1 Reactive oxygen species ( ROS ) emissions and formation pathways in residential wood smoke under different combustion and aging conditions

Wood combustion emissions can induce oxidative stress in the human respiratory tract caused by reactive 15 oxygen species (ROS), either directly or after oxidation in the atmosphere. To improve our understanding of the ROS generation potential of wood combustion emissions, a suite of smog chamber (SC) and potential aerosol mass (PAM) chamber experiments were conducted under well determined conditions for different combustion devices and technologies, different fuel types, operation methods, combustion regimes, combustion phases and aging conditions. The ROS content as well as the chemical properties of the aerosols were quantified by a novel ROS 20 analyzer and a high resolution time of flight aerosol mass spectrometer (HR-ToF-AMS). For all eight tested combustion devices, primary ROS concentrations substantially increased upon aging. The level of primary and aged ROS emission factors (EFROS) were dominated by the combustion device (within different combustion technologies) and to a greater extent by the combustion regimes: the variability within one device was much higher than the variability of EFROS from different devices. Aged EFROS under bad combustion conditions were ~2-80 times higher 25 than under optimum combustion conditions. EFROS from automatically operated combustion devices were on average one order of magnitude lower than those from manually operated appliances, which indicates that automatic combustion devices operated at optimum conditions to achieve near-complete combustion should be employed to minimize ROS emissions. The parameters controlling the ROS formation in secondary organic aerosol were investigated by employing a regression model, including the fractions of the mass spectrometric signatures m/z 44 30 and 43 in SOA (f44-SOA and f43-SOA), the OH exposure, and the total organic aerosol mass. The regression model results of the SC and PAM chamber aging experiments indicate that the ROS content in SOA seems to increase with the SOA oxidation state, which initially increases with OH exposure and decreases with the additional partitioning of semi-volatile components with lower ROS content at higher OA concentrations, while further aging seems to Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2017-1068 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 21 November 2017 c © Author(s) 2017. CC BY 4.0 License.


Introduction
Numerous studies have shown a link between exposure to airborne particulate matter (PM) and worldwide morbidity and mortality (Beelen et al., 2013;Dockery et al., 1993;He et al., 2016), as well as a strong correlation of airborne 40 PM with lung function (Lee et al., 2011;Pope et al., 2002;Adam et al., 2015;Hwang et al., 2015).The adverse health effects of PM are related to the aerosol chemical composition (Kelly and Fussell, 2012;Baltensperger et al., 2008).Residential wood combustion can contribute 5-44 % to the total ambient PM 2.5 (particulate matter with a diameter smaller than 2.5 μm), depending on the environment (Zhang et al., 2010;EPBE, 2005;USEPA, 2000;EEA, 2013a;Ciarelli et al., 2017).In addition to PM, wood combustion emits a wide range of gaseous pollutants, 45 including volatile organic compounds, which upon oxidation can form secondary organic aerosol.Although wood is considered to be a climate neutral source of energy, epidemiological studies suggest that wood smoke may contribute significantly to premature mortality (Boman et al., 2003;Johnston et al., 2012), because of its association with respiratory disease, cerebrovascular diseases and impaired lung function (Liu et al., 2017;Yap, 2008;D. G. Fullerton, 2011).Liu et al. (2017) found a 7.2 % increase in the risk of respiratory hospital admissions during days 50 with high wildfire-specific PM 2.5 compared to non-wildfire smoke event days.Exposure to wood combustion particles may cause moderate inflammatory activity, cell death and DNA damage, and adverse effects to airway epithelia (Krapf et al., 2017;Tapanainen et al., 2012;Muala et al., 2015;Marabini et al., 2017).These adverse effects may be related to oxidative stress caused by free radicals induced by inhaled PM, which overwhelms the antioxidants in the body (Lobo et al., 2010).In turn, free radical formation may be due to reactive oxygen species 55 (ROS) present in atmospheric aerosol, transition metals undergoing Fenton reactions, or redox cycling organic compounds like quinones.The content of ROS in wood combustion emissions is largely unknown, with the contribution from secondary organic aerosol being particularly uncertain.This limits our understanding of the adverse health effects of wood smoke.Acellular assays enable the assessment of particulate ROS by methods that are easily applicable to field and laboratory studies.One such assay monitors the rapid decay of 60 2',7'-dichloorofluorescin (DCFH) to a fluorescent compound (DCF) in the presence of horseradish peroxidase (HRP) (King and Weber, 2013;Fuller et al., 2014b;Huang et al., 2016;Wang et al, 2011).The DCFH assay has been shown to be sensitive towards a broad range of ROS, and to have fast response rates and linear response to varying ROS concentrations, thus being suitable to evaluate the overall oxidative activity of PM (Zhou et al., 1997;Venkatachari and Hopke, 2008;King andWeber, 2013, Zhou et al., 2017).

65
Functionalized species formed from oxidation often have more deleterious effects on human health than parent analogues (Wang et al., 2007;Fu et al., 2012).Therefore, studies of both primary and aged wood combustion aerosols are needed to advance our understanding of their impact on human health.To investigate the aged aerosol products in a controlled and reproducible environment, atmospheric simulation systems such as smog chambers (SC) and potential aerosol mass (PAM) chambers are commonly used.In the present study, a suite of SC and PAM experiments were conducted.As different types of wood, combustion appliances and combustion conditions result in varying levels of emissions (Johansson et al., 2004;Schmidl et al., 2011;Fitzpatrick et al., 2007;Heringa et al, 2011), eight wood combustion devices with variable combustion conditions were tested.Primary and aged biomass smoke generated under different combustion and aging conditions were characterized by an online ROS analyzer based on the 2',7'-dichlorofluorescin (DCFH) assay coupled with an aerosol collector.Observations from this study 75 provide the more detailed evidence of the influence of combustion technology on the oxidative potential of the emitted PM compared to a previous similar study (Miljevic et al., 2010).We also show the variation of the ROS content from primary and aged aerosols under different operation conditions.Further, the contribution of reactive oxygen species to aged organic aerosol generated with different aging tools was investigated to clarify the ROS formation potential upon photo-oxidation.Results from these experiments may be directly compared with ambient 80 measurements.

Experimental setup and methodology
We performed two sets of measurement campaigns, utilizing several wood combustion devices with different combustion conditions and two aging tools.First we present the different devices, then give a description of the PAM chamber and the Paul Scherrer Institute (PSI) mobile smog chamber (PSI-MSC, ~ 7 m 3 ) and the PSI stationary 85 smog chamber (PSI-SSC, 27 m 3 ) (Platt et al., 2013(Platt et al., , 2014;;Paulsen et al., 2005), including the experimental procedures, and finally discuss the combustion conditions and measurement strategy.An experiment schematic is shown in Figure S1.The combustion devices, experiment aging tools, as well as the test aspects are listed in Table 1.

Combustion devices
Eight combustion devices with different technologies were tested, including a pellet boiler (PB, automatic), a 90 moving grate boiler equipped with electrostatic precipitator (MGB, automatic), a updraft combustion pellet stove (PS, automatic), a two-stage combustion downdraft log wood boiler (LWB, manual), two advanced two-stage combustion log wood stoves (LWS1, manual, updraft; LWS2, manual, updraft combustion when cold and downdraft combustion when hot), and two conventional single-stage combustion log wood stoves (LWS3: manual, and LWS4: manual).In the following, we describe the different combustion devices.

95
PB. Automatically operated pellet boiler, with two-stage updraft combustion and a nominal heat output of 15 kW, using wood pellets (EN certified, moisture content 7 %) as the combustion fuel.Under optimum combustion conditions, the ideal air to fuel ratio (λ) is achieved leading to near-complete combustion and, consequently, the particle emissions are dominated by inorganic components which are contained in the pellets.The PB was also altered to enable the variation of the air to fuel ratio to investigate the influence of this parameter on the emissions.

100
In this way, different combustion regimes could be achieved with this device, details are described in Sect.2.2.

MGB.
Automatically operated industrial moving grate boiler with nominal heat output of 150 kW, operated with wood chips (30 % moisture content).The grate has several zones where primary and secondary combustion air can be regulated.for the injection of the combustion air.However, due to a relatively simple air control, the PS is operated at high λ.
We also investigated part load conditions at 3 kW.LWB, LWS1, LWS2, LWS3 and LWS4 are manually operated devices, with the norminal heat outputs of 30 kW, 8 kW, 4.6 kW, 8 kW and 4.5 kW, respectively.Further, the 110 LWS1 is equipped with a storage container for logs, which slide on the grate due to gravity.For all four two-stage combustion devices (PS, LWB, LWS1, and LWS2) and one single-stage combustion device (LWS3), ROS emissions from starting, flaming and burn out phases were investigated (details of the combustion phases are described in Sect.2.3).In the case of the LWS4, only the flue gas from the flaming phase was injected into the smog chamber, where the EF ROS under different aging temperatures of -10 o C and 15 o C were tested.In three of the log 115 wood operated devices (LWS1, LWS2 and LWS3) dry (13-16 % moisture content) and wet logs (24-42 % moisture content) were investigated.In the PS, wheat pellets (manufactured from milling residues, moisture content 9 %) were tested in addition to conventional wood pellets (EN certified, moisture content of 7 %).In the LWS4, beech wood logs with a moisture content of 18 ± 3 % were used.

Combustion conditions
Two parameters are used to describe the combustion conditions namely the combustion regimes and the combustion phases.Combustion regimes are defined by the air fuel equivalence ratio (λ) (Nussbaumer et al., 2000).Depending mainly on the level of excess air three combustion regimes are distinguished: lack of oxygen (λ -), optimum combustion condition (λ opt ), and (high) excess of oxygen (λ ++ ).Each of these is characterized by a different type of combustion particles, i.e., comprising mostly soot, salts, and condensable organic compounds, respectively (Nussbaumer and Lauber, 2010).It should be noted that in wood combustion λ is always > 1.
130 Consequently, λ -and λ ++ only describe λ-values which are clearly (for λ ++ at least 1.5 fold or higher) below or above The three combustion regimes were achieved by changing the air to fuel ratio in the pellet boiler (PB).Optimum combustion conditions (λ opt ) were easily achieved by operating the PB under the designed optimum operation mode.
High excess of oxygen (λ ++ ) compared to λ opt was obtained by additionally blowing air into the combustion chamber 135 via the ignition tube.The lack of oxygen (λ -) regime was obtained by manually closing the secondary combustion air inlet.It should be noted that in real life operation λ ++ and λ -conditions only occur with severe mal-operation.These conditions were investigated since they result in distinct emission characteristics (high NMVOC emissions during λ ++ and high soot emissions during λ - (Nussbaumer and Lauber, 2010).
In the MGB, part load (50 kW) and full load (150 kW) conditions were tested, as well as the influence of an 140 electrostatic precipitator (ESP) installed downstream of the combustion unit.ESPs are widely used in both large and small scale wood combustion devices to reduce PM emissions (Bologa et al., 2011;Nussbaumer and Lauber, 2010).
Combustion phases in the log wood stoves, log wood boiler and pellet stove were classified using the modified combustion efficiency (MCE), defined as the molar ratio of the emitted CO 2 divided by CO plus CO 2 (CO 2 /(CO+CO 2 )), in the flue gas after wood combustion (Ward and Radke, 1993).Each full combustion cycle 145 includes three combustion phases: start phase (beginning of the burning cycle before MCE reaches 0.974), flaming phase (between start and burnout phase, with MCE > 0.974) and burnout phase (after flaming phase, with MCE < 0.974).As mentioned in Sect.2.1, all three phases were obtained in the PS, LWB, LWS1, LWS2 and LWS3.In the PS, LWB and LWS1, experiments started with a cold start, followed by a flaming phase and burn out.In the LWS2 and LWS3, after the first complete combustion cycle starting with a cold start, several full combustion cycles 150 followed by adding new logs into the combustion chamber after the burn out was finished (warm start).In devices where the combustion phases were rapidly changing the ROS analyzer was not able to separate these combustion phases due to a slow response time (~ 8 min).Consequently, the single combustion phases, including the start, flaming and burn out, as well as the combined combustion phases start + flaming or flaming + burn out were used for the ROS analysis.In the LWS4, with which the experiments were conducted in the PSI-MSC (at temperatures of 155 263 K and 288 K), and the PSI-SSC (at a temperature of 288 K), only emissions from the flaming phase were sampled.were sampled through a heated line (473 K), diluted by a factor of ~100-150 using two ejector diluters in series (VKL 10, Palas GmbH), and then injected into the PAM chamber (see Figure S1 in the Supporting Information).

Experimental procedures and aging tools
The original concept of the PAM chamber is described by (Kang et al., 2007).Briefly, the PAM chamber is a single, 0.015 m 3 cylindrical glass chamber, flanked by two UV lamps.Prior to entering the PAM chamber, pure air (1.6 L min −1 , humidified with a Nafion membrane, Perma Pure LLC) used as an OH precursor and a stream of diluted 165 d9-butanol (98%, Cambridge Isotope Laboratories) were merged with the incoming reactant flow.The OH exposure during aging was defined as the integral of the OH concentration over the reaction time, and was calculated from the decay of the d9-butanol, measured by a proton transfer reaction-mass spectrometer (PTR-MS 8000, Ionicon Analytik GmbH) (Barmet et al., 2012).The total flow rate in the PAM chamber was maintained at ~ 7 L min -1 , which was the sum of the flow rates of the instruments and a supplementary flow, resulting in a residence time of 170 approximately 2 minutes.The OH exposure was controlled by adjusting the UV light intensity to obtain different OH concentrations.An outer ring flow (~0.7 L min -1 ), which was discarded, was used to minimize wall losses and the instrument sampled only from the inner flow of the PAM chamber (~6.3 L min -1 ).The temperature in the PAM chamber was around 38 o C due to the lamps.Primary wood combustion emissions were characterized either before or after the PAM chamber when the lights were switched off.Aged emissions were characterized after the PAM 175 chamber with lights on.All the experiments were conducted under OH exposures of (1.1-2.0)×10 8 molec cm -3 h which corresponds to ~ 4.5-8 days of aging in ambient by assuming a mean daily OH concentration of 1×10 6 molec cm -3 .The applicability of the PAM chamber to measure wood combustion emissions has been shown in a previous study (Bruns et al., 2015).

Smog chamber aging 180
The second set of experiments was conducted in the PSI mobile smog chamber (PSI-MSC, ~ 7 m 3 ) at temperatures of 263 K and 288 K, and the PSI stationary smog chamber (PSI-SSC, 27 m 3 ) at 295.5 K.An overview of the experimental setup is also shown in Figure S1.Generally, 3 pieces of dry beech logs, 4 pieces of kindling and 3 fire starters and 9 pieces dry beech logs, 8 pieces kindling and 4 fire starters were combusted in LWS4 for average (2.9 ± 0.3 kg) and high (5.1 kg) load experiments, respectively (details in Sect.2.1).The wood moisture content was 19 ± 2 185 %.Only emissions during the flaming phase with a modified combustion efficiency (MCEs) in the range from 0.974 to 0.978 were sampled.Emissions were sampled for 11-21 min and injected into the PSI-MSC using an ejection diluter, yielding a total dilution factor of 100 to 200.Hydroxyl radical (OH) concentrations in the chamber are controlled by continuous injection of nitrous acid into the smog chamber (after the characterization of the primary emissions as described below in Sect.3.1), which produces OH upon irradiation by UV lights (Platt et al., 2013).

190
The OH exposure was estimated by monitoring the decay of d9-butanol (butanol-D9, 98%, Cambridge Isotope Laboratories) following a single injection before the UV lights were turned on.In all five experiments conducted in the PSI-MSC, the aging time lasted 4.5-6 h.The OH exposure was 2.6-4.8×10 7molec cm -3 h which corresponds to ~1-2 days of aging in ambient by assuming a mean daily OH concentration of 1 × 10 6 molec cm -3 .More details about some of the PSI-MSC experiments of this campaign can also be found in Bruns et al. (2016Bruns et al. ( , 2017)) additional experiment was conducted in the PSI-SSC, with an OH exposure up to 4.0× 10 8 molec cm -3 h, equivalent to ~17 days of aging assuming a mean daily OH concentration of 1 × 10 6 molec cm -3 , extending the aging range beyond the range achieved by the PAM chamber (~1-8.5 days).
The oxidative potential of the aerosol particles was characterized by an on-line ROS analyzer (flow rate: 1.7 L min -1 ) (Zhou et al., 2017).The aerosols particles were collected in a mist chamber type aerosol collector, dissolved into water and mixed with a 2',7'-dichlorofluorescin (DCFH)/horseradish peroxidase solution.The ROS converts DCFH to DCF, which is detected by fluorescence and quantified as nM-H 2 O 2 equivalents.The time resolution of the 210 on-line ROS analyzer was ~ 8 minutes, preventing resolving brief discrete combustion phases.Therefore, different methods were used to calculate the average ROS emissions under different conditions: 1) average (Figure S2a): utilized when the combustion conditions were relatively stable and sufficiently long to yield a stable ROS signal; 2) integrated average (Figure S2b): in cases of variable combustion conditions, the ROS signal was integrated over 215 the measurement period which could include one or several phases from the same burn; 3) extrapolation + integrated average (Figure S2: panels 2c_1 and c_2): when the combustion conditions were variable and the background could not be measured between two combustion conditions due to the time resolution of the ROS instrument.We extrapolate each measurement to the background value and then make the integrated average calculation as described above.

220
The various definitions for ROS and related aerosol characteristics are presented below:

ROS emission factors (EF ROS )
. ROS emission factors (EF ROS ) were calculated as the amount of ROS in nmol-H 2 O 2 equivalents per kilogram wood burnt, using Eq. ( 2): Secondary organic aerosol (SOA) and secondary ROS (ROSs) were calculated by subtracting primary organic aerosol (POA) and primary ROS (ROSp) from the total OA and aged ROS, respectively, assuming ROSp and POA to be only lost to the chamber wall at the same rate as eBC but otherwise to remain constant during aging.Although both quantities may not be conserved, a decrease of both does partially compensate in the ROS fraction.In the SC experiments, POA is defined as the OA mass before lights on, while SOA is estimated as the difference between 240 total OA and the time dependant POA mass accounting for particle wall loss.Wall loss rates for POA and SOA were assumed to be the same as that of the measured eBC.In PAM aging experiments, each experiment had a certain POA (measurements before PAM or after PAM with lights of) and SOA (measurements after PAM with lights on).
where Org 44-SOA is the difference of total Org44 and primary Org44, Org 43-SOA is the difference of total Org43 and primary Org43 and using the same procedure as for the SOA calculation mentioned above.

Wall loss correction.
The wall loss correction in the SC was done by assuming the same losses for all particle components as for the inert tracer eBC.The wall loss corrected concentration of OA or ROS (X) can be derived 250 using the equation Eq. ( 5): where   () refers to the concentration of X measured at time t.BC (t 0 ) and BC (t) are the concentrations of 255 BC when lights were switched on and at time t, respectively.

Gas phase characterization
During the PAM chamber experiments, total volatile organic compounds (VOC) and CH 4 (using a flame ionization 3 Results and discussion

Primary and aged ROS emission factors (EF ROS )
The ROS and OA emission factors are presented in Table 2 for all combustion condition, together with the number 270 of tests, the combustion efficiency (MCE), the air to fuel ratio (λ), and the aerosol bulk properties determined with the AMS (OM:OC, O:C and H:C ratios).The given values are the 25 percentile and 75 percentile of averages from several experiments and the data points considered for the calculations were restricted to the time period of the ROS measurements.As shown in Fig. 1, ROS emission factors (EF ROS ) for primary and aged OA were highly variable depending on the combustion conditions and devices.For all devices and combustion conditions, a substantial 275 enhancement in the EF ROS is observed with aging, indicating the importance of secondary ROS production.The ROS enhancement factor, defined as the ratio between aged and primary EF ROS , range between 4 and 20, with lower values for MGB (~ 4) and PB under λ opt combustion condition (~ 6), and higher values for PB under λ -and λ ++ combustion conditions (> 10).The ROS enhancement factors for all log wood stoves as well as LWB are comparable, with an average value around 10.

280
The variability in the EF ROS in primary and aged OA for one device is much higher than the variability between average emission factors for different devices, spanning almost two orders of magnitudes.Despite this, EF ROS from PB and MGB (80-8 890 nmol kg -1 wood and 2 440-1.83×10 5 nmol kg -1 wood for primary and aged emissions, respectively) are on average one order of magnitude lower than those from PS, LWB and LWS1-4 (220-1.89× 10 6 nmol kg -1 wood and 3 570-1.1×10 6nmol kg -1 wood for primary and aged emissions, respectively).These results

285
clearly indicate differences due to the combustion technology, as a general rule, EF ROS were lowest for automatically operated devices and higher for manually operated devices: PB and MGB are automatically operated and the primary and secondary air supply as well as the fuel feeding is controlled permanently, while LWB and LWS1-4 are manually operated.The PS is automatically operated but is operated at high λ and exhibits similar EF ROS to the manual devices.Part of the EF ROS variability within each device can be ascribed to the combustion phase, with 290 higher emission factors for the starting and burn out phases compared to the flaming/stable phase.This is especially true for the aged emissions from the PS (EF ROS of the start phases was on average 13 times higher than the flaming phase; Mann-Whitney p-value = 0.06), the LWS2 (EF ROS of the start phases was on average 1.7 times higher than the flaming phase, Mann-Whitney p-value = 0.24, not significant) and the LWS3 (EF ROS of the start and burn out phases were on average 1.5 times higher than the flaming and flaming + burn out phase, Mann-Whitney p-value = 295 0.07).For the automatically operated MGB, the primary EF ROS did not statistically differ between part and full load operation (Mann-Whitney p-value = 0.95).However, the aged EF ROS was a factor of ~3 higher for part load than for full load (Mann-Whitney, p-value = 0.23).The use of the electrostatic precipitator had little effect on the primary and aged ROS emissions, with the differences being within burn-to-burn variability.

300
For PB, the combustion operation could be systematically varied to investigate the influence of air to fuel ratio on ROS and OA emission factors before and after aging.The EF ROS were highest under λ ++ condition for both primary and aged emissions, with average values of 4 100 and 5.8 × 10 4 nmol kg -1 wood burnt, respectively (Fig. 1 and Table 2).Primary ROS emissions under λ opt conditions did not statistically differ from λ -conditions (Mann-Whitney, p-value = 0.43), but on average 7 and 3 times lower than that obtained under λ ++ condition, respectively 305 (Mann-Whitney, p-value < 0.005 for both cases).The aged EF ROS under λ opt and λ -were also quite similar (Mann-Whitney, p-value = 0.20), but with average values 8 and 5.5 times lower than obtained under λ ++ conditions, respectively (Mann-Whitney, p-value = 0.02 for both cases).This shows that the air to fuel ratio has a significant effect on the ROS emissions, which will be investigated for all devices hereafter.

EF ROS under different combustion regimes
Fig. 2 shows the aged EF ROS of the eight devices as a function of λ.Similar to PB, as already described above, a clear increase of EF ROS in the aged aerosol can be observed with increasing λ values, with ~2-80 times higher 335 aged EF ROS values under bad combustion conditions than under optimum combustion conditions, although the extend of the increase and the overall trend were not the same for all individual devices.In the MGB all the burns occurred at 2.0 < λ < 2.2, leading to aged EF ROS in line with those from the PB between λ opt (λ = 1.6) and λ ++ (λ ranged from 2.7-3.4).The combustion in all stoves (PS, LWS1-4) exhibited higher λ (λ > 2.2) due to a less controlled air supply leading to less efficient combustion.In this excess of oxygen range, aged EF ROS 340 ranged between 1.68×10 4 nmol kg -1 wood and 1.38 × 10 6 nmol kg -1 for λ values between 2.2-17.6,where all aged EF ROS were high but without any systematic trend with λ, suggesting that other parameters may influence ROS emissions as well.The LWB follows a different trend, where the aged EF ROS increase sharply with λ, starting at lower λ values than the other manually operated devices.Aged EF ROS for LWB ranged from 3530 to 5.79 × 10 5 nmol kg -1 wood within the λ-range of 1.5-2.6.Although trends in Fig. 2 show differences between 345 devices, they highlight quite readily the important influence of the combustion conditions on aged EF ROS .While the combustion efficiency was found to have a strong influence on aged EF ROS , the latter varies considerably, by a factor of 3-50, within the same combustion regime but for different combustion devices.In   The organic aerosol mass may influence the fraction of ROS in SOA, by affecting the amount of condensing semi-volatile species, which might be characterized by different f ROS-SOA compared to low-volatility species 385 dominating at low organic aerosol mass.The aim of the multiple regression analysis used here is to extract the influence of different aging factors on the observed variance in f ROS-SOA (the 2.6 factor variance described in Fig. 3), and to assess the magnitude of their influence.We do, however, not seek to propose using the model and the model coefficients for a deterministic explanation of ROS formation.increasing the ROS content, especially in the beginning of the experiment.Therefore, the actual effect of OH exposure on f ROS-SOA could only be revealed when it was isolated from the f 44-SOA effect (see Fig. 4).
The analysis suggests that f 43-SOA and the organic mass concentrations exhibit a low, but statistically significant, effect on f ROS-SOA (Fig. S5).Their increase results in a decrease in the secondary ROS content, consistent with 430 the increased partitioning of moderately oxygenated components, which seem to contain less ROS.
Comparison between SC and PAM chamber data.The conditions in the PAM chamber are different from those in the SC.PAM chamber experiments were conducted at high OH exposures of ~10 8 molecules cm -3 h, where the resulting aerosol was highly oxygenated.However, the secondary ROS content of the aerosol in the PAM chamber was largely within the expected range, following consistent trends with high OH exposures and high 435 f 44-SOA as in the SC (Fig. 4).We examined in more detail whether the regression model parameters obtained from the SC could faithfully represent the f ROS-SOA measured in the PAM chamber.Indeed, the model was capable to predict, within uncertainties (2σ), the f ROS-SOA measured in the PAM chamber for low organic aerosol concentrations (average 21 µg m -3 ), but considerably (factor of three on average) overestimated f ROS-SOA at higher concentrations (average 68 µg m -3 ).This is because such a range of concentrations at high OH exposures 440 and high f 44-SOA was not included in the training dataset, and as a result the model slightly underestimated the effect of OA concentration on f ROS-SOA (e.g., a three-fold increase in OA concentration in the PAM chamber results in a decrease of f ROS-SOA by 45 %, while the model suggests that the same increase would only result in a 10 % decrease).Despite this, for similar conditions f ROS-SOA measured in the PAM chamber and the SC were similar within our uncertainties.We also note that this slight bias does not affect the main conclusions of the 445 analysis: the secondary ROS content seems to initially increase with the SOA oxidation state, which increases with OH exposure and decreases with the additional partitioning of semi-volatile components with lower secondary ROS content at higher SOA concentrations, while further aging seems to result in a decay of ROS.

Summary and Conclusions
In this study, eight wood combustion devices for log wood, pellets and wood chips, denoted as log wood boiler The variability in the EF ROS in primary and aged OA for a single device was much higher than the variability between emission factors from different devices.A part of this variability within each device could be ascribed to the combustion phase, with higher emission factors for the starting and burn out phases compared to the 470 flaming phase.This was especially true for the aged emissions from the PS, LWS2 and LWS3.Despite this, EF ROS from the PB and MGB were on average one order of magnitude lower than those from the PS, LWB and LWS1-4.This indicates that applying automatic combustion devices operated at optimum conditions, to achieve near-complete combustion, is most effective to minimize ROS emissions.Although the EF ROS showed somewhat different trends between devices with varying λ, a clear increase of EF ROS in the aged aerosol can be 475 observed from optimal to high lambda values; this emphasizing the important influence of the combustion conditions on EF ROS .For the PB, the EF ROS under λ opt (λ = 1.6) did not statistically differ from that under λ -(λ ≈ 1.3) conditions for both primary and secondary emissions (Mann-Whitney, p-value = 0.43 and 0.20, respectively).When comparing the EF ROS under λ opt and λ -conditions with λ ++ (2.7 < λ < 3.4) condition, primary EF ROS under λ opt and λ -conditions were on average 7 and 3 times lower than that obtained under λ ++ condition, 480 respectively (Mann-Whitney, p-value < 0.005 for both cases).Aged EF ROS under λ opt and λ -conditions were on average 8 and 5.5 times lower than obtained under λ ++ condition, respectively (Mann-Whitney, p-value = 0.02 for both cases).In the MGB all the burns occurred at 2.0 < λ < 2.2, leading to EF ROS in line with those from the PB between λ opt (λ = 1.6) and λ ++ (where λ ranged from 2.7-3.4).The combustion in all stoves (PS, LWS1-4) exhibited higher λ (λ > 2.2) due to a less controlled air supply leading to a lower combustion temperature and 485 increased products of incomplete combustion (less efficient combustion).In this range of oxygen excess, all aged EF ROS were high but without any systematic trend with λ, suggesting that also other parameters influence ROS emissions.We further revealed that this variability was related to the bulk OA emissions, implying that this variation is inherent to the combustion conditions.
Nonetheless, the ROS content still varied by a factor of 2.6 on average for the same OA emission factor (EF OA ).

490
We used a regression model on the data of SC and PAM chamber aging experiments to identify the different parameters that control the ROS secondary formation and content in OA upon aging.This regression model showed that the ROS contents in SOA (represented as f ROS-SOA ) depended significantly on all the aging parameters investigated, including the fractions of m/z 44 and m/z 43 in SOA, f 44-SOA and f 43-SOA , respectively, the OH exposure and the organic aerosol mass concentration.The greatest share of explained variability in f ROS-SOA 495 was attributed to f 44-SOA, which indicates that the more oxygenated compounds are preferentially ROS active compared to others.The OH exposure was the second most important parameter controlling the aerosol ROS content under our condition where the anti-correlation between OH exposure and f ROS-SOA indicated that initially formed ROS are prone to further reactions.The organic mass and f 43-SOA exhibited a low, but statistically significant effect on f ROS-SOA .In summary, the ROS content seems to increase with the SOA oxidation state, 500 which increases with OH exposure and decreases with the additional partitioning of semi-volatile components with lower ROS content at higher OA concentrations, while further aging seems to result in a decay of ROS.
2) where [  ] is the background-corrected concentration of ROS (nmol m -3 ) in the emitted particles either before 225 (primary ROS) or after aging (aged ROS), [  ] are the carbon mass concentrations calculated from the background-corrected, carbon-containing species where x includes CO 2 , CO, CH 4 , volatile organic compounds (VOC), eBC, and particulate organic carbon (OC).M C is the carbon mass burnt and   represents the average Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1068Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 21 November 2017 c Author(s) 2017.CC BY 4.0 License.carbon fraction of the wood fuel, ~ 0.46, measured in this study using an elemental analyzer.OC data were obtained from AMS measurements.Similarly, the organic aerosol (OA) emission factors (EF OA ) were calculated by replacing 230 the ROS concentration by OA.ROS fraction.In order to study the ROS formation during aging, the secondary ROS fraction (f ROS-SOA ) is introduced.It expresses the amount of secondary ROS (ROS S = aged ROS -primary ROS) per amount of secondary organic aerosol (SOA) formed during aging and as calculated from Eq. (3)

f 44 -
SOA and f 43-SOA .To express the degree of oxygenation of SOA, the fraction of secondary Org44 and Org43 in SOA (represented as f 44-SOA and f 43-SOA ) is introduced, which is calculated from Eq.

*
Values of each parameter are described as[a, b], where a and b represent the 25 th and 75 th percentile of the averages from several experiments and the data points considered for the calculations were restricted to the time period of the ROS measurements Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1068Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 21 November 2017 c Author(s) 2017.CC BY 4.0 License.

Figure 1 .
Figure 1.ROS emission factors (EF ROS ) for all tested combustion devices under different operating and aging conditions.Open circles symbols represent the average values of all the 330

Figure 2 .
Figure 2. Aged ROS emission factors (EF ROS ) from different combustion regimes and combustion devices.The grey dashed line represents the EF ROS increase with λ for the PB.The error bars of the y-axis of the data points denote the propagation of the uncertainty ( = √ 1 2 +  2 2 , with  1 and  2 representing the standard deviation of the averaged aged ROS and aged

Fig. 3 ,
Fig.3, we investigate to which extent this variability in aged EF ROS is related to the variability in the bulk OA 355

Figure 3 .
Figure 3. Aged ROS emission factors vs. aged OA emission factors.Marker color correspond to the air to fuel ratio (λ).Fitting equation: log 10 (EF ROS ) = 0.92log 10 (EF OA ) indicating that the relationship between aged ROS and aged OA is almost linear.The geometric standard deviation obtained from the fit is 2.6, suggesting that the aged ROS content of aged OA may 365

Figure 4 :
Figure 4: Variation of the fraction of ROS in SOA, f ROS-SOA , with the fraction of m/z 44 in the total signal SOA as measured by the AMS (f 44-SOA ) color coded with the OH exposure estimated from the decay of d9-butanol measured by the 370 Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1068Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 21 November 2017 c Author(s) 2017.CC BY 4.0 License.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1068Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 21 November 2017 c Author(s) 2017.CC BY 4.0 License.The comparison and evolution of ROS with different combustion and aging conditions in this study might be used for a speedy assessment of potential health risks of wood combustion emissions from different combustion and aging conditions.

Table 1 .
Overview of combustion devices and test aspects.

Table 2 .
Characterization of primary emissions from PAM chamber and SC aging experiments* 325