Stable isotope measurements confirm volatile organic compound oxidation as a 
major urban summertime source of carbon monoxide in Indianapolis, USA

Atmospheric carbon monoxide (CO) is a regulated pollutant in urban centers. Oxidation of volatile organic compounds (VOCs) has been hypothesized to 20 contribute substantially to the summertime urban CO budget. We performed measurements of CO stable isotopes on air samples from three sites in and around Indianapolis, USA over three summers to investigate the VOC contribution to urban CO. One of the sites is located upwind of the city, allowing us to quantitatively remove the background air signal and isolate the urban CO enhancements. The 25 distinct isotopic signatures of CO produced from fossil fuel combustion and VOC oxidation allow us to separate contributions from these two sources. Our results provide the strongest empirical evidence to date of large contributions from VOC oxidation to the urban summertime CO source and show that this contribution varies in time and location between 0 and 58%. We attribute the remainder of the 30 Indianapolis summertime CO budget to fossil fuel combustion. We assess the reactivities of different VOCs and determine that biogenic sources are likely Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-506 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 22 June 2018 c © Author(s) 2018. CC BY 4.0 License.

responsible for the majority of CO produced by VOC oxidation reactions within Indianapolis.

5
On the global scale, mole fraction of carbon monoxide (CO) has four major sources which include biomass burning, oxidation of methane (CH4), the incomplete combustion of fossil fuels and the oxidation of biogenic volatile organic compounds (BVOC's, Logan et al., 1981;Duncan et al., 2007, Table 1). These sources are countered by the oxidation of CO by hydroxyl radicals (OH) resulting in a mean 10 residence time of CO in the atmosphere of roughly 2 months (Logan et al., 1981;Duncan et al., 2007). On a regional scale in urban areas, CO mole fractions are often significantly enhanced due to the incomplete combustion of fossil fuels (https://www.epa.gov/air-emissions-inventories/air-emissions-sources, Mak and Kra, 1999;Popa et al., 2014;Turnbull et al., 2015;Vimont et al., 2017). Additionally, 15 during the summer months, previous literature suggests that there may be large urban source of CO from the oxidation of biogenic volatile organic compounds (BVOC's) (Guenther et al., 1993(Guenther et al., , 1995Carter and Atkinson, 1996;Kanakidou and Crutzen, 1999). 20 Tangential evidence of a significant urban CO source in addition to fossil fuel combustion is provided by Turnbull et al. (2006) and Miller et al. (2012). These studies aimed to predict fossil fuel CO2 (CO2,ff) by using CO as a proxy gas, but noted that the ratio of CO:CO2,ff was higher in the summer than the winter (Turnbull et al., 2006;Miller et al., 2012). A higher CO:CO2,ff ratio is not consistent with a sink 25 process, such as an increase in OH during the summer months. Instead, an increase in a non-fossil fuel source provides the most likely explanation for the increase in the CO:CO2,ff ratio. These studies hypothesized, but could not confirm, that oxidation of VOC's may be the source of this summertime increase in CO.
30 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-506 Manuscript under review for journal Atmos. Chem. Phys. Studies that model the effect of CO sources on the measured CO mole fraction have also suggested oxidation of VOC's contribute significantly to the CO budget, in particular BVOC emissions (e.g. Kanakidou and Crutzen, 1999). Griffin et al. (2007) used an atmospheric chemistry model to investigate CO production by VOC oxidation at a regional scale in the United States. Their model determined that 20-5 40 nmol mol -1 out of an approximate enhancement of 100-200 nmol mol -1 of CO in air was derived from VOC oxidation, and attributed the majority of this to isoprene. Cheng et al. (2017) measured O3 and CO mole fractions and then modeled CO production from the various sources using O3-to-CO ratios. Their model suggested the oxidation of isoprene might account for the total anthropogenic production of 10 CO within the urban region of Baltimore, USA.
No matter the scale, the attribution and quantification of CO sources and sinks is difficult. Forward and inverse modeling estimates that simply use CO mole fraction measurements to apportion the relative impact of sinks and sources have high 15 uncertainty due to additional uncertainties in transport and chemistry (e.g. Kanakidou and Crutzen, 1999;Duncan et al., 2007). Measuring stable isotopes of CO ( 13 CO and C 18 O) can provide a robust method to directly quantify the relative strengths of the different sources of CO and avoid some of the complications encountered with simply modeling the CO mole fraction (Brenninkmeijer, 1993;20 Röckmann and Brenninkmeijer, 1997;Brenninkmeijer et al., 1999). However, accurately apportioning the various sources of CO from atmospheric observations of the isotopic content can only be done if the isotopic signatures of the sources are well known . The large differences between the signatures of different sources in both 13 CO and C 18 O make source attribution 25 possible despite the substantial uncertainties associated with the isotopic signatures of the sources and the OH sink Gros et al., 2001Gros et al., , 2002 (Table 1). For example, the C 18 O signature of oxidation sources (~0‰) is significantly lighter than combustion sources (16-24‰) Gros et al., 2001; Table 1). 30 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-506 Manuscript under review for journal Atmos. Chem. Phys. Only a few prior studies of CO isotopes in urban regions exist (Stevens et al., 1972;Sakagawa and Kaplan, 1997;Mak and Kra, 1999;Kato et al., 1999;Tsunogai et al., 2003;Saurer et al., 2009;Popa et al., 2014;Vimont et al., 2017). These studies generally attribute much of the urban pollution to combustion of fossil fuels via automobiles and biomass burning (Saurer et al., 2009;Kato et al., 1999). This, along 5 with variations in fuel sources, can cause CO source isotopic signatures to change through time and differ regionally, particularly in urban areas (Popa et al., 2014).
Although none of the previous urban CO isotopic studies have identified VOC oxidation as a significant urban source of CO, Sakugawa and Kaplan (1997) noted an 10 "unknown", non-fossil source of CO within Los Angeles, CA, USA. Stevens et al. (1972) suggest oxidized BVOC's as a source of rural CO during the spring and summer months, but attribute urban CO to engine emissions alone. Mak and Kra (1999) found a "burst" of CO in the spring during their measurement 15 campaign at Long Island, NY, USA, and attribute this to a rapid increase in fossil fuel combustion from tourist activity in the region and peak energy consumption during the summer months. This conclusion was based on known energy data and tourist information (Mak and Kra, 1999). However, the isotopic data presented by Mak and Kra (1999) do not preclude increased oxidation of VOC's, though the signal is 20 complicated by the increase of OH oxidation of CO and CH4. Vimont et al. (2017) examined the wintertime urban CO budget of Indianapolis and characterized the overall isotopic signature of CO emissions from fossil fuel combustion.
In this study, we use stable isotopes of carbon monoxide measured over two summers to assess the contribution from oxidized VOCs to the CO budget at 25 Indianapolis. We then identify the likely CO precursors by assessing expected mixing ratios and atmospheric oxidation rates. Indianapolis, Indiana is a metropolitan area of over one million people in the Mid-West region of the United States. It is surrounded by mostly agriculture interspersed with trees and foliage both inside and outside of its borders ( Figure 1).

Tower Sampling at Indianapolis
It has distinct seasons, with hot summers (25 to 30° C) and cold winters (-8 to 1° C), 5 which result in a distinct growing season, with the winter being relatively devoid of biogenic fluxes of CO and CO2 (Turnbull et al., 2015). The Indianapolis FLUX project (INFLUX) aims to develop and assess methods for determining urban greenhouse gas emissions. CO, though not a primary greenhouse gas, is measured and used as a tracer for fossil fuel CO2 emissions, and to provide information for source 10 attribution. This study uses CO isotope measurements on samples from the existing INFLUX network and attempts to better quantify the urban CO source budget.
INFLUX has twelve instrumented towers within and around the urban boundary ( Figure 2) . The flask-sampling regime was described in detail by 15 Vimont et al. (2017) andTurnbull et al. (2015). In brief, discrete air samples are collected at six of the towers, three of which are sampled for CO isotopes (towers 1 -3 on Figure 2) approximately six days per month, during the early afternoon when the strongest boundary layer mixing occurs (19:00 UTC, 14:00 local). Stable isotope measurements of CO were made on samples collected from July 2013 to July 2015. 20 However, this study only considers the summer samples that were collected in July fraction measurements used in this study (Novelli et al., 2003).
For the samples in this study, collection was done when the wind was approximately from the west, so that Tower 1 provides a clean-air background for the towers further to the east (Turnbull et al., 2012). Tower 2 is east of the city, 30 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018- with only a small residential influence and one major highway nearby, with significant foliage within its influence footprint (Turnbull et al., 2015;Figure 2).
Tower 3 is in the downtown, urban center, and is strongly influenced by anthropogenic fossil fuel emissions, with relatively fewer biogenic sources (Turnbull et al., 2015;Figure 2). The distance between towers 1 and 2 is larger (51 5 km) than the distance between towers 1 and 3 (36 km). Therefore, there is more time for reactions to occur between tower 2 and tower 1, and any atmospheric reaction source of CO will have greater influence at tower 2 relative to tower 3.

Stable Isotope Analysis
The stable isotopic measurement procedure is described in detail in Vimont et al. (2017). Briefly, the air is extracted from the PFP by vacuum transfer through a cryogenic trap at -60° C that removes water vapor. Next, a mass flow controller is 25 used to regulate the flow of the sample through a second cryogenic trap at -196° C that removes CO2, N2O, and any other condensable species. The remaining air is passed through acidified I2O5 suspended on a silica gel matrix (Schutze's reagent, Schutze, 1949)  remove any excess sulfuric acid that has evolved from the reagent, and finally the CO-derived CO2 is trapped on a third cryogenic trap (also at -196° C) while the remaining gasses are pumped away. The CO-derived CO2 is then transferred to a cryogenic focusing trap, and finally released through a GC column (PoraBond Q) to the isotope ratio mass spectrometer (GV Instruments IsoPrime 5KeV). 5 Following convention, we use delta notation to report our isotopic results: where Rs is the ratio of 13-carbon to 12-carbon in the sample, and RVPDB is the ratio of 13-carbon to 12-carbon in the international standard Vienna Pee Dee Belemnite. 10 The same relationship describes d 18 O except the international standard of reference is Vienna Standard Mean Ocean Water (VSMOW). Because we are oxidizing CO to CO2 in this analysis, we must correct our CO2 d 18 O data to account for the added oxygen, as described in Mak and Yang (1998): We note that a significant deviation from the standard CO2 17 O correction has been observed and quantified for CO, particularly in the high northern latitudes (Röckmann and Brenninkmeijer, 1998;Röckmann et al., 1998). This so called " 17 O 30 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-506 Manuscript under review for journal Atmos. Chem. Phys. excess", or Δ 17 O, is a result of mass-independent fractionation (MIF) that arises in OH and O3 photolytic formation (Röckmann et al., 1998;Huff and Thiemens, 1998).
This effect can introduce error of up to 0.35‰ in the corrected d 13 C values, and is only quantifiable by measuring d 17 O (Röckmann and Brenninkmeijer, 1998).
However, though we do not measure d 17 O for our samples, our analysis (section 2.5) 5 precludes the need for this correction because both background and urban samples will see similar Δ 17 O effects.

10
One of the advantages to the INFLUX experiment is the ability to remove background signals from the urban measurements, and thereby derive the urban enhancement. This approach also allows the CO budget to be simplified. Both the oxidation of CH4 to CO and the oxidation of CO to CO2 via the OH radical are reactions that proceed slowly relative to the experimental scale of a few hours 15 transit time between the background and urban sites. Because of this, we calculate that these two processes have negligible impact on our urban CO enhancements, and can be disregarded given the short reaction time being considered.
The reaction time period can be calculated simply by considering the distance 20 between tower 1 and towers 2 or 3 and the average wind speed. Given the average wind speed during sampling for this study was 4.4 ± 1.3 m s -1 , a 2.7-hour transit time is required. In this experiment, we correct our results to account for the incoming background CO and examine the urban contribution alone. This short transit time scale allows us to place constraints on the CH4 oxidation source and the 25 OH oxidation sink of CO.
Oxidation of CH4 by OH is a major source of CO globally but CH4 is long lived in the atmosphere relative to CO (Sander et al., 2006;Atkinson et al., 2006;Duncan et al., 2007). The approximate rate for the reaction of CH4 with OH is 6.4x10 -15 cm 3 s -1 at 30 standard pressure and our mean ambient temperature of 26° C (Atkinson et al., 2006). OH mole fraction has been determined at urban sites in similar latitude bands and ranges from 1x10 6 cm -3 in cool, winter time conditions to 2x10 7 cm -3 in hot, summertime conditions (Warneke et al., 2007(Warneke et al., , 2013Atkinson and Arey, 2003;Park et al., 2011). We do not have OH mole fraction measurements at Indianapolis, 5 and therefore use the highest reported literature value for OH of 2x10 7 cm -3 (Park et al., 2011) to assess the maximum CH4 oxidation contribution to CO (Park et al., 2011, Table 2). We calculated the change in mole fraction of CO due to oxidation of CH4 by OH by: where ΔXCO is the change in CO mole fraction due to CH4 oxidation by OH, γ is the CO yield for the CH4+OH reaction (0.96 mole CO produced per mole CH4), XCH4,i is the initial CH4 mole fraction (the average CH4 mole fraction during the sampling period, We further assessed the impact on CO isotopes (Table 2) by using the reported isotopic values for CH4 oxidation (Table 1). We calculated the change in δ 13 C and d 18 O by 20 where Δδ is the change in either δ 13 C or d 18 O, δCOi is the initial delta value at the polluted towers (average of the two towers (non-enhancement) of -29.6‰ for δ 13 C and 5.1‰ for d 18 O), XCOi is the CO mole fraction at the two polluted towers (average value of 166 nmol:mol), dCH4 is the d 13 C or d 18 O value of CO produced by CH4 25 oxidation (-52.6‰ and 0‰ for d 13 C and d 18 O respectively), and XCOCH4 is the mole fraction of CO produced from oxidation of CH4 by OH, calculated above. Using these parameters and the average transit time between the towers of 2.7 hours, we calculate that during the transit across the city, CH4 oxidation could contribute up to 1.4 nmol:mol CO, changing d 13 C by up to -0.21‰, and d 18 O by up to -0.04‰. These values are below our 1σ measurement uncertainties (0.23‰ d 13 C and 0.46‰ d 18 O), and thus we do not consider CH4 oxidation to be a significant 5 source of CO in our analyses.
OH oxidation is the main sink of CO, and will directly impact the isotopic signatures of CO measured within the city (Röckmann and Brenninkmeijer, 1997;Duncan et al., 2007). Using the same method and OH mole fraction as for CH4 oxidation above, 10 and a reaction rate for CO+OH of 1.44x10 -13 cm 3 s -1 (Atkinson et al., 2006), we calculated the net loss of CO during the transit of an air mass across the city.
However, to calculate changes to the isotopic budget, we use the fractionation factors for OH oxidation found in Table 1 where XCOT is the total CO mole fraction measured at tower 1 (mean value of 146 nmol:mol), and XCOlost is the amount of CO removed by oxidation with OH. α is the fractionation factor for either d 13 C or d 18 O from the literature ( Table 1) Biomass burning can be a source of CO in urban regions, though it is primarily used as a heat source (Saurer et al., 2009). Within Indianapolis, 2/3 of residential and commercial heating is done by natural gas combustion, and the remaining 1/3 is electrical (Gurney et al., 2012). Vimont et al. (2017) estimated that biomass burning for heat was only about 1% of the CO budget during the winter, and did not impact 5 the isotopic budget significantly. As there should be much less (if any) biomass burning for heat during the summer, we assume that biomass burning is not a significant source of CO. Any biomass burning outside the city (burning off of crop fields or forest fires) is accounted for by removing the background.

10
The remaining sources of CO that must be considered are oxidation of VOC's (both biogenic and anthropogenic), and fossil fuel combustion. Fossil fuel combustion has long been considered the primary source of CO within urban regions (Stevens et al., 1972; EPA NEI 2011), whereas only recently has biogenic VOC oxidation been shown to be a significant urban source (Cheng et al., 2017). 15

Data filtering based on meteorological conditions
During the summer months, Indianapolis experiences thunderstorm activity, which is associated with convective air movement. When this convection occurs, it is 20 possible the towers are influenced by air that is not representative of the urban enhancements due to entrainment of clean, free tropospheric air. Using data provided by NOAA's National Center for Environmental Information (NOAA NCEI), we obtained daily meteorological data for each sampling day (http://www.ncdc.noaa.gov/qclcd/QCLCD?prior=N). We removed days with winds 25 that were calm, or had highly variable wind direction, as well as days with thunderstorm activity before, during, or directly after sampling occurred (Table 3).
This filtering was necessary because these days with thunderstorms were also large outliers on our regression plots (figure 4). After removing these data, there were 16 (out of the initial 30) usable days for analysis. 30

Regression Plot Analysis
At Indianapolis the CO enhancements measured at towers 2 and 3 are small relative to the background CO at tower 1 (17 nmol:mol on average at tower 2, and 22 5 nmol:mol on average at tower 3 relative to average background CO at Tower 1 of 146 nmol:mol). Because of this, it is necessary to remove the background signal from the polluted towers to accurately constrain the urban signals. Using the method described by Miller and Tans (2003), we calculate the isotopic signature of the urban source: 10 δ s = δ meas X CO meas -δ bkg X CO bkg X CO meas -X CO bkg (7) where ds is the d 13 C or d 18 O of the urban source (Figure 4), the subscript meas indicates the d 13 C (or d 18 O) and CO mole fraction measured at either tower 2 or 3.
The subscript bkg indicates the d 13 C (or d 18 O) and CO mole fraction measured at tower 1. We note that in equations (7) and (8) (below), the relationships hold only 15 for the common mass isotopologue, but for small changes in the 13/12 C and 18/16 O ratios, the errors in these equations are negligible. In order to obtain a 'best-fit' solution using (3) for all the data, we regressed the numerator against the denominator using an ordinary least squares (model 1) Y|X approach, which assumes mole-fraction to be independent (Isobe et al.,1990;Zobitz et al., 2006). 20 To account for uncertainty in our measurements, we used a Monte Carlo technique.
Using the propagated measurement uncertainties, we assigned an error distribution to each point. We assumed a normally distributed error curve based on QQ plot analysis (not shown). 10,000 regressions were run, randomly selecting values for 25 each data point from that point's error distribution. The reported slopes are the median values from the 10,000 regressions. We use the median of the regression slopes rather than the mean because the median is more robust to outlier points than the mean (Miller et al., 2012). Because of the high scatter in our data set  (Figure 4), the median provides a better estimate of the overall ds in equation (7).
The errors on the slope are 1σ for the slopes of each simulation. Finally, to assess how well our regression analysis results represent a solution for each point, we use the median slope and intercept to determine the residuals for each data point, and calculate an r 2 for each tower and isotope. r 2 is a metric for the strength of the 5 correlation between x and y, and both uncertainty in the measured values and real atmospheric variability in the relationship between x and y will work to reduce r 2 .
Therefore, in this context, r 2 is a metric for determining the likelihood of a single source (or sources with identical isotopic signatures) or multiple sources with different isotopic signatures contributing to the CO signal on different days. High r 2 10 correspond to a single isotopic signature, whether by a single source or multiple sources, and a low r 2 corresponds to multiple isotopic source signatures varying through time.

Mass Balance Source Attribution 15
Through our calculations and reasoning above, we are able to neglect the CH4 oxidation source, the biomass-burning source, and the OH oxidation sink. In order to constrain the remaining two sources (fossil fuel combustion and VOC oxidation, Duncan et al., 2007), we use a simple mass balance approach. We assume that the ds 20 calculated at each polluted tower (section 2.5, equation (7)) can be represented by: where fVOC and dVOC are the mole fraction and isotopic signature of VOC oxidation, and fFF and dFF are the mole fraction and isotopic signature of fossil fuel combustion. 25 The isotopic signatures of VOC oxidation are -32 ± 2‰ and 0 ± 3‰ for d 13 C and d 18 O respectively Gros et al., 2001). These values have not been determined for this study area, and therefore have high uncertainty. The d 13 C and d 18 O values were originally determined by Stevens and Wagner (1989)  changing background air (via variable source distributions outside of their experimental area). Their approach assumed all CO added to the background was solely from VOC oxidation, but other sources may also have contributed. Further work by Brenninkmeijer (1993) and Röckmann and Brenninkmeijer (1997) suggest that the d 13 C value is reasonable, but assign ~2‰ uncertainty bounds (1σ). Conny 5 et al. (1997) invoke a -3‰ fractionation factor for the oxidation of isoprene to CO, based on an average d 13 C of -29‰ for isoprene (Sharkey et al., 1991). However, Sharkey et al. (1991) suggest that the d 13 C may vary as plants are stressed, which is possible given the above mentioned urban heat island induced heat stress on urban trees (Califapietra et al., 2013). Moreover, they focused only on isoprene rather 10 than a larger set of VOCs. From this perspective, the relatively large uncertainty suggested by Röckmann and Brenninkmeijer (1997) seems reasonable. Brenninkmeijer (1993) and Röckmann and Brenninkmeijer (1997) deduced a d 18 O value of 0‰ from their work in the high northern and southern latitudes. This value is generally accepted in the literature, yet disagrees substantially with Stevens 15 and Wagner (1989), who calculated 15‰ from their study in rural Illinois. We use the value of 0 ±3‰ because it best explains the signals seen in several high latitude atmospheric studies (e.g. Röckmann et al., 2002;Park et al., 2015). We note that the uncertainty in these VOC isotopic signatures contribute a large portion of the overall uncertainty in the conclusions of this paper. More precise isotopic values for this 20 oxidation process would drastically improve the estimates of VOC produced CO in isotope studies. The isotopic signatures of fossil fuel combustion at Indianapolis are -27.7 ± 0.5‰ and 17.7± 1.1‰ for d 13 C and d 18 O respectively (Vimont et al., 2017).
Previously, we found that the isotopic signature in the winter did not vary with temperature significantly, and that the primary source within the city was emissions 25 from transportation (Vimont et al., 2017). Therefore, we use these values as the fossil fuel produced CO isotopic signatures for Indianapolis. By solving (4a) and (4b) for fVOC we are able to place a constraint on the VOC oxidation contribution to the urban CO budget.

Time Series
The full-time series of CO mole-fraction, d 13 C, and d 18 O has been shown previously 5 by Vimont et al. (2017) and figure 5 from that paper is reproduced here for completeness. Briefly, CO mole-fraction exhibits a seasonal trend that is similar to that found in other studies, with maximum CO occurring during the winter (February-March) and minimum values occurring in late summer (August) (Gros et al., 2001;Röckmann et al., 2002). Tower 2 and 3 are systematically enhanced 10 relative to tower 1 (background), demonstrating consistent urban enhancement. this differs from the high latitude background signals reported by Röckmann et al. (2002). High latitude CO seasonal cycles are driven both by the OH sink and transport to and from the mid-latitudes (Röckmann et al., 2002), whereas midlatitude continental sites have stronger influences from local or regional sources (Gros et al., 2001). 20 Towers 1 and 2 exhibit more negative isotopic signatures than tower 3 during the spring and summer months, likely due to higher fossil fuel contribution at the downtown location. The trend is consistent with VOC oxidation sources being more important at both tower 1 and 2. d 18 O shows a larger spread between all three 25 towers relative to d 13 C, which is also consistent with the larger difference in d 18 O between the fossil fuel combustion and VOC oxidation sources. The Monte Carlo regression analysis produced urban source isotopic results of -28.8 ± 2.3‰ and -27.7 ± 2.1‰ for d 13 C, and 10.9 ± 3.2‰ and 13.0 ±4.9‰ for d 18 O at towers 2 and 3 respectively (Figure 4). In both isotopes, tower 2 is associated with lower urban source isotopic values than tower 3. With the higher variability in d 18 O 5 measurements than the d 13 C, the r 2 was lower (r 2 > 0.9 for d 13 C, r 2 ~ 0.4 for fossil fuel combustion to the overall urban CO enhancement. For example, isoprene oxidation is a highly variable source of CO because isoprene emissions depend exponentially on the ambient temperature, and the rate at which isoprene is oxidized will increase as NOx increases (Guenther et al., 1995;Carter and Atkinson, 1996). Both the temperature and boundary layer mixing will vary day to day. Using the mass balance approach described in section 2.5, we calculated the contribution to urban CO enhancements arising from oxidized VOC's during the summer months to be 38 ± 20% at tower 2, and 27±29% (1σ) at tower 3, resulting in a VOC source contribution range of 0-58% of the urban enhancements during the summer. The remainder of the CO enhancement (40-100%) is due to fossil fuel 5 combustion. The variability in the determined VOC contributions is likely due in large part to the variable emission of BVOC's (discussed above), but this is not apparent in our small dataset For this calculation, we only considered the d 18 O signatures at each tower. Due to 10 the small difference in d 13 C between fossil fuel combustion and VOC oxidation (-27.7±0.5‰ vs. -32±2‰) and the relatively large uncertainties in the ds signatures calculated from our regression analysis, no meaningful source attribution is possible with our mass balance approach. However, the shift we see in d 13 C at tower 2 is qualitatively consistent with an increased VOC oxidation source (figure 4). 15

Partitioning fossil fuel combustion and VOC oxidation in
Tower 2 exhibits lighter d 18 O than tower 3 on average, implying a larger contribution of VOC oxidized CO at this site than at tower 3. The distance between towers 1 and 2 is larger than the distance between towers 1 and 3, which allows more time for oxidation reactions to occur. Further, the footprint of tower 2 20 contains more biogenic sources than tower 3 (section 2.1; Turnbull et al., 2015), which is also consistent with an increased VOC oxidation source at tower 2.

25
The most likely sources of the oxidized VOC's present in this study are biogenically produced isoprene, methanol, and monoterpenes. BVOC's are primarily produced from deciduous and coniferous trees. These emissions exceed anthropogenic VOC emissions globally by a factor of four (Lamb et al., 1987;Guenther et al., 1991;Warneke et al., 2010). Isoprene is produced from deciduous trees during the spring 30 and summer growing season, and its emissions are estimated to comprise 50-60% of the BVOC budget, (Guenther et al., 1993(Guenther et al., , 1995Helmig et al., 1998;Harley et al., 1999). The emission of isoprene increases exponentially with temperature (Guenther et al., 1995). Within urban environments, temperatures are amplified by the urban heat island effect, therefore subjecting urban trees to higher temperatures 5 than rural trees (Oke, 1973;Takebayashi and Moriyama, 2009;Califapietra et al., 2013). Concrete and asphalt can reach temperatures of 60° C, re-radiating heat into the urban atmosphere and raising the air temperature by 5-10° C (calculated) relative to the same location were the city not present (Oke, 1973;Takebayashi and Moriyama, 2009). It is thought that isoprene emission is due to heat stress 10 (Califapietra et al., 2013).
Once isoprene has been emitted, it is rapidly destroyed through reactions with OH and ozone (O3) (Carter and Atkinson, 1996). Within polluted regions such as cities, increased OH and O3 levels result in a 30-60 minute lifetime for isoprene (Warneke 15 et al., 2010). This lifetime is short compared to what is estimated from unpolluted forest, using the global average OH mole fraction (1x10 6 cm -3 ) and the reaction rate of isoprene with OH (3.1x10 -11 cm 3 sec -1 ) (Atkinson et al., 2006) which is about 3 hours. CO is not a direct product of isoprene in the atmosphere (Carter and Atkinson 1996). Isoprene oxidizes to products such as formaldehyde, methacrolien 20 (MACR), and methyl-vinyl-keytone (MVK), which are rapidly oxidized by OH, O3, or hν to form CO (Carter and Atkinson, 1996). The resultant yield of CO from isoprene oxidation (n molecules CO:1 molecule isoprene) ranges from 1:2 in NOx depleted conditions to 3.2:1 in high NOx conditions (Miyoshi et al., 1994;Holloway et al., 2000;Duncan et al., 2007;Grant et al., 2010). 25 The other main CO-producing BVOC's are methanol, monoterpenes, and acetone. (Duncan et al., 2007). Methanol is the second largest BVOC source of CO globally, followed by monoterpenes such as α and β pinene, and lastly acetone (Duncan et al., 2007). Monoterpenes, like isoprene, have short lifetimes in the atmosphere, ranging 30 from around a minute (α-terpenine + O3/NO3) to over a day (β-pinene + O3) (Atkinson and Arey, 2003). However, methanol and acetone have much longer lifetimes than isoprene: 12 days (methanol + OH) and 61 days (acetone +OH) (Atkinson and Arey, 2003).

5
Anthropogenic volatile organic compounds (AVOC's) are also a source of CO via oxidation reactions, and in urban regions, AVOC emissions can be a larger source of VOCs than BVOC emissions (Atkinson and Arey, 2003;Borbon et al., 2013;Ammoura et al., 2014). Further, isoprene, MVK, MACR, methanol and acetone all have minor anthropogenic sources (relative to their biogenic sources), the most prominent 10 being automobile exhaust (Biesenthal et al., 1997;Cheung et al., 2014). AVOC emissions are less variable than BVOC emissions throughout the year, but some, such as evaporative emissions from gasoline processes, increase during the summer (Jordan et al., 2009). Therefore, their contribution to the CO budget must be considered. 15 However, biogenic and anthropogenic VOC sources cannot currently be separated using stable isotopes. Only a combined isotopic signature has been estimated for VOC oxidation to CO, and CO isotopic signatures associated with specific VOC oxidation pathways have not yet been quantified. Therefore, we have assessed the 20 likely CO yields from the most prevalent VOC compounds using estimated abundances and reaction kinetics (Table 4).
While BVOC emissions dominate globally, AVOC emissions are important in urban regions, specifically for O3 and secondary organic aerosol prediction modeling 25 (Warneke et al., 2007;Bourbon et al., 2013, references therein). As with BVOC emissions, AVOC emissions increase in the summer months; however, AVOC's are present year round, with smaller seasonal variations than BVOC's (Jordan et al., 2009). Several compounds, such as isoprene and methanol, have both biogenic and anthropogenic sources, though for both isoprene and methanol, the biogenic source 30 is much larger than the anthropogenic source (Singh et al., 2000;Jordan et al., 2009;Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-506 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 22 June 2018 c Author(s) 2018. CC BY 4.0 License. Wagner et al., 2014). Further, in the late 1980's, Chameides et al. (1988) showed that BVOC's were of equal or greater importance than AVOC's in Atlanta, GA. As emission controls have steadily improved over the last three decades, there has been a continual reduction in urban AVOC emissions (Dollard et al., 2007;von Schneidemesser et al., 2010;Wagner et al., 2014). The INFLUX sampling regime 5 (section 2.1) provides only 2.7 hours (on average) for oxidation reactions to produce CO, and thus only the fastest reacting VOC's will contribute to the urban enhancements (Table 4). Though AVOC mole-fractions are often elevated relative to BVOC's in urban regions, very few have short enough lifetimes to produce significant amounts of CO in this experiment (Table 4, Atkinson and Arey, 2003;10 Jordan et al., 2009;Warneke et al., 2013).
In addition to being highly reactive, a VOC must be present in high enough mole fraction to significantly impact the CO budget. Jordan et al. (2009) produced a longterm time series from rural New Hampshire for many common anthropogenic and 15 biogenic VOC's. Of those species measured, the highest summer time mole fractions were seen in isoprene (and its immediate products, methyl-vinyl keytone (MVK) and macroelien (MACR)), monoterpenes, methanol, and acetone (Table 4, Jordan et al., 2009). The largest mole fractions are seen midday to mid afternoon (Karl et al., 2003;Jordan et al., 2009;Park et al., 2011;Wagner et al., 2014). Karl et al. (2003) 20 andPark et al., (2011)  compounds, we approximated the mole fractions using the VOC:CO emission ratios from Warneke et al. (2013).
Lastly, the chemical yield of each reaction must be considered. In the case of isoprene, under high NOx conditions (such as those found in urban environments), 5 the CO yield from isoprene is 3.05:1 (61% carbon) (Grant et al., 2010) (Table 4).
Therefore, not only is isoprene oxidized rapidly, but also each oxidized isoprene molecule may produce three molecules of CO. Methanol has a chemical yield of 0.98:1 (98% carbon), and therefore produces one CO molecule per methanol molecule oxidized. 10 In order to assess the overall impact of various VOC's on the CO budget, we used mole-fraction measurements (where available) and / or emission ratio estimates to approximate the VOC budget for 20 of the most abundant biogenic and anthropogenic VOC's (Table 4). We then calculated the net loss of each VOC based 15 on their OH, O3, and NO3 reactivities (Atkinson et al., 2006;Sander et al., 2006) ( Table 4). In order to assess the maximum yield, we used high-end values of OH, O3, and NO3 mole fractions found in the literature. We took a value of 2x10 7 cm -3 for the OH mole fraction found in Park et al. (2011). Both the O3 mole fraction (7x10 11 cm -3 ) and the NO3 mole fraction (2.5x10 8 cm -3 ) were taken from Atkinson and Arey 20 (2003). Finally, we applied the chemical yield for the VOC's to CO (Altshuller, 1991;Grant et al., 2010).
Through these calculations, we determined that isoprene, methanol, monoterpenes, toluene and ethene oxidation could be significant in the CO budget, adding 6.2 25 nmol:mol of CO to the urban enhancement (Table 4) during the 2.7 hour transit time. This result is in good agreement with the VOC produced CO estimates from our isotopic analysis. Using the average CO enhancement for towers 2 and 3 (19.5 nmol:mol), the predicted VOC oxidation contribution from our isotopic analysis is 6.3 nmol:mol (section 3.1). These calculations were done using high mole fraction 30 estimates for the various oxidants, suggesting that these oxidants may be present at

Declining Anthropogenic Emissions
Anthropogenic emissions of CO and VOCs in the United States and Europe have been declining for several decades due to emission control campaigns (e.g. Bishop and Stedmann, 2008;Bourbon et al., 2013). For CO, anthropogenic emissions still 10 dominate the overall urban budget, as evidenced by this study, as well as others (e.g. EPA NEI 2011;Turnbull et al., 2015;Vimont et al., 2017). However, as early as the late 1980's, urban studies of VOCs showed that biogenic VOCs could be a significant portion of the urban VOC budget (e.g. Chameides et al., 1988). Chameides et al. (1988) also note that fast reacting biogenic VOC's, such as isoprene, can have a 15 greater impact on species produced through oxidation effects. These include O3 and CO, and thus BVOC's can contribute more strongly to the urban CO budget than anthropogenic VOC's. It is possible that urban planning has increased the number of trees within cities, although we do not have direct evidence to support this at Indianapolis. Cheng et al. (2017) found that the CO budget in the Washington DC 20 area was similarly dominated by anthropogenic CO emissions and oxidation of isoprene, with near equal contributions. They used a modeling approach comparing O3 and CO (using O3 and CO observations), a method independent of our own. They also attribute this increased influence of isoprene oxidation to the gradual decrease in anthropogenic emissions, largely in the mobile sector (Cheng et al., 2017). We 25 find that fossil emissions most likely still dominate the Indianapolis summertime CO budget. However, VOC oxidation sources can exceed anthropogenic emissions within our 1σ uncertainty, which is consistent with Cheng et al. (2017).

Conclusions 30
Atmos. Chem. Phys. Discuss., https://doi.org /10.5194/acp-2018-506 Manuscript under review for journal Atmos. Chem. Phys. Our CO isotope results from Indianapolis provide the strongest empirical evidence to date that VOC oxidation represents a major source of CO in summertime urban environments. The determined contribution of VOC oxidation to the total urban CO source ranged from 0 to 58% (1s range) depending on date and location with the remainder (42-100%) from fossil fuel combustion. While we were unable to 5 confirm the VOC source directly, oxidation of VOC's to CO in conjunction with fossil fuel emissions provides the most plausible explanation for our isotopic results.
Estimates of the likely mole fractions of different VOC's and their respective reactivities suggests that biogenic VOC's (primarily isoprene, methanol, and 10 monoterpenes) rather than anthropogenic VOCs are likely responsible for the majority of the VOC-produced CO. Throughout the year, fossil fuel emissions still dominate the urban budget. However, our study makes it apparent that during the summer, the VOC oxidation source can be of similar magnitude to the fossil fuel combustion source in an urban area and therefore must be considered as important 15 during the growing season in future urban CO inventories and studies. Finally, the uncertainties highlighted in much of this study, particularly with respect to the oxidized VOC isotopic signatures, underscore the need for continued research into CO stable isotopes, and the isotopic signatures of the CO sources.

Author contributions:
IJV performed the measurements, data analysis, and wrote the article. JCT assisted in data analysis and provided multiple coauthor revisions. VVP provided assistance with measurement issues, data analysis, and multiple coauthor revisions. PFP assisted in several of the measurements. CS provided several coauthor revisions. 25 NM and SR provided logistical support for sample collection for the measurements.
BHV and JWCW provided laboratory and equipment support.