Temperature and VOC concentration as controlling factors for chemical composition of alpha-pinene derived secondary organic aerosol

15 This work investigates the individual and combined effects of temperature and volatile organic compound precursor concentration on the chemical composition of particles formed in the dark ozonolysis of α-pinene. All experiments were conducted in a 5 m Teflon chamber at an initial ozone concentration of 100 ppb and initial α-pinene concentrations of 10 ppb and 50 ppb, respectively, at constant temperatures of 20 °C, 0 °C, or -15 °C, and at changing temperatures (ramps) from -15 °C to 20 °C and from 20 °C to -15 °C. The chemical composition of the particles was probed using a High-Resolution 20 Time-of-Flight Aerosol Mass Spectrometer (HR-ToF-AMS). A four-factor solution of a Positive Matrix Factorization (PMF) analysis of the combined HR-ToF-AMS data from experiments conducted under different conditions is presented. The PMF analysis as well as elemental composition analysis of individual experiments show that secondary organic aerosol particles with the highest oxidation level are formed from the lowest initial α-pinene concentration (10 ppb) and at the highest temperature (20 °C). Higher initial α-pinene concentration (50 ppb) and/or 25 lower temperature (0 °C or -15 °C) result in lower oxidation level of the molecules contained in the particles. With respect to carbon oxidation state, particles formed at 0 °C are more comparable to particles formed at -15 °C than to those formed at 20 °C. A remarkable observation is that changes in temperature during or after particle formation result in only minor changes in the elemental composition of the particles. The temperature at which aerosol particle formation is initiatedinduced, thus seems to be a critical parameter for the particle elemental composition. 30 Comparison of the HR-ToF-AMS derived estimates of the content of organic acids in the particles based on m/z 44 in the mass spectra show good agreement with results from off-line molecular analysis of particle filter samples collected from the same experiments. While higherHigher temperatures are associated with a decrease in the absolute mass concentrations of organic acids (R-COOH) and organic acid functionalities (-COOH), while the organic acid functionalities account for an increasing 35 fraction of the measured SOAparticle mass at higher temperatures. Formatted


Introduction
Atmospheric aerosol particles can alter air quality (WHO 2016) and visibility ) on a regional scale. On a global scale, particles affect cloud formation, the radiative balance, and thus climate (IPCC 2013).
Atmospheric particles are chemically diverse entities, often with a significant mass fraction of organic compounds (Zhang et 5 al. 2007;). Secondary Organic Aerosol (SOA) is formed from condensation of oxidation products of volatile organic compounds (VOC) emitted from both anthropogenic and biogenic sources .; Seinfeld and Pandis 2016). α-pinene is a biogenic VOC emitted from e.g. foliage of coniferous trees , and it has it been identified as the most common monoterpene in boreal forestforests all year round . In the atmosphere, α-pinene is oxidized primarily by ozone (O3), hydroxyl radicals (OH -),·), and nitrate radicals (NO3 -).·). Due to 10 their low vapor pressures, some of the gas phase oxidation products may partition onto already existing particles by condensation or reactive uptake, and contribute to particle growth ). In addition, some low vapor pressure oxidation products of α-pinene are able to nucleate (Kirkby et al. 2016) and are likely to play an important role in the initial growth of new particles in the atmosphere Ehn et al. 2014;Tröstl et al. 2016).
It is well established that the particle mass available for condensation of gases affects the partitioning of organic species 15 between gas phase and particle phase (Pankow , 1994b,) although, the traditional partitioning theory is limited in relation to non-liquid, more viscous particles e.g., such as α-pinene derived SOA formed at low relative humidity , because of slow diffusion Pöschl, 2011).
The fraction (F) of a given semi-volatile species in the particle phase at a given temperature has been formulated in an absorptive equilibrium partitioning framework as 20 where c* is the gas phase mass concentration at saturation and M is the mass concentration of absorbing material . Thus, the chemical composition of a particle that is in equilibrium with the surrounding gas phase is affected by both c* and M. The c* of a gaseous compound is generally inversely related to its level of oxidation ).
The particle composition can be shifted towards species with higher c* values (i.e. less oxidized, more volatile species) by 25 increasing the mass concentration of pre-existing particles, i.e. the value of M; conversely, lower M values result in particle phase compositions that are dominated by species with lower c* values (i.e. more oxidized, less volatile species). This has been experimentally confirmed by e.g. ), who showed that the oxidation level of SOA from α-pinene ozonolysis decreases with increasing particle mass loadings.
The equilibrium partitioning of a gas with a given c* (volatility of gases (c*)) also depends on temperature, which affects 30 equilibrium partitioning and thereby particle mass as demonstrated by , ), based on chamber studies of α-pinene derived particles formed at different constant temperatures between -30 °C and 45 °C. Partitioning has also been addressed in chamber studies where the temperature was changedramped after the initial (constant temperature) formation of SOA.  increased the temperature from 22 °C to a maximum of 40 °C, and in some experiments decreased the temperature back again to 22 °C. During heating, they observed a decrease in 35 SOA size, an indication of evaporation, and during cooling an increase in SOA size, an indication of condensation. In experiments by , where the temperature was cycled in the ranges of 5 °C to 27 °C and 27 °C to 45 °C, heating was associated with a decrease in particle mass, and cooling associated with an increase in particle mass. In a recent study by Zhao et al. (2019), where the temperature was cycled between 5 °C and 35 °C (RH varied between 10 and 80 %), it is suggested, that condensation during cooling is smaller than predicted by equilibrium partitioning.
The chemical composition of gas and particle phase in α-pinene ozonolysis experiments is determined by a combination of thermodynamic and kinetic aspects (Zhang et al. 2015;. The effect of temperatures below room temperature (around (~20 °C), in particular below 0 °C, on gas phase oxidation products, nucleation, SOA growth, and particle chemical composition, however, remains a largely unexplored area. , Simon et al. 2020. Since low temperatures are of high atmospheric relevance due to their prevalence, e.g. low temperatures are prevailing at the latitudes of 5 the boreal forests and at higher elevation, it is important to quantify SOA formation and properties under cold conditions. Furthermore, vertical transport can lead to changes in temperature within short time frames, affecting reaction kinetics, condensation processes, and particle properties relevant for the climate effect of particles .
The lack of knowledge on how the chemical composition, of both the gas phase and particle phase, vary with temperatures 10 was the motivation behind the Aarhus Chamber Campaign on HOMs and Aerosols (ACCHA) introduced in the companion paper Kristensen et al. (manuscript submitted2020). The ACCHA campaign focuses on temperatures from 20° C to -15° C, corresponding to conditions relevant in the boreal forest regions (Portillo-Estrada et al. 2013). As in most chamber experiments VOC concentrations in the ACCHA campaign (10-50 ppb) were one to two orders of magnitude higher than typical ambient conditions (Kourtchev et al. 2016). These conditions were chosen to speed up aerosol formation in the experiments and we 15 believe the data provides valuable and atmospherically relevant information, applicable to e.g. the boreal forest areas.
The impact of temperature on the yield of highly oxygenated organic molecules (HOMs) yield is presented in Quéléver et al.
(2019)), and more details on the volatile organic compounds are presented in Rosati et al. (2019).
The goal of the current paper is to investigate and quantify the individual and combined effects of α-pinene precursor concentration and temperature on SOA mass concentration and chemical composition. For this purpose, we here describe and 20 discuss a subset of the data collected during the ACCHA campaign, focusing on results obtained from a High-Resolution Time-of-Flight Aerosol Mass Spectrometer (HR-ToF-AMS).

Experimental 25
This work is based on experiments conducted in the Aarhus University Research on Aerosol (AURA) chamber; a ~5 m 3 bag made of 125 µm FEP Teflon film located in an enclosure, where the temperature is controllable between -16 °C and 26 °C.
The AURA chamber has been described in detail by .
The experiments were conducted as part of the ACCHA campaign and focus on SOA formed in dark ozonolysis of α-pinene. at various temperatures. An overview of the campaign is provided in Kristensen et al. (manuscript submitted2020). To ease 30 the reading of the current paper, a short summary of the ACCHA campaign is given here, and an experimental a modified version of the overview table withof the experiments from Kristensen et al. (2020), is presented in Table 1 where focus is on the parameters relevant in this work is presented (Table 1).,. At a constant chamber temperature of either 20 °C, 0 °C, or -15 °C, ozone was injected into the chamber to a concentration of ~100 ppb, followed by injection of either 10 ppb α-pinene (low concentration) or 50 ppb α-pinene (high concentration).) α-pinene. The chamber was operated at atmospheric pressure, and 35 neither seed particles nor OH-scavengers were introduced.
Three series of constant temperature experiments, all consisting of an experiment at 20 °C, 0 °C, and -15 °C, were conducted.
OneIn one of the series was performed at the initial 10 ppb α-pinene concentration of 10 ppb was injected into the chamber (experiments 1.1-1.3), andwhile two similar series of experiments were initiatedperformed at 50 ppb α-pinene (experiments 2.1-2.3 and 3.1-3.3). 220-272 minutes after α-pinene injection (i.e. after the SOA mass concentration has peaked) 40 inAdditionally, the series consisting of 10 ppb α-pinene experiments 2.1-2.3,includes two temperature ramp experiments, where the temperature was in a continuous ramp changed from the original setting, either by cooling the chamber decreased from 20 °C to -15 °C (experiment 2.1), by cooling.4) and increased from 0 °C to -15 °C followed by heating to 20 °C (experiment 2.2), or by heating from -15 °C to 20 °C (experiment 2.3). In two additional experiments with an initial α-pinene concentration of 10 ppb (experiments 1.4 and 1.5) the temperature was ramped from the initial temperature at 20 °C to -15 °C and from -15 °C to 20 °C, respectively, ~35 minutes after α-pinene injection (i.e., which corresponds to the period during SOA 5 formation and before themass peak in SOA mass concentration)..
In this work, we present data from a subset of instruments involved in the ACCHA campaign: a temperature and humidity sensor (HC02-04) attached to a HygroFlex HF320 transmitter (Rotronic AG) placed in the center of the chamber, a scanning mobility particle sizer (SMPS)), consisting of a Differential Mobility Analyzer (DMA, TSI 3082) and a nano water-based 10 condensation particle counter (CPC, TSI 3788), and a High-Resolution Time-of-Flight Aerosol Mass Spectrometer (HR-ToF-AMS, Aerodyne Research Inc.) DeCarlo et al. 2006;). In the following, the HR-ToF-AMS will be referred to as AMS. Both the SMPS and AMS were placed at room temperature next to the chamber outlets, and the connecting tubing was temperature insulated.
By the end of each experiment, a particle sample was collected on a Teflon filter (0.45 µm pore size, Chromafil). Particle 15 samples were extracted and analyzed by an Ultra High Performance Liquid Chromatograph/Electrospray Ionization quadrupole Time-of-Flight Mass Spectrometer (UHPLC/ESI-qTOF-MS, Bruker Daltonic)), as described in Kristensen et al. (manuscript submitted(2020), where also the analytical method and results are presented in detail. Herein, we compare the findings from the UHPLC/ESI-qTOF-MS, hereafter referred to as LC-MS, andto the AMS measurements.

Data analysis 20
Positive Matrix Factorization (PMF)  has traditionally been used to investigate contributions of different sources to ambient particles, and the application of PMF to AMS data from chamber experiments was first demonstrated by Craven et al. (2012). In the present work, PMF analysis is applied to chemical composition data from SOA particles that are produced in the ozonolysis of α-pinene, but formed and aged under different temperatures and precursor concentrations, and consequently different particle loadings. High-resolution AMS mass spectra of SOA particles 25 from the various experimental conditions were analyzed in one matrix, allowing for monitoring spectral and elemental chemical composition changes that occur as conditions change. The PET tool (V 2.09A) was used to perform the PMF analysis on high -resolution AMS mass spectra, according to the principles described in detail by . PMF is a model that can be used to express measured mass spectra as a linear combination of factors that are the products of 30 constant mass spectra and related time profiles as follows: The measured mass spectral data is the matrix X, an m×n matrix with n ion masses measured at m different time points, and 35 xij is an element of this matrix. p is the number of factors chosen for the solution, gip is an element of the matrix G containing time series of the factors, and fpj is an element of the matrix F of constant factor mass spectral profiles. The matrix elements eij correspond to the error matrix, E, of residuals not explained by the model ). Equation (2) is solved using the PMF2 algorithm ), which uses linear least-squares fitting together with the constraints that the values of matrix F and G have to be non-negative. The solution is found by minimizing the fit parameter Q: 40 where σij is an element of a matrix containing the standard deviations for each element of X ).
Estimation of standard deviations was performed as outlined in  with "weak" ions (i.e. ions with signal/noise (S/N) < 2) being down weighted by a factor of two, and "bad" ions (i.e. ions with S/N < 0.2) being down weighted 5 by a factor of ten. Additional sources of uncertainty that are not accounted for in the PMF analysis of high-resolution mass spectra, are uncertainties related to high-resolution fitting, including errors in peak shape, and m/z calibrations . The number of factors (p) areis chosen based on a combination of evaluation of residuals, Q values, and a priori knowledge about the dataset ). In the result section, a four-factor solution of the PMF analysis of high-resolution AMS HR data is presented. Although the five-factor solution and six -factor solutionssolution have 10 lower Q/Qexpected (Qexpected ≈ m×n, i.e. the number of points in the data matrix )) compared to the four-factor solution, the higherlarger number of factors doesare not selected because they do not provide additionalany more interpretable information. Plots of about the Q/Qexpected and the residuals ofparticle composition. The background for choosing the four,factor solution over the five,-factor and six -factor solutions, as well as the results of the five and six factor solutions are shown is explained more detailed in the supplementary materialinformation (S1-S8S11). 15 Since previous laboratory experiments show that the collection efficiency (CE) and relative ionization efficiency (RIE) of laboratory SOA isare variable , the mass concentrations presented in the PMF analyses are estimated from the total SOA mass concentration, as obtained from integrated SMPS size distributions, assuming spherical particles and densities calculated from the AMS-derived elemental ratios . Densities are derived as averages based on AMS data from the last 30 minutes of each experiment, except for experiments 2.1-2.3 where only. Uncertainties related to 20 the partdensity calculation are described in .
Mass spectra of the experiment wherefactors from the temperature is kept constant is includedPMF analysis are compared using the methods described by Wan et al. (2002) and , respectively. The comparisons focus on both the entire high-resolution mass spectra, m/z 12 to m/z 115, and the range of m/z > 44 to prevent an impact from the most intense peaks, especially m/z 43 and m/z 44, which are the ones associated with the largest variation between the factors. 25 Low SOA concentrations in the beginning of the experiments increase the uncertainty of the AMS measurements. Therefore, the first 4 to 16 minutes of the experiments (longest in the 10 ppb α-pinene experiments) are omitted from the elemental analysis of the AMS data.

Results and discussion
To provide an overview of the course of a typical experiment, Figure 1a shows the evolution in particle mass concentration, 30 density, and the elemental composition, illustrated by the oxygen-to-carbon (O:C) ratio in experiment 2.3, which is conducted at -15 °C and with an initial α-pinene concentration of 50 ppb. InitiallyIn the beginning of the experiment, both the densitymass concentration and O:C ratio increase significantly, but fromafter ~50 minutes onwards, the O:C ratio stabilizes and only slightly increasestend to stabilize while the density slightly decreases. After ~175 minutes the particle mass concentration continues increasing and peaks after ~175 minutes (not corrected for wall loss). Under all conditions (i.e. in all The reported particle 35 mass concentration is obtained using a density of 1.12 g cm -3 . According to Table 1 and supplementary material S12, by the end of the experiments), the AMS derived SOA densities are inof the range 1.1 toorder 1.1-1.3 g cm -3 . Figure S9 however revealsThere are indications of a slight increase in density with higher particle formation and agingexperimental temperature, and as well as a slightly higher density for the particles formed at low α-pinene concentration (10 ppb) compared to high α-6 pinene concentration. (50 ppb). For reproducibility with respect to SOA formation (mass concentration) as well as loss rates of α-pinene and ozone, see Kristensen et al. (2020). Figure 1b is a mass spectrum of Experimentexperiment 2.3 obtained from the high -resolution AMS data at the highest particle mass concentration (not wall loss corrected). It shows that fragments, which belong to the so-called hydrocarbon family (CH)), are distributed throughout the mass spectrum, with some of the most prominent peaks (and ions) being m/z 39 (C3H3 + ), 41 5 (C3H5 + ) + ), and 55 (C4H7 + ). The oxidized compounds, which belong to the CHOCHO1 and CHOgt1 (gt means greater than) families, dominate at m/z 28 (estimated from CO2 + according to Aiken et al. (2008)), 29 (CHO + ), 43 (C2H3O + ), 44 (C2H4O + , CO2 + ), 55 (C3H3O + ), and 83 (C5H7O + ). m/z 43 is the most significant peak and it also has the highest contribution of the CHOCHO1 family, while the more oxidized CHOgt1 family dominates in m/z 44. Previous studies have shown that these two peaks provide useful information about particle oxidation level . The patterns described above are also observed 10 in the mass spectra of the other experiments and, which are overall the mass spectra are highly comparable across experimental conditions (Supplementarysupplementary material S10-S20S13-S23) as well as comparable to mass spectra of particles formed in the α-pinenedark ozonolysis of α-pinene in other chambers (Bahreini et al. 2005;).

PMF analysis of Aerosol Mass Spectra 15
While PMF analysis is traditionally utilized to identify distinct sources in ambient measurements, and the factors are As shown in Figure 2, the relative contributions of the different factors to the observed SOA particle mass concentration vary across different temperatures and α-pinene concentrations (i.e. particle mass loadings). At high temperature, Factor 1 30 dominates whereas Factor 4 is less significant, while at low temperature Factor 4 dominates and Factor 1 is less significant.
Correspondingly, at high α-pinene concentration, Factor 2 dominates and Factor 3 is less significant while the opposite is the case at low precursor concentration. This means, that the profiles of the factors represent characteristics of the particle chemical composition associated with temperature and precursor concentration.
According to their appearance Factors 1 and 4 will in the following discussion be referred to as temperature factors and Factors 35 2 and 3 will be referred to as concentration factors. For each factor, Table 2 provides the oxygen-to-carbon (O:C) ratio, hydrogen-to-carbon (H:C) ratio, as well as the ratio between the absolute intensities of the fragment ions at m/z 43 (C3H7 + , C2H3O + ) and the total organic ion intensity (f43), and the ratio between the absolute intensities of the fragment ions at m/z 44 (C2H4O + , CO2 + ) and the total organic ion intensity (f44). Figure 3 shows the mass spectrumThe mass spectra are colored according to contributions from the various types of elemental 40 compositions (i.e. ion families) that appear at each ion signal. Factor 1 (high temperature) is the factor that is mostly dominated by ions from oxidized species (i.e. high intensity of CHO and CHOgt1 ion groups at m/z 28, 29, 43, 44, 55, and 83). In Factor 1, the intensity of the ions at m/z > 44 is in general low in comparison with the other factors. Among all factors, Factor 1 has the highest O:C ratio (0.56), f43 (14 %), and f44 (9 %) and therefore the chemical species represented by Factor 1 are likely the most oxidized and least volatile of all those present in the SOA. Factor 4 (low temperature) has around the same level of f43, (13 %) as Factor 1 and relative high intensities of m/z 55 and m/z 83, which are the fragment ions larger than m/z 43 that are most intense in the CHO1 ion family. On the other hand, Factor 4 is low in the more oxidized f44 (4 %) and consequently the 5 O:C ratio (0.34) is much lower compared to Factor 1 (high temperature). As expected from the comparison of the mass spectra from the individual experiments (supplementary material S13-S23), the factor profiles show a high degree of similarity ( Figure   3 and S24) with small differences in the relative intensities of ions. Using the method described by , on a scale from 0 to 1, with 1 indicating highest similarity, the similarity is calculated to be between 0.86 and 0.97 across all factors in the m/z range from 12 to 115 with Factor 3 and Factor 4 being the most similar and Factor 1 and 3 being least similar. By 10 focusing only on the m/z > 44, similarities between 0.85 and 0.97 are obtained and Factor 2 and Factor 3 are the most similar and Factor 1 and Factor 2 are least similar. S24 also show the corresponding results of a comparison using the method described by Wan et al. (2002).
Factor 3Table 2 summarizes how the four factors do differ in oxygen-to-carbon (O:C) ratio, hydrogen-to-carbon (H:C) ratio, average carbon oxidation state (OSC, Kroll et al. 2011) as well as the ratios between the absolute intensities of the fragment 15 ions at m/z 43 (C3H7 + , C2H3O + ) and m/z 44 (C2H4O + , CO2 + ), respectively, and the corresponding total organic ion intensity (f43 and f44, respectively). Variations in these parameters can help explain the relative contribution of each factor to the SOA mass under different experimental conditions (i.e. temperatures and α-pinene concentrations (particle mass loadings)).
As shown in Figure  According to their appearance and relative contribution to total SOA mass, Factors 1 and 4 will in the following discussion be referred to as "temperature factors", and Factors 2 and 3 will be referred to as "concentration factors". For example, according to Figure 2, at both α-pinene concentrations Factor 1 makes up a significant fraction of the particle mass in the 20 °C 25 experiments, but plays a minor role in the colder experiments. Therefore, Factor 1 will be referred to as "high temperature factor". The significant contribution of Factor 1 to the SOA mass at high temperature is in agreement with the fact that this factor is mostly dominated by ions from oxidized species (i.e. high intensity of CHO1 and CHOgt1 ion groups at m/z 28, 29, 43, 44, 55, and 83) (Figure 3). Among all factors, Factor 1 has the highest O:C ratio (0.56), OSC (-0.53), f43 (14 %), and f44 (9 %) (Table 2), and therefore the chemical species represented by Factor 1 (i.e. related to high temperature) are likely the most 30 oxidized entities present in the SOA.
Factor 2 (high 3 makes up a significant fraction of the particle mass formed in the 50 ppb α-pinene concentration) has a slightly lower f43 (10 % represents the least oxidized material. It particle mass is noticed thatformed, Factor 3 is practically nonexistent (Figure 2).
Based on this appearance, Factor 2 has relative3 will be referred to as a "high concentration factor". Looking into the profile of Factor 3 (high α-pinene concentration), it has relatively high contributions from the CHO1 family at m/z 55 and m/z 83, and among all factors, it has the highest contribution of m/z 91 from the CH family, which has been used as a tracer of biogenic emissions in ambient measurements (Lee et al. 20162016) (Figure 3). Factor 3 also has the lowest f44 (3 %), O:C (0.26), and OSC (-1.08) among all factors, i.e., it represents the least oxidized material (Table 2).
OverallFactor 4 (low temperature) appears at low temperature in SOA formed in both 10 ppb and 50 ppb α-pinene experiments ( Figure 2). It has around the same level of f43, (13 %) as Factor 1 and relatively high intensities of m/z 55 and m/z 83, which 5 are the fragment ions larger than m/z 43 that are most intense in the CHO1 ion family (Figure 3). On the other hand, Factor 4 is almost as low as Factor 3 in the more oxidized f44 (4 %) and also in O:C ratio (0.34) and OSC (-1.03) ( Table 2).
Generally seen, the factors related to temperature variation (Factors 1 and Factor 4) show a larger difference in oxidation level than the factors related to α-pinene concentration, i.e. particle mass loading (Factors 2 and 3). This suggests that within the investigated conditions, differences in temperature (20 °C to -15 °C) have a larger effect on particle chemical composition 10 than VOC concentration (10 and 50 ppb α-pinene).
It is interesting to notice that the O:C and H:C ratios of Factor 1 (high temperature) and Factor 3 (low concentration) are similar to results presented by  whodecoupled, overall, the four PMF factors represent different main characteristics of the particle chemical composition associated with temperature and VOC precursor concentration, and provide a useful 15 framework for discussing effects of temperature and VOC concentration on SOA formation and properties in chamber experiments. Figure 2 and S7 show that within each experiment, the relative contribution of the factors changes with time.
These changes in relative ratios likely reflect the changes in SOA composition from nucleation (beginning of experiment), condensational growth (increase in mass concentration), and wall loss (decrease in mass concentration towards the end of the experiment). In addition, ongoing gas phase chemistry may also affect observed trends in composition (Kristensen et al., 2020). 20 Furthermore, recent studies (Pospisilova et al. 2020) have shown that particle phase processing continues after condensation, but more work is needed to understand the extent and mechanisms of such processes, and therefore we cannot conclude on such effects.
In each chamber experiment, the correlation between the relative contribution of each factor and the SOA mass concentration 25 can be utilized to infer information about the relative volatilities of the species in each factor. For example, Figures 2 and S7 show that within each of the experiments 1.1 and 1.2, the relative mass concentration ratio of Factor 2 to Factor 1 is largest at lower SOA mass concentrations. This suggests that the volatility of species related to Factor 2 is lower than Factor 1 species which is interesting since Factor 1 is more oxidized than Factor 2 ( Table 2). Figure 2 and S7 show that for each of the experiments 3.2 and 3.3 the trend in the relative mass concentration ratio of Factor 3 to Factor 4 is largest at time periods in 30 the experiment with lower SOA mass concentrations. This indicates that the volatility of Factor 3 species is lower than Factor 4 species. The relative volatilities of Factor 1 and Factor 3 can be assessed by examining the time trends in experiments 3.1 and 3.2 where the fraction of Factor 1 is higher at low mass loading. This together with the fact that the relative ratio of Factor 3 to Factor 1 is higher at lower temperatures suggest that Factor 3 is more volatile than Factor 1. Taken together these results suggest that the volatility (c*) of the four factors increases in the following way: Factor 2 < Factor 1 < Factor 3 < Factor 4. 35 While we do not see any systematic indications of a specific factor being coupled to relative humidity in our experiments, we cannot rule out that the changes in relative humidity during the experiments might have some impact on SOA composition.
As PMF analysis is traditionally used on ambient data, it is relevant to compare the findings from the AURA chamber 40 experiments to ambient studies. Here the analysis presented by  is relevant, as they used PMF analysis to explore the SOA sources in a coniferous forest mountain region in British Columbia., where SOA concentrations reached up to 5 µg m -3 and the temperature varied from ~5 °C to ~25 °C, corresponding to the temperature in the upper range of the experiments presented in this paper. PMF factors obtained from the ambient AMS data showed a background source and two biogenic SOA sources: BSOA1 from terpene oxidized by ozone and nitrate radical during nighttime, and BSOA2 from terpene oxidized by ozone and OH-radical during daytime. Overall, the mass spectrum of BSOA1 from  is highly comparable to the spectra obtained in these experiments. In the BSOA1 factor, the O:C ratio is 0.56 and the H:C ratio 1.56.
These levels have a high similarity with Factor 1 (high temperature) and Factor 3 (low concentration), consistent with the fact 5 that in the ambient measurements by

Elemental analysis
Studying the evolution of the elemental composition of SOA can provide insight into the chemical changes occurring during chemical and physical processes. SOA formed from low α-pinene concentration (10 ppb) and at higher temperature are associated with higher O:C ratios, compared to SOA formed from high α-pinene concentration (50 ppb) at lower temperatures. In all experiments, an initial increase in the O:C ratios, which subsequently level off, is observed -most significantly in the 50 ppb α-pinene experiments 30 (2.1-2.3) probably due to higher reaction rate in these experiments. Although ageing of oxidized organic particles in ambient measurements is associated with an increase in O:C ratio ) at higher particle mass concentrations, the O:C ratio is usually observed to decrease during particle aging Denjean et al. 2015b), because of increased partitioning of less oxidized semi volatile compounds into the particle phase.
In the 10 ppb α-pinene experiment (1.4), where the temperature is lowered 36 minutes after experimental start, the temperature 35 change from 20 °C to -15 °C is associated with a small decrease in the O:C ratio, which corresponds to condensation of less oxidized, i.e. more volatile, species (Figure 4) 1-1.3, 2.1-2.3, and 3.1-3.3) are shown in Figure 4. The slight An important outcome of Figure 4 is that the O:C ratios at the end of the temperature ramps are closer to the O:C ratios of the particles in the experiments conducted at the temperature, where the ramps start than where they end. This observation suggests that the composition of α-pinene derived SOA particles is, to a large extent, controlled by the temperature at which they are 5 initially formed, and that subsequent changes in temperature, even as dramatic as 35 °C during 100-130 minutes, only affect the particle chemical composition to a minor extent. Even though the newly formed particles are exposed to this change in temperature (~35 minutes after experimental start, and ~1 hour before SOA mass peak (Figure 2) Although ageing of oxidized organic particles in ambient measurements is associated with an increase in O:C ratio (Ng et al. 25 2011), in laboratory experiments at higher particle mass concentrations, the O:C ratio is usually observed to decrease during particle aging Denjean et al. 2015b) because of increased partitioning of less oxidized semi volatile compounds into the particle phase. Figure 4b5b shows that during all constant temperature 10 ppb α-pinene experiments, the O:C ratio and the H:C ratio are almost constant.
Interestingly, in the 50 ppb α-pinene experiments conducted at lower temperatures (0 °C and -15 °C, experimentexperiments 30 2.2, 2.3, 3.2, and 3.3)), the H:C and O:C ratios areincrease simultaneously increasing during the experiment. ThisAs this is not a commonly reported trend, neither in ambient measurements

nor in chamber experiments
focusing on α-pinene derived SOA .2011) it demonstrates the importance of investigating SOA particles at low, atmospheric relavant, temperatures. Several mechanisms could potentially explain the observed evolution of SOA elemental composition in the Van Krevelen plot, and in fact, it could be due to a combination of different simultaneous 35 mechanisms, e.g. oxidation and oligomerization. Since no OH -scavenger is added in our experiments, one explanation could be related to OH chemistry;.  demonstrated that exposure of ozonolysis generated α-pinene SOA formed by ozonolysis , to OH-radicals increased, increases the O:C ratio, and also leadleads to higher H:C ratio, because of OH addition to the unsaturated VOC. Modelling suggests, however, that the OH oxidation is not more pronounced at low temperature (0 °C)), compared to high temperature (20 °C) (Quéléver et al., 2019), which makes this a less likely explanation for the 40 continuous increase in the O:C ratio and H:C ratio in the cold experiments. More specifically, the ratio of VOC oxidized by ozone relative to that oxidized by OH-radicals was ~2:1, independent of precursor concentration and temperature (Quéléver et al., 2019).
While the simultaneous increase in H:C ratio and O:C ratio could also be associated with hydration reactions ) withof carbonyls , condensation of water does not influence the elemental ratios derived from the AMS spectra, since the calculation does not directly utilize measured H2O-related ion signals, as they typically have large interferences from gas phase H2O in air . It should be mentioned, that the observed increase in H:C ratio could potentially be due to impurities condensing to the particle phase in the cold experiments, although this seems highly 5 unlikely, as the chamber was cleaned thoroughly before the experimentseach experiment (see Kristensen et al., manuscript submitted. 2020), and the observed changes in the H:C ratios would need an excessive amount of impurities as the particle mass is high (see Table 1, and Figure 2). Finally, since the increase in H:C ratio is only observed in the cold experiments, it is possible that the H:C ratios for the chamber SOA species that condense at the lower temperatures are less accurate, because these species are not well represented in the calibration dataset that was used to formulate the method ) 10 by which the H:C ratio (and O:C ratio) is derived from AMS spectra. the experiments conducted at the temperature where the ramps start than where they end. This observation suggests that the composition of α-pinene-derived particles is, to a large extent, controlled by the temperature at which they are initially formed and that subsequent changes in temperature, even as dramatic as 35 °C during 100-130 minutes, do not change the particle chemical composition significantly. Even though the newly formed particles are exposed to this change in temperature (~35 minutes after experimental start, and ~1 hour before SOA mass peak (Figure 2)), only slight changes in the chemical 35 composition are observed (Figure 5a).

Oxidized organic tracer ions
As described in relation to the mass spectra obtained from the PMF analysis (Figure 3), differences in VOC precursor concentration (i.e., particle mass loading) and temperature primarily result in intensity differences in the dominant oxygencontaining ions, m/z 43 and m/z 44. m/z 43 (dominated by C2H3O + (CHOCHO1 family)) likely derives from organic 40 compounds containing non-acid oxygen , while the signal at m/z 44 (primary CO2 + (CHOgt1 family)) arises from carboxylic acids (Alfarra 2004). Both the number of acid groups and the length and functionalization of the carbon chain in the compounds affect the intensity of the signal at m/z 44 Canagaratna et al. 2015). Figures 6a and 6b are "triangle plots" ), showing f44 (the fraction of m/z 44 relative to the total mass in the spectra) as a function of f43 (the fraction of m/z 43 relative to the total mass in the spectra)), obtained from unit mass resolution data from the AMS. Figure 6a shows the values at the peak of mass concentration (five data point average) of constant 5 temperature experiments (1.1-1.3, 2.1-2.3, and 3.1-3.3) and temperature ramp experiments (1.4 and 1.5)), while Figure 6b shows the evolution through the constant temperature experiments. As observed in the Van Krevelen plots (Figures 5a and   5b), data from the repeated 50 ppb α-pinene experiments (2.1-2.3 and 3.1-3.3,)) conducted at similar temperatures show goodoverall reproducibility, though not being identical.
The triangle plots show that particles formed at higher temperature have a higher f44 (i.e. CO2 + , acid-derived functionalities) 10 than particles formed at lower temperature. No clear tendenciestrends with temperature are observed for f43 (i.e. C2H3O + , nonacid-derived functionalities). Particles formed at lower α-pinene concentration (10 ppb) have higher f44 and a lower f43 than particles formed at higher α-pinene concentration (50 ppb). This suggests that acid-derived functionalities are more prevalent in α-pinene SOA formed at low precursor concentration. (and thus low particle mass loading), which is consistent with less partitioning of the more volatile, less oxidized material to the particle phase. In all experiments, f44 values are between 0. concentration.concentrations. Moreover, the non-evolving f43 at 20 °C is also in agreement with literature exploring α-pinene SOA at comparable concentrations and room temperature . As the increase in f43 is only observed in the cold experiments, especially -15 °C, this suggests that formation of species that give rise to high f43 values are highly temperature dependent.

Estimated particle content of organic acids 25
In AMS mass spectra, m/z 44 has been shown to be a good tracer for the content of organic acids in SOA Yatavelli et al. 2015). Yatavelli et al. (2015) investigated how the mass concentration of molecules (R-COOH) containing one or more acid functionalities, can be related to the AMS derived mass concentration of m/z 44 multiplied by scaling factors. In this work Yatavelli et al. (2015) estimated that 10 to 50 % of the organic particle mass in the northern hemisphere can be attributed to molecules containing the carboxylic acid functionality. Inspired by Yatavelli et al. (2015)), we 30 here explore how the intensity of m/z 44 in the AMS mass spectra compares to the mass concentration of organic acids (R-COOH) and organic acid functionalities (-COOH) obtained by the off-line LC-MS based on results from the off-line LC-MS analysis of filter samples obtained by the end of the AURA chamber experiments. As described in detail in Kristensen et al. For the 50 ppb α-pinene experiments 3.1, 3.2, and 3.3, conducted at 20 °C, 0 °C, and -15 °C, respectively, Figure 7a shows the mass concentration of organic acids (R-COOH) identified from LC-MS analysis (Kristensen et al., manuscript submitted)2020), and the mass concentration of the m/z 44 signal in the AMS mass spectra, scaled to the SMPS mass 40 concentration and corrected for density, as previously described. For both techniques (AMS and LC-MS), the mass concentration of organic acids is lower at higher temperatures. The AMS m/z 44 mass concentrations are lower than the organic acid concentrations obtained from the LC-MS by factors of 2. 55, 4.11, and 4.65 at 20 °C, 0 °C, and -15 °C, respectively. In the following, these numbers will be referred to as scaling factors. For comparison, Yatavelli et al. (2015) reported the m/z 44 AMS signal being a factor of ~2.32 lower than the mass concentration of organic acids in SOA during summertime in a forest area dominated by pine trees near Colorado Springs, USA. Their result is in very good agreement with the scaling factor obtained in the experiment conducted at 20 °C, which supports the hypothesis that the most important organic acids in α-5 pinene SOA are determined by the LC-MS method. The variation in scaling factors at the different temperatures likely reflects that organic acids with different numbers of acid functionality (-COOH) and/or different multifunctional moieties exhibit different degrees of thermal decomposition to the m/z 44 signal in the AMS Yatavelli et al. 2015).
The similarity of the scaling factors obtained in the 0 °C (4.11) and -15 °C (4.65) experiments is consistent with the fact that the SOA chemical composition at those temperatures have a higher degree of comparability relative to the 20 °C experiment, 10 where a lower scaling factor (2.55) is obtained (recall Figures 4 and 6).
Since lower SOA mass is produced at the higher temperatures, it is also relevant to investigate how the mass fractions of organic acids vary with temperature ( Figure 7b). The mass fractions are obtained by dividing the LC-MS and AMS results presented in Figure 7a by the total SOA mass concentration measured in the chamber, prior to the filter sampling, and corrected as described previously. By application of the scaling factors found above, the two techniques are in good agreement, though 15 slight differences appears at 20 °C and -15 °C. While the mass concentration of organic acids (R-COOH) obtained from the LC-MS decreased significantly with higher temperature (Figure 7a), no trend is observed in organic acid mass fractions ( Figure   7b). Interestingly, for m/z 44 from the AMS mass spectra, the temperature dependent trend changes from decreasing with higher temperature (Figure 7a) to increasing when focusing at the mass fraction of m/z 44 to total SOA mass concentration ( Figure 7b). 20 Some of the organic acids as well as the dimers observed from the LC-MS data (Figures 7a and 7b) contain multiple acid functionalities (-COOH) (Kristensen et al., manuscript submitted).2020). Therefore, it is also relevant to investigate how the mass concentration and mass fraction of acid functionalities (from the suggested molecular structures (Kristensen et al., manuscript submitted) (Zhang et al. 2015;Kristensen et al. 20172015). 35 Overall, the comparison of the m/z 44 signal from the AMS mass spectra and SOA acid content obtained from LC-MS data shows that organic acids and organic acid functionalities are important constituents of α-pinene derived SOA, and that it is relevant to investigate and compare different techniques for their quantification.

Conclusion
ChemicalThe chemical composition of α-pinene derived SOA was investigated using HR-ToF-AMS in a series of experiments 40 performed at different -pinene concentrations (10 ppb and 50 ppb) and temperatures (20, 0, and -15 °C respectively)., and ramps in the range 20 to -15 °C and -15 to 20 °C). PMF analysis was applied to a combined AMS dataset representing all of theseeight different experimental conditions. The PMF analysis revealed that the chemical composition of the SOA particles could be described by four factors, which differ in their dependence on VOC concentration and experiment temperature. To our knowledge, this is the first study using PMF analysis on AMS chamber data to reveal distinct factors sensitive to temperature.
This analysis demonstrates that α-pinene SOA oxidation level is dependent on both temperature and -pinene concentration: 5 SOA oxidation level increases with higher temperature and with lower SOA mass loading. The dataset suggests that particles formed at 0 °C are more chemically similar to particles formed at -15 °C than to particles formed at 20 °C. Temperature ramps over a range of 35 °C were only accompanied by slight changes in chemical composition, with increasing oxidation levels during heating ramps and decreasing oxidation levels during cooling-ramps. The investigation demonstrates that the temperature at which particles are formed is decisive for aerosoltheir properties during α-pinene SOA lifetime. This is 10 interesting from an atmospheric perspective as secondary organic aerosol particles are formed and age over a wide range of temperatures.
This work confirms that the particle chemical composition is dependent on precursor concentration and particle mass loading.
From an atmospheric perspective, it is equally interesting that temperature has a high impact on aerosol chemical composition.

Author contributions
The ACCHA campaign was supervised by MB, ME, MG, and HBP. Experiments were performed by LJ, KK, LLJQ, BR, and RT. The Aerosol Mass Spectrometer (AMS) measurements and AMS data analysis, including Positive Matrix Factorization (PMF) analysis, was carried out by LJ and supervised by MC and MB. LJ prepared the manuscript with 20 contributions from all co-authors.

Competing interests
The authors declare that they have no conflict of interest.