

112
PMF model assumes that the quantity of the input species is conserved, and the source profile is unchanged. In order 113 to minimize the impact of organic matter degradation on the deviation of mass conservation hypothesis, organic species 114 with low volatility and low reactivity are selected as input. The requirement of constant source profiles is not strictly met 115 when the receptor model is applied to measurement data covering a long duration (e.g., months or longer). However, 116 understanding/progress can be achieved despite the non-strict adherence to the requirements of the constant source profile.

117
The source profile parsed by PMF can be viewed as the averaged profile over the entire sampling period. In an atmospheric 118 environment, both primary organic aerosol (POA) and secondary organic aerosol (SOA) have the problem of changing 119 source profiles. Therefore, it is necessary and vital to obtain high time resolution data, preferably several hours for a sample 120 of data or shorter, as an input file for PMF model. The input files in this study are hourly data and the time span of whole 121 campaign is less than one month. As such, the source type information will not change significantly.

122
In this study, a total of 289 samples has been collected. The hourly chemical species selected as input to PMF model

128
The uncertainty of each data point is calculated according to Eq 3: where MDL is the method detection limit and EF is the error fraction determined by the user and associated with the 131 measurement uncertainty. The concentration data of species below the detection limit were replaced by 1/2 of the MDL, 132 and the uij was calculated by 5/6 of the MDL. For the concentration data of missing species, the missing value is the 133 geometric average value of the concentration of this species, and its uij is four times the geometric average value.

142
The pollution episodes occurred mostly in winter, due to adverse atmospheric conditions(such as more frequent 143 stagnation of atmospheric movement)and enhanced the impact on air quality from local and regional emissions. The 144 hourly meteorological parameters and PM2.5 concentration during the monitoring time is shown in Fig. 2

155
In this study, PMF source analysis was conducted in two scenarios. They were MM-PMF with organic tracers and PMFt without organic tracers included and the results were compared in detail. The abundance and nomenclature of the 157 organic tracers used are shown in Table 2 Table S3. Nevertheless, the base run results still have certain degrees of factor mixing. As the source specific 170 tracer compounds have similar temporal variations and the diversity of species components contained in the source, 171 chloride, sulfate and certain metal elements were found in different PMF-resolved profiles.

172
Factor recognition was based on the highest loaded species of each factor. The factor profiles of the 11-factor solution 173 are shown in Fig. 3

185
These results suggest that F1 may represent condensation of oxidation products of local emissions in the nighttime plus         The factor profiles of PMFt of the 8-factor solution is shown in Figure S6, and the difference of individual factor 358 contribution to PM2.5 and to OC from MM-PMF and PMFt are show in Fig.11