Building envelopes impact indoor air quality in three general ways: (i) primary emissions of air pollutants from materials used in the enclosure to the interior space, (ii) incidental removal of outdoor air pollutants that cross the enclosure through reactive deposition of oxidants (such as ozone, or O 3 ) or ltration of particles, and (iii) secondary emissions of air pollutants resulting from reactions between inltrating oxidants (e.g., O 3 ) and materials used in the enclosure. Whether the impact of the building enclosure on indoor air quality is benecial or detrimental to the exposure of occupants is dependent on a large number of factors, including (but not limited to) material selection, environmental conditions, airow characteristics, and the quantity and type of indoor and outdoor air pollutants present.
Through emission, removal, and transformation, building enclosures impact the balance of indoor volatile organic compounds (VOCs) and O 3 . Volatile organic compounds are a well-studied class of indoor air pollution, and VOCs oen exceed chronic health standards indoors. 1 Ozone is a major driver of indoor chemistry and one of the most studied oxidants in indoor air. 2 Elevated concentrations of outdoor, ground-level O 3 are consistently associated with increases in a number of adverse health effects including mortality, [3][4][5] exacerbation of asthma symptoms, 6 and infant respiratory and cardiovascular effects. 7 In 2005, ambient O 3 was estimated to account for $4700 deaths and $36 000 years of life lost in the United States alone, suggesting that despite signicant improvements in outdoor air quality in recent decades, levels of outdoor O 3 still pose a risk to public health. 8 Primary emissions of VOCs from materials in building enclosure cavities contribute substantially and directly to indoor VOC concentrations. [9][10][11] Building enclosure materials also act as a "hidden" transformation pathway as inltration air enters a building. For example, in the vast majority of residences in the U.S., which typically do not have mechanical ventilation systems with dedicated outdoor air supply, 12 occupants are exposed to O 3 and O 3 reaction byproducts (including VOCs) only aer O 3 -laden air penetrates through leaks in the building enclosure. [13][14][15] In U.S. residences, limited data suggests that windows are seldom open in most climates, less than 15% of the time in most cases. 16,17 Therefore, inltration across the building envelope is oen the primary path by which O 3 and O 3 -building enclosure reaction byproducts enter occupied residential spaces. 18 Cracks and gaps in the building enclosure where inltration occurs create the potential for O 3 chemistry with interior enclosure materials, such as exterior cladding, insulation, and structural materials, depending on the reactivity of the materials used and the nature of crack geometries. 15,19,20 Reactions within the building enclosure can serve to reduce the amount of outdoor O 3 that transports indoors through surface chemistry that alters the balance of O 3 and may generate harmful or irritating O 3 reaction byproducts found indoors.
To date, O 3 penetration factors have been measured in only a very limited number of buildings. In a sample of eight homes in Austin, TX, the rst measurements of O 3 penetration factors (measured at an articially high indoor/outdoor pressure difference) ranged from as low as $0.6 to as high as $1.0. 21 Subsequent measurements of O 3 penetration factors in a multifamily apartment unit during natural inltration conditions revealed a mean value of only 0.54. 22 These data suggest that most homes relying on inltration for ventilation air likely have O 3 penetration factors lower than unity (i.e., P O 3 # 1). In homes under these conditions, as much as 40-50% of total outdoor O 3 loss occurred because of reactions within the building enclosure, offering substantial protection from indoor ozone exposure, but with implications for subsequent exposure to byproducts of O 3 reactions within the building enclosure that may be transported indoors.
If O 3 reactions occur primarily at the enclosure, indoor exposures to inltrated O 3 are likely lower than if reaction losses occur primarily with indoor materials and reactive gasphase compounds. Indoor exposure to O 3 reaction byproducts formed from homogeneous or heterogeneous indoor O 3 chemistry are also likely to be different, due to the distinct materials that are present in building enclosures versus in interior occupied spaces. Some known byproducts of indoor O 3 reactions with common surfaces and/or gases in typical indoor environments include organic acids, carbonyls, free radicals, and secondary organic aerosols, [23][24][25][26] all of which can yield varied inammatory responses in humans. Understanding routes of indoor O 3 removal and transformation are warranted given that O 3 oxidation products may be partially responsible for the health impacts of ambient O 3 observed in epidemiology studies. 16 Ozone reaction probabilities of typical building enclosure materials available in the literature range several orders of magnitude, from $10 À4 for brick to $10 À8 for aluminum. 15 Key material properties that inuence O 3 reaction probability are porosity, thickness, and composition. 27 However, there exists very limited data in the literature reporting O 3 reaction probabilities to materials used inside building enclosures (e.g., various insulation types, soundproong, etc.) and little data reporting the emissions of volatile byproducts stemming from these materials in the presence or absence of O 3 . Therefore, this study evaluates primary VOC emissions, O 3 reaction probabilities, and O 3 reaction byproduct yields for eight common, commercially available building insulation materials to further understanding of how insulation selection in building enclosures may affect indoor air through these mechanisms.
Eight commercially available insulation materials that are commonly used in building enclosures were selected to span a wide range of chemical composition and physical properties: berglass, cellulose, stone wool, recycled denim, polystyrene, polyurethane spray foam, polystyrene with a thermal backing, and polyisocyanurate foam. An overview of properties of the test materials is provided in Table 1. Images of each tested material in its raw form and as prepared for testing are provided in the Table S1 in the ESI. † All tested materials were purchased new. A sub-sample of each material was randomly taken from the larger quantity of each purchased product. Samples were made to accommodate placement in a benchtop scale environmental chamber (see Section 2.2) for measurement of VOC emissions, O 3 dry deposition, and reaction byproducts emitted aer being exposed to O 3 . Materials were of varying morphology and bulk structure and were prepared to ensure a known, projected surface area was exposed to the bulk chamber air. Materials made of loose ll (cellulose, berglass, stone wool, denim) were placed in a glass enclosure (Pyrex) with taped edges. Solid materials were cut with the sides and backing sealed with aluminium tape to expose only the top surface to the chamber environment.
Samples were tested for source and sink behaviour in a 11.4 L laboratory chamber apparatus; detailed descriptions of a similar experimental apparatus are available in Gall and Rim. 28 Briey, the experimental apparatus was an electropolished stainless steel chamber (CTH-24, Eagle Stainless) in which owrate, temperature, and humidity conditions were controlled to maintain environmental conditions. Air was supplied by laboratory compressed air supply and passed through a particle lter and granular activated carbon lter to remove particles and volatile organics in inlet air. Air was humidied to a setpoint by control of two ows, one passed through an impinging column lled with puried water and a second bypass ow. Ozone was injected into dry air (bypass ow) using a stable O 3 generator (97-0067-01, UVP). Chamber temperature was controlled by circulating the outow of a temperature-controlled water bath (Neslab RTE 10, Thermo Scientic) through vinyl tubing wrapped around the exterior surfaces of the chamber.
All ows were controlled and measured using mass ow controllers (GFC17A, Aalborg). Inlet and chamber temperature and relative humidity were measured via sensors (S-THB-M-002, Onset) inserted into the chamber inlet line and through a septum in a chamber access port, respectively; the sensor which protruded slightly into the chamber was included as part of the chamber background. Ozone was monitored using a UV absorbance federal equivalent method instrument (106-L, 2BTech). The chamber was operated at a owrate of 1.88 L min À1 , and target conditions of inlet O 3 , temperature, and relative humidity of 100 ppb, 22 C, and 50% RH. Actual chamber conditions were (mean across all experiments AE 1 s.d.) 100.2 AE 5.2 ppb, 21.95 AE 0.91 C and 50.9 AE 0.51% respectively. An inlet O 3 concentration of 100 ppb was chosen to represent a high, albeit realistic, outdoor O 3 concentration that exterior building enclosures are subjected to in a high ambient O 3 environment.
Primary emissions of volatile organic compounds (VOCs, primary emissions calculated in the absence of O 3 ) and byproduct formation yields (VOCs emitted due to O 3 surface ux) were calculated with established methods (Lamble et al. 29 ) using concentration data measured via proton transfer reaction-time of ight-mass spectrometry (PTR-TOF1000, Ionicon). The PTR-TOF-MS scanned across 17-250 amu for compounds with a proton affinity higher than that of H 2 O. The dri tube conditions were 600 V, 60 C, and 2.28 mbar. The PTR-TOF-MS was operated at an E/N value of 130. The mass axis was calibrated to three peaks: NO (m/z ¼ 29.9974), C 3 H 7 O + (m/z ¼ 59.0497) and a C 6 H 4 I 2 fragment (m/z ¼ 203.944) that was continuously injected into the dri tube via a heated permeation device (PerMaScal, Ionicon). The PTR-TOF-MS inlet was maintained at 60 C and the supplemental inlet ow to the dri tube was 50 mL min À1 . Mass spectra were stored in 10 s intervals.
Compounds were rst identied using a peak table resolved with unit mass resolution. This method was selected given the limitations associated with the mass resolving power of the PTR-TOF1000 (m/Dm $ 1000). Ions known to be associated with instrument operation were removed from the analysis (m/z 29, 30, 32, 34, 36, 37, 38, 39, 46, 50). 30 Compounds meeting a threshold of statistically signicant difference due to the material or material and O 3 presence (Section 2.5.3) are rst identied according to their unit mass. For observed primary uxes and yields, the ve largest contributors to ux or yield are assigned an exact mass from the measured mass, determined from the centroid of the peak of interest by manual inspection of mass spectra (PTR-MS Viewer 3.2.12, Ionicon). The ve largest contributors to ux are discussed in terms of their putative chemical identication (ID), determined from evaluation of potential chemical formulas that may result in the exact mass of each signal, analysis for presence of expected isotopes for a given chemical ID, and a review of the literature for compounds expected to be emitted from test materials in the presence or absence of O 3 .
Following assignment of the ve largest uxes or yields, the remaining mass was allocated into groups based on unit mass by carbon and hydrogen containing compounds (C x H y ) and oxygenated compounds containing one (C x H y O) and two oxygen atoms (C x H y O 2 ). The approach was similar to that of Inomata et al.; 31 compounds were classied by attribution of a general chemical formula based on the known series of families of Volatile organic compounds were quantied with the PTR-TOF-MS using a relative transmission method similar to methods described elsewhere. 32 A transmission curve was generated using eight calibration compounds spanning protonated mass of 33 to 135 (methanol, 1,3-butadiene, methyl vinyl ketone, benzene, toluene, p-xylene, 1,3,5-trimethyl benzene, 1,2,3,5-tetramethyl benzene). Calibration compounds were diluted to 100 ppb from an initial nominal mixing ratio of 2 ppm from a compressed gas cylinder (Airgas) using a dilution system that mixed a known owrate of zero air (Airgas) with a known owrate from the compressed cylinder containing gas standards. The relative transmission method requires an estimate of the mixing ratio of H 3 O + isotope, typically approximated from the mass signal of the isotope at m/z 21.022. We determined this parameter using the generated transmission curve from the data analysis soware for the instrument (PTR-MS Viewer 3.2.12, Ionicon) and calculating a best-t value of transmission at m/z 21.022 that resulted in the minimization of the sum of squared errors between the reported concentration of the eight compounds used in generation of transmission curve and the known concentrations from the diluted calibration standard. Using this method, reported concentrations for all compounds in the calibration curve were estimated with the transmission method to within 20% of calibration value. For compounds present in the calibration standard, reaction rate constants were taken from Zhao and Zhang. 33 For other compounds, quantication was made using the transmission factor and the Ionicon default reaction rate constant of 2 Â 10 À9 cm 3 per s per molecule. Aer calculation via the transmission curve, quantication of the ve largest primary emission uxes or yields was corrected for isotopologues by correcting the major signal for its contribution to the total mass of the compound. 34 Reported concentrations for the ve largest primary emissions sources or sinks and yields are corrected by manual inspection of peak assignment, correction for known isotopic interferences when greater than 1%, and deconvolution of overlapping peaks where instrument resolution is sufficient (e.g., separation of protonated methanol (m/z 33.0335) from one oxygen-17 isotopologue of O 2 + (m/z 32.9971)). These analyses were all performed in PTR-MS Viewer 3.2.12 (Ionicon).
Prior to initiation of experiments, the stainless-steel chamber surfaces and test material glass enclosures were cleaned and passivated to reduce background reactions between O 3 and chamber surfaces. The chamber and enclosures were rst cleaned with water and soap and rinsed three times. Surfaces were rinsed with reagent grade isopropyl alcohol followed by methanol, followed by hexane (all compounds Sigma-Aldrich, 98% or greater purity). Surfaces were allowed to dry in a fume hood overnight and were then heated with a heat gun. The day before an experiment, the chamber was passivated by introducing elevated O 3 (>500 ppb) into the chamber for at least 10 hours.
The experimental protocol included measurement of VOCs and O 3 levels in the inow and outow of an empty chamber (background) before and aer experiments where materials were tested in the presence or absence of O 3 . Volatile organic compounds and O 3 levels were measured at the chamber outlet or inlet following the timeline described in Fig. 1. The duration of each sample period was selected by calculation of the predicted time to reach steady-state in a non-reactive chamber. It was calculated that an unreactive chamber would reach 99% of steady-state O 3 levels aer $30 minutes; surface reactions will reduce this time to reach steady-state. The steady-state condition was conrmed for each experiment by evaluating if chamber O 3 levels deviated more than 2 ppb in the nal 20 minutes of each chamber monitoring period, a similar criteria to prior studies. 35 All experiments met the steady-state criteria for chamber O 3 levels. Following the completion of each experiment, the chamber was prepared for the next experiment by passivating the chamber overnight.
2.5 Data analysis 2.5.1 Quantication of volatile organic compound source and sink strength. Primary emissions of VOCs are those compounds emitted due to the presence of the material itself, and in the context of this study, in the absence of O 3 . Primary emissions from the test samples were calculated according to eqn (1):
where l is the air exchange rate (s À1 ), V is the volume of the stainless steel chamber (cm surface. 36 The molar yield was calculated following the method described by Lamble et al., 29 shown in eqn (2):
where The steady-state O 3 deposition velocity is calculated as described previously 37 and shown in eqn (3):
where C inlet and C outlet represent the O 3 concentrations in the inlet and outlet air ow of the chamber, respectively (ppb), and v d and v d,BG are the O 3 deposition velocities for test material and chamber, respectively (cm s À1 ) and all other terms as described previously.
Background O 3 deposition velocities (v d,BG ) are calculated by performing an experiment with an empty chamber for a xed air exchange rate until steady-state O 3 concentrations are achieved. Inlet and outlet concentrations of O 3 averaged over the nal 20 minutes of data collection are used to solve eqn (3) for v d,BG when v d ¼ 0 and there is no exposed test material area (A e ). To measure the deposition velocity to the insulation material (v d ), the test procedure is repeated for experiments with insulation materials placed in the chamber, and eqn (3) is solved for the unknown values of v d . An estimate of uncertainty was calculated using a propagation of errors, incorporating uncertainties of the O 3 monitors of 2% of reading and ow controllers of 1.5%.
Ozone deposition was further parameterized by determining the material reaction probability (g, dimensionless), or the fraction of O 3 molecule-surface collisions that result in a reaction. To calculate the reaction probability, the transport limited deposition velocity (v t , cm s À1 ) was rst determined by applying potassium iodide to surfaces using previously described protocols 27 and g was calculated using eqn (4) as described by Cano-Ruiz et al. (1993):
where hv b i is the Boltzmann velocity, and is equal to 3.6110 cm s À1 for O 3 at 22 C. As shown in Table 1, we determined the transport-limited deposition velocity experimentally for four different materials. For materials for which v t was not directly calculated, we assigned a v t value for the material with the most similar surface morphology (see Table 1), similar to the approach taken by Lamble et al. 29 Uncertainty in reaction probabilities was calculated from a propagation of errors from experiments conducted to determine v d,BG , v d , and v t .
2.5.3 Statistical testing for signicant VOC emissions. Data generated by the PTR-TOF-MS, when initially analysed with unit mass resolution, resulted in time-series mass spectra with >200 peaks each. These spectra required subsequent analysis for identication of peaks with signicant differences due to material presence (primary emissions) or material and O 3 presence (molar yield). For statistical testing to identify primary emissions, we selected 100 steady-state time series data points from each of the test conditions of empty chamber outlet and chamber with material outlet. Statistical signicance was determined by comparing datasets for each mass unit with a ttest with a ¼ 0.05.
For molar yields, comparisons required three groups: the empty chamber in the presence of O 3 , the chamber with only the material present, and the chamber with the material and O 3 present. As with statistical testing for primary emissions, 100 steady-state time series data points were selected for each of the three test conditions. Statistical testing required consideration of multiple comparisons; thus, 3-group ANOVA was used to determine if statistically signicant differences existed across comparisons. The ANOVA F-test was rst calculated for each mass signal to determine if at least two of the means across groups were signicantly different. This test was performed with a ¼ 0.05. If the F-test determined at least two comparisons within the three groups were different, a post hoc Bonferroni-corrected t-test was performed for each of the three possible combinations of t-tests between the groups. The Bonferroni correction resulted in a p-value for the t-test statistic of 0.017. Yields were included as statistically signicant only if three conditions were met: (1) the F-test met the signicance threshold, (2) t-test comparison between chamber with material and O 3 and empty chamber with O 3 met t-test threshold, and (3) t-test comparison between chamber with 3 Results and discussion
An example of the resulting estimation of molar uxes across unit mass resolved mass spectra from the PTR-TOF-MS is shown in Fig. 2 for cellulose and berglass. Note that these emissions were those determined from the presence of the material alone, i.e., in the absence of O 3 . From Fig. 3, it can be observed that cellulose emits a larger quantity and more diverse range of volatile organic compounds due to the material itself compared to berglass, expected given the organic nature of cellulosic material vs. the higher inorganic content present in the berglass material. Mass to charge ratios where no ux is reported are those m/z ratios where the comparison of the empty chamber to the chamber with material did not meet the statistical threshold for signicance (a > 0.05). A summary of primary emission source and sink behaviour is shown in Fig. 3 for all tested materials. Detailed tables showing mass accuracy, putative chemical ID, and additional notes can be found in the ESI in Table S2. † In general, the tested materials acted as a source of VOCs to the chamber outlet, although in some circumstances statistically signicant decreases in chamber levels for specic VOCs compared to background tests were detectable for some materials. Note that for each statistically signicant mass signal (see Section 2.5.3) from the PTR-TOF-MS, the compound was considered as a source if eqn (1) was positive for that mass signal and a sink if eqn (1) was negative. Thus, as shown in Fig. 3, materials exhibit may act as a source for certain compounds while acting as a sink for others.
Cellulose insulation was the largest emitter of VOCs followed by recycled denim. Interestingly, both materials are made from recycled materials, as the cellulose was primarily derived from recycled newsprint. Cellulose is also one of the major components of denim, 39 explaining the similar magnitude and composition of observed primary VOC uxes. PSTB, when subtracting positive VOC uxes (sources) from negative VOC uxes (sinks), had the lowest VOC emissions of all tested materials, due in part to the modest sink effect observed for compounds m/z 47.0131, m/z 45.031, and m/z 61.0289 (see Table S2
A thorough exploration of the potential chemical mechanisms impacting primary emissions for each material is beyond the scope of this paper but may be warranted in the future given the relatively high primary emissions observed here for some materials. In the case of cellulose, the largest emitter, there exists a body of research demonstrating the instability of cellulose and release of VOCs. As noted previously, cellulose is also a common recycled insulation material, and is present in many other consumer products present indoors. For the ve largest signicant uxes from cellulose, we speculate that chemical assignments are protonated methanol (CH 3 OH), an acid fragment possibly associated with isopropyl alcohol, 40 acetaldehyde (C 2 H 4 O), formic acid (CH 2 O 2 ) and acetic acid (CH 3 COOH). Note that because of the limitations with the mass resolving power of the instrument and resulting potential for interferences, these assignments should be taken as tentative chemical identications.
Cellulose is known to be a reactive material, subject to a wide variety of degradation routes including chemical, thermal, and radiation induced reaction routes; 41 the largest primary emissions from cellulose may be explained from a variety of cellulose degradation mechanisms. Low molecular weight organic acids are known to be formed from degradation of polysaccharide chains in cellulose materials, forming acetic acid and formic acid. 42 Acetic acid was the highest measured ux (24.2 mmol m À2 h À1 ) from cellulose insulation, and has been observed as an emitted product from degradation of cellulose-containing museum materials. 43 Methanol and acetic acid have also been observed in FLEC cell studies of emissions from various solid wood products. 44 Note that wood is typically on the order of 40-50% cellulose. 45 Acetaldehyde is a commonly identied indoor air pollutant; 1 the presence of acetaldehyde ux from cellulose insulation may derive from aerobic microbial activity, possibly from the oxidation of ethanol present as a solvent in adhesives. 46 The presence of ethanol in the material is plausible as the cellulose insulation is made from recycled newsprint; inks contain a variety of solvents including ethanol. 47 We attribute the peak at m/z 47 to formic acid and not ethanol due to its alignment with the exact mass of formic acid (m/z 47.0128) and the presence of isotopes of intensity and at exact masses (m/z 48.01611 and m/z 49.017) expected for formic acid in the mass spectra. However, it is possible that volatile ethanol is present in our air sample at relatively low mixing ratio but is not distinguishable from the formic acid peak. Furfural is a known marker of cellulose degradation, 48 and has been detected previously with PTR-TOF-MS at m/z 97.0287 (protonated parent compound) and m/z 62.0334 (fragment). 49 Both signals are statistically signicantly elevated in the cellulose mass spectra, and in combination total a ux of $1.1 mmol m À2 h À1 of furfural; or the 7 th largest observed VOC ux.
Interestingly, polyurethane primary VOC emissions appeared similar to that of cellulose containing materials (cellulose and recycled denim). Some polyurethane spray foams may include cellulosic materials, 50 although the material safety data sheet (MSDS) for the polyurethane spray foam material used here did not list cellulose as part of the composition. There is limited data in the peer-reviewed literature on VOC emissions from polyurethane spray foam; a NIST report using microchambers on four spray foams reported the largest chemicals Other compounds were out of range of the PTR-TOF-MS mass range (TCPP), not detected (1,4-dimethylpiperazine and 1,2dichloropropane), or possibly detected at very low uxes (1,2dichlorobenzene). However, three of the four spray foams tested in the NIST study were sampled 5-24 months aer spraying. In this work, we tested the spray foam within 48 hours of spraying. Thus, we speculate that the major contributor to the observed VOCs is a result of the blowing agents used (or B-side components of this do-it-yourself spray foam kit), which may contain, e.g., formic acid. 52 A NIST report notes that "a wide range of aldehydes" were detected in spray foam samples from a test house, however, the house was aged for 1.5 years and these compounds may have originated from other sources, adsorbed to the spray foam, and subsequently desorbed during sampling. 53 Polyisocyanurate was characterized by substantial VOC emissions from m/z 41.038577 and m/z 42.033826. We speculate that m/z 41.038577 is protonated propyne (C 3 H 4 ), likely a fragment of a larger molecule based on prior studies in the literature. 54,55 The signal at m/z 42 may also be associated with acetonitrile in PTR-MS studies; 56 the exact mass of acetonitrile aligned well with the measured mass and acetonitrile is a solvent which may be used in the production of polyisocyanate polymers. 57 However, it is also possible that this signal represents the fragment of propanal. 58 Other materials (polystyrene, berglass, stone wool, PSTB) had generally lower primary VOC emissions. Polystyrene was the only material with a large peak at m/z 105.0699, which we attribute to styrene.
On a mass basis, total primary VOC emissions summed for all statistically signicant unit masses yields a total VOC (TVOC) mass ux ranging from $0.3 mg m À2 h À1 (from PSTB) to $3.3 mg m À2 h À1 (from cellulose). These estimates are based on the summation of statistically signicantly elevated molar uxes, assuming that the molecular weight of each compound is one amu less than the protonated mass. The complete list of statistically signicant positive molar and estimated mass uxes for cellulose and PSTB is available in Table S3. † These estimates of summed TVOC emissions are within the range of TVOC emissions reported for many polymeric building materials; for example, TVOC emissions from plywood have been shown to range 0.04-1.5 mg m À2 h À1 . 59,60 To explore the potential implications of these primary TVOC uxes, consider a well-mixed 150 m 2 single-story home with 3 m height walls and dimensions of 10 m  15 m. The total wall enclosure assembly area would be $150 m 2 . If windows contribute 20% of the wall enclosure area, the total wall area would be $120 m 2 . Assuming $90% of the wall cavity is lled with the tested insulation materials (10% accounting for studs and other construction elements), and that approximately half of the wall assembly contributes VOC uxes directly to the interior of the space (e.g., if stack-driven ow dominates and leakage areas are evenly distributed along the vertical height, half would be a reasonable approximation 61 ), the TVOC ux from the wall cavity could contribute between $16 mg h À1 and $180 mg h À1 , depending on material (and assuming mass transfer characteristics are consistent between the cavity and the chamber test conditions; a useful but somewhat unrealistic approximation). If the air exchange rate in the space is 0.5 h À1 , the resulting TVOC concentration would be between $70 mg m À3 and $800 mg m À3 . These approximations are reasonably consistent with limited prior estimates of the contribution of wall cavity materials to indoor spaces of which we are aware. 9,11 However, future work beyond the scope dened herein should integrate these ndings into more mechanistic models of emissions and transport from enclosures to interior spaces.
Results of the calculation of deposition velocity and reaction probability are presented in Fig. 4. Ozone deposition tests were conducted in duplicate; error bars shown in Fig. 4 are the larger of propagated uncertainty or the range across duplicate tests. Results show that g varied by more than an order of magnitude across materials, from $1 Â 10 À6 to $3 Â 10 À5 . These values are in the range that Liu and Nazaroff 15 predict could yield highly varying O 3 penetration factors through insulation materials alone under realistic pressure differences, ranging from $70% penetration to <10% penetration.
There exist relatively few measurements of O 3 reactivity of building enclosure insulation materials in the literature. To the best of our knowledge, reaction probabilities have been reported only for berglass materials, which is discussed subsequently. Given the recent acknowledgement that building envelopes can act as a protective barrier for O 3 , 21,62,63 we expect these data to be useful for modelling indoor-outdoor O 3 transport and as an aid in building enclosure material selection. In general, brous and loose materials (cellulose, berglass, stone wool and recycled denim) appeared to have generally higher g while rigid, smoother materials (polyurethane, polystyrene, polyisocyanurate, and polystyrene with thermal backing (PSTB)) appeared to have lower g. Given that estimates reported here are "effective" reaction probabilities, meaning they are derived from a deposition velocity parameterized to horizontal projections of test material surface area, these values are over-predictions relative to estimates that consider the internal surface area of a material. 64 As will be discussed, this consideration is of particular importance given the potential for complex and varied ow paths across or through materials in building enclosure assemblies.
We measured the transport-limited deposition velocity for four materials, selecting materials that varied in structure and morphology. Measured v t are reported in Table S4 † of the ESI, † and ranged from 0.14-0.27 cm s À1 , indicating potential for transport of O 3 to the surface to impact overall ozone removal for some tested materials. For the chamber conditions used, resistance due to transport and surface reaction were on the View Article Online same order of magnitude for ve of eight tested materials (PU, DM, SW, C, FG) while for three tested materials (PI, PSTB, PS) the surface resistance dominated. Implications for uncertainty in calculated reaction probabilities are discussed the ESI. † As described in Sections 2.1 and 2.2, materials were tested in a well-mixed continuous ow reactor (CFR) with one surface of the material prepared to allow interaction with the bulk chamber air. The parameterization of the deposition velocity requires a surface area for calculation, normally taken to be the horizontal projection of surface area. Air transport across or through materials within building enclosures is complex, but it is expected that the majority of transport occurs through pressure-driven ow across larger (0.2-1 mm in crack height) size cracks and gaps in the enclosure. 15 Liu and Nazaroff 15 discuss the likely ow paths of air entering a building enclosure lled with berglass insulation and conclude the airows are likely to either traverse through a berglass mat in the direction orthogonal to a vertical wall or bypass insulation in the air space between the insulation and the frame. In the former case, it is likely that a brous mat acts as lter, with greater internal surface area available for interaction than in the latter case of airow predominantly passing through void space.
Reaction probabilities calculated here are more likely to be representative of a scenario where air movement occurs in the void space adjacent to a material in a two-phase material-air system. While limited data exists reporting O 3 reaction probabilities to building insulation materials, Liu and Nazaroff 15 empirically estimate the reaction probability of berglass bers to be 6 Â 10 À6 , lower than the value of 2.8 Â 10 À5 reported here. Note that the comparison of these reaction probabilities is for that of a ber of berglass compared to the effective reaction probability made using a horizontally projected area of a material sample. In contrast, the reaction probability reported here is similar to the value of $3 Â 10 À5 reported by Lamble et al. 29 for berglass ceiling tiles tested in a similar manner to the apparatus used in the present study. Liu and Nazaroff 15 use a plug ow reactor, described in detail by Morrison and Nazaroff, 65 that exposes a greater internal surface area of the material to O 3laden ow than the continuous ow reactor (CFR) type apparatus used here and in Lamble et al. 29 Future studies of pollutant transport and transformation occurring in building enclosures should carefully consider the formulation, and assumptions inherent to, surface and transport resistance terms, as ow paths in building enclosures are likely complex and both spatially and temporally variant.
Major contributors to the O 3 byproduct yield with the tested materials appear to be oxygenated VOCs, primarily aldehydes; O 3 byproduct yields are shown in Fig. 5. Yields ranged from 0.25 (polystyrene) to 0.85 (recycled denim) moles of byproduct formed per mole of O 3 consumed (mol mol À1 ). For the experiments conducted here, these yields are equivalent to secondary uxes that range from 0.71 (polystyrene) to 10 (recycled denim) mmol m À2 h À1 . While lower in magnitude than the primary emissions discussed in the scaling analysis in Section 3.1, it is plausible that secondary products from O 3 reactions on wall cavity materials could impact indoor air quality. We believe these ndings compel further study of the oxidation pathways in wall cavities, including coupling of outdoor-enclosure-Fig. 4 Summary of ozone deposition velocities and reaction probabilities calculated for each of the eight test materials. PSTB ¼ polystyrene with thermal backing. *Note that g for these materials were calculated using v t measured for another tested material with similar surface morphology. The assumed v t results in an additional source of uncertainty in the calculation of g for these materials, described in greater detail in the ESI in Table S4 indoor uid dynamics models with studies of emissions and transformation.
Detailed tables showing mass accuracy, putative chemical ID, and additional notes for compounds contributing signicantly to the byproduct yield can be found in the ESI in Table S5. † Buhr et al. 58 showed that most aldehydes will lose a molecule of water leading to a main fragment at m/z (MH + -18). Therefore, we attribute the peak at m/z 83 to a fragment of hexanal with a loss of a molecule of water (mass measured ¼ 83.0804), consistent with the presence of a peak at m/z 101 possibly corresponding to the protonated parent compound. For the same reason, we also attribute the peak at m/z 55 to a fragment of butanal (MH + -18) (exact mass ¼ 55.054227, measured mass ¼ 55.0454), conrmed by the presence of a peak at m/z 73 corresponding to protonated parent compound (exact mass ¼ 73.06534, measured mass ¼ 73.0411). However, this assignment should be taken with caution, as fragmentation of hexanal can lead to a signal at m/z 55, which may also be true of heptanal (Buhr et al., 2002). 58 The presence of heptanal was also conrmed by the presence of a peak at m/z 97, which corresponds to a loss of water from the parent compound as well. Aldehyde fragments were found in cellulose, polyisocyanurate, berglass and stone wool insulation as well as recycled denim insulation, although the hexanal fragment (m/z 83) was not among statistically signicant compounds for recycled denim.
Further limitations include the potential for presence of the structural isomers, such as propanal and acetone as oxygenated byproducts from reaction of the materials with O 3 . Mass signals at m/z 59 were present for all materials. Only recycled denim and polyurethane contained signicant signals at m/z 41, although at higher levels than would be expected due to fragmentation from a propanal parent at m/z 59. Thus, we conclude that for recycled denim and polyurethane, both acetone and propanal are likely present in the sampled air. Given the precedence for O 3 -building material interactions to result in the formation of carbonyl compounds and acetone, 66 this nding is expected. The PTR-TOF-MS method does not enable the separation of these structural isomers, and so assignment of these compounds to signals in a matrix possibly containing both compounds is challenging.
Interestingly, cellulose appears to have a different secondary emission prole and yield from that of recycled denim although they had similar magnitude and composition of primary emissions. This is most likely due to differences in material or surface-bound chemical composition of the two materials, leading to different heterogeneous chemistry that yields distinct oxygenated products. The largest byproduct yield for cellulose corresponds mainly to aldehyde fragments (m/z ¼ 55, 69 and 83), acetone, and more widely to un-attributed mono-or dioxygenated compounds. Heterogeneous reactions of recycled denim and polyurethane with O 3 led to high formation yields for acetaldehyde (with contributions of acetaldehyde-water cluster at m/z ¼ 63.05 according to Herbig et al. 67 ), acetone, and, more generally, mono-and di-oxygenated compounds ($0.85 and $0.53 mol mol À1 consumed for denim and polyurethane, respectively), making them the highest emitters of byproducts in the presence of O 3 . Recycled denim and polyurethane had similar yield proles. A peak at m/z 41 likely corresponds to a propanal fragment, pentanal fragment, and/or alcohol fragment according to Wyche et al. 68 (exact mass ¼ 41.038577, measured mass ¼ 41.0358). This peak was observed only for denim and polyurethane with a yield of 0.05 mol mol À1 for both materials and could be related to the oxidation of polyols present in polyurethane (polyols react in excess with isocyanates to make polyurethane and can still be present in the nal polymer and give aldehydes that react with O 3 and could lead to carboxylic acid in presence of water). A peak at m/z 62 was found as a byproduct only in those two materials with relatively high yields (0.07 and 0.04 mol mol À1 for denim and polyurethane, respectively). We attributed this peak to the protonated carbamic acid (CH 3 NO 2 ) due to its alignment with the exact mass (exact mass ¼ 62.023655, measured mass ¼ 62.0266). The formation of those compounds with only those two materials in the presence of O 3 might be explained by the presence of polyurethane in denim, which could explain the similar composition in VOCs primary emission and byproducts formation.
Formic acid (exact mass ¼ 47.012756, measured mass ¼ 47.0128) was common and one of the most elevated yields among ve of the eight materials (present in berglass, stone wool, polyisocyanurate, polystyrene, and PSTB). Yield values ranged from 0.04 to 0.16 mol mol À1 . Polyisocyanurate, berglass, and stone wool had a similar prole of byproduct formation yield, with aldehyde fragments (for a total of 0.08, 0.09, and 0.08 mol mol À1 , respectively), formic acid (0.06, 0.04, and 0.06 mol mol À1 , respectively), and acetone (0.05, 0.07, and 0.13 mol mol À1 , respectively), with stone wool presenting a relatively high yield of acetaldehyde (0.06 mol mol À1 ). Polystyrene and PSTB had a very similar prole, with higher yields for formic acid, acetone, and acetic acid with PSTB.
When considering both the removal of O 3 and the resulting byproduct formation yield, there exists a range of behaviors. In general, O 3 reactivity does not appear to be predictive of byproduct formation yield. Both cellulose and berglass have shown relatively high O 3 sink strengths compared to the other materials tested in this study, but their secondary TVOC formation rate is lower than other insulation materials. A potential reason for this is the cellulose and berglass surface compounds may be structured to allow for a catalytic decomposition pathway for O 3 , resulting in the production of CO 2 and O 2 . Recycled denim and polyurethane appear to have the opposite behavior. Lower deposition velocities were found for these two materials as compared to cellulose, but a higher formation yield was detected. Due to the fact that they have similar primary emission proles as cellulose, the difference is most likely due to differences in surface morphology and the composition of these materials.
The work presented here represents, to our knowledge, a substantial expansion of the ozone reactivity and primary and secondary emission behavior of building insulation materials. However, there exist important sources of uncertainty in this investigation that compel future studies of this important class of building material. First, while we selected eight materials to span a range of commonly used insulation materials, there exist other types of wall enclosure products that should be tested. In addition, there are multiple manufacturers of any given type of insulation material, and variation in VOC emissions and ozone chemistry of a material type for any given manufacturer is also possible. Future work could investigate variability among similar products across different manufacturers as well as within-manufacturer variability (e.g., subsampling across a single lot of material as well as acquiring samples from a single manufacturer with different manufacture dates).
The scope of this study was to investigate the VOC emissions and ozone reactivity of common building insulation materials, as manufactured. Future work would also be well-served by conducting additional characterization of materials to quantitatively capture differences in material chemical composition and morphology. These data would advance understanding of the drivers of observed differences in emission and reactivity behavior across materials. Primary emissions were measured over a period of 55 minutes, ideally, emissions would be measured over a longer time period to ensure full desorption of compounds that may have adsorbed to the material from other sources (e.g., air in storage warehouse or in sample storage bag air). We attempted to minimize this source of error by keeping materials in their manufactured bags and limiting the amount of time the samples were exposed to laboratory air. Finally, while ionization via PTR-TOF-MS is, in theory, "so", fragmentation of aldehydes is known phenomena that complicates the calculation of emission rates of this class of compounds. Furthermore, the PTR-TOF-MS is not a universal detector, future studies should inter-compare emission rates estimated with PTR-TOF-MS complemented by, and in comparison with, other analytical methods such as TD-GC-MS and HPLC-UV.
This study investigated the primary emissions, ozone (O 3 ) reactivity, and O 3 reaction byproduct emissions from eight commonly used building insulation materials. Results demonstrate that cellulose insulation was the largest emitter of primary VOCs, followed by recycled denim. Polystyrene, berglass, and stone wool had relatively low primary VOC emissions, and polystyrene with thermal backing actually served as a sink for some VOCs. The O 3 reaction probability of these materials ranged more than an order of magnitude, and total reaction byproduct yields ranged from $0.25 to $0.85 moles of byproduct formed per mole of O 3 consumed. A number of secondary VOCs resulting from O 3 reactions were logically deduced (and varied by material), but further analysis should be done to clearly identify the secondary byproducts formed due to oxidation of insulation. To our knowledge, this study provides the rst characterization of the aforementioned parameters for a range of common insulation materials. The data presented herein could serve as the basis for informing quantitative comparisons of trade-offs between different enclosure insulation materials, e.g., consideration of thermal resistance in conjunction with material emissions, O 3 removal, and byproduct formation. These data can also inform building enclosure transport modelling efforts, which we recommend be improved upon in future work to incorporate the ability to predict the impacts of oxidation chemistry in building enclosures on both primary and secondary pollutant uxes into the space.
Published on 25 March 2019. Downloaded by Portland State University on 2/7/2020 11:18:46 PM.View Article Online