The AERosol and TRACe gas Collector (AERTRACC): an online-measurement-controlled sampler for source-resolved emission analysis

. Probing sources of atmospheric pollution in complex environments often leads to the measurement and sampling of a mixture of different aerosol types due to ﬂuctuations of the emissions or the atmospheric transport situation. Here, we present the AERosol and TRACe gas Collector (AERTRACC), a system for sampling various aerosol types independently on separate sampling media, controlled by parallel online measurements of particle, trace gas, and mete-orological variables, like particle number or mass concentration, particle composition, trace gas concentration, and wind direction and speed. AERTRACC is incorporated into our mobile laboratory (MoLa) which houses online instruments that measure various physical and chemical aerosol properties, as well as trace gas concentrations. Based on preparatory online measurements with the whole MoLa setup, suitable parameters measured by these instruments are used to de-ﬁne individual sampling conditions for each targeted aerosol type using a dedicated software interface. Through evaluation of continuously online-measured data with regard to the sampling conditions, the sampler automatically switches be-tween sampling and non-sampling for each of up to four samples, which can be collected in parallel. The particle phase and gas phase of each aerosol type, e.g., source emissions and background, are sampled onto separate ﬁlters with PM 1 and PM 10 cutoffs and thermal desorption tubes, respectively. Information on chemical compounds in the sampled aerosol is obtained by means of thermal desorption chemical ionization mass spectrometry (TD-CIMS) as the analysis method. The design, operation, and characterization of the sampler are presented. For in-ﬁeld validation, wood-ﬁred pizza oven emissions were sampled as targeted emissions separately from ambient background. Results show that the combination of well-chosen sampling conditions allows more efﬁ-cient and effective separation of source-related aerosols from the background, as seen by the increases of particle number and mass concentration and concentration of organic aerosol types, with minimized loss of sampling time compared to alternative sampling strategies.


S1 Calculation of PM1 mass concentrations
The PM1 mass concentrations were calculated from the combined particle number size distributions of FMPS (dp = 5.6 -560 nm) and OPC (dp = 0.25 -32 μm) assuming spherical particles with a density calculated based on the AMS and black carbon data using the equation of Kuwata et al. (2012) for organic density and Salcedo et al. (2006) for overall density.Since we found that the FMPS under-measures the concentrations in the uppermost size channels these were corrected using the lower OPC size channels.Details on OPC data treatment like the conversion from optical diameter to geometric diameter are provided in Drewnick et al. (2020).The uncertainty for the calculated PM1 mass concentration is 25%.It was calculated by error propagation from the uncertainty of the density (15%) and the uncertainty of the FMPS and OPC data merging (20%).

S2 AMS and PMF data analysis
For the AMS data analysis all standard analysis procedure steps were performed with SQUIRREL 1.63I and PIKA 1.23I.A collection efficiency of 0.5 (Canagaratna et al., 2007) was applied and ionization efficiency (IE) and relative ionization efficiency (RIE) were determined in calibrations before the measurements.Elemental ratios were calculated based on the improved calibration method (Canagaratna et al., 2015).
For PMF analysis of the organic aerosol, the high-resolution data with error matrix were prepared with PIKA 1.23I.Ions with signal-to-noise ratio (SNR) < 2 were downweighted through increase of the corresponding error by a factor of 2, while ions with SNR < 0.2 were discarded from the data.The CO2 + ion and related ions (m/z 16, 17, 18 and 28) were downweighted by a factor of SQRT(5) as they all contain the same information.Additionally, "noisy" ions without contribution to the total measured signal were discarded.To find a robust solution the analysis was run for 1 to 7 factor solutions, with fpeak -1 to 1 (steps of 0.1) and seed 0 to 50 (steps of 1).For further analysis, the three-factor solution was chosen with fpeak=0 and seed=0.
The solution was chosen based on comparison of the time series (Fig. S1a) with those of other instrument data and of the mass spectra (Fig. S1b) with literature references (Fig. S2).The residual mass is smaller than 1 %.

S3 CIMS data analysis
For the CIMS data analysis, the software Tofware 3.2.3(Aerodyne Inc., USA) and custom data procedures were used.All standard analysis procedure steps were performed including m/z calibration (with I(H2O) -, I(CH2O2) -, I(HNO3) -, I2 -and I3 -, deviation < 3 ppm), background correction using the field blanks and normalization to the iodide signal.

S4 Error calculation for Fig. 4
To determine the reproducibility, several samples were prepared with equal sample amounts by simultaneously sampling the same aerosol onto multiple filters and TDTs.For the overall reproducibility, the standard deviation over all samples for all individual compounds, which were identified in this study (Section 4), was calculated and then these standard deviations were averaged over all compounds.
As error for the signal intensity of individual compounds the uncertainty, derived from the reproducibility determination, and the error from a Gaussian error propagation of the standard deviation of the blanks and the samples were compared and the larger one was chosen.The signal intensity from compounds found on blank filters was negligible in contrast to source and background samples.The error for the ratios was calculated using Gaussian error propagation from the errors of signal intensity of source and background samples.Error bars of the overall source ratios represent the standard error of the ratios of all ions assigned to the respective sources.

Figure S1 :
Figure S1: Site map with the location of the institute (MPIC) within the city and a wind rose plot (a) and a magnification to show 25

Figure S2 :
Figure S2: Time series (a) and mass spectra with the source and background aerosol sampling times highlighted in blue and red (b) of the chosen 3-factor solution of the PMF analysis representing the three different aerosol types, observed during the field-validation measurement.

Figure S3 :
Figure S3: Comparison of the three PMF factor mass spectra with reference spectra.Shown are Pearson R values from correlation of the PMF factor mass spectra with different reference mass spectra from the AMS Spectral Database (Ulbrich et al., 2022) as color-coded boxes.

Figure S4 :
Figure S4: User interface of the AERTRACC software in manual sampling mode.For each sampling path, a sampling time and mass limit can be set.To sample, the "Active" checkbox needs to be checked; then the red indicator for "non-sampling" turns green 55

Figure S5 :
Figure S5: Size-dependent transport losses for the (a) PM1-only and (b) PM1/PM10 sampling mode.For the calculations, it was 60

Figure S6 :
Figure S6: Time series for the relevant parameters black carbon and PM1 mass concentration as well as CPC and OPC particle 65

Figure S7 :
Figure S7: Ion signal intensities normalized to the respective sampled volume for the PM1 and PM10 filters from pizza oven and background sampling.Error Bars show the larger uncertainty, either the reproducibility or the uncertainty estimated by error propagation from the standard deviation of the blank and the ambient measurements.