The Measured Impact of Wildfires on Ozone in Western Canada From 2001 to 2019

The impacts on atmospheric ozone (O3) due to wildfires are difficult to characterize due to the many factors that affect O3's formation rate and the episodic nature of fire events. This study uses a very large set of air quality data (518,987 6‐hr data points) collected in Western Canada from 2001 to 2019 to determine the prevalence and severity of fire‐driven increases to measured O3 values. Wildfire events are identified using the automated Trajectory‐Fire Interception Method (TFIM), looking for interceptions between HYSPLIT back‐trajectories and wildfire hotspots. As with other studies, which have used more restricted sets of measurements, the results from this large‐scale, data‐driven approach indicate increases in the O3 mixing ratio with wildfire impact, on average ∼2 ppbv across all wildfire time periods. To understand the factors which lead to the largest increases, and to better compare to other studies looking at individual fire events, wildfire events are classified using their distance from the air quality measurement location, time of measurement, and corresponding PM2.5 value. Increases to O3 are largest during the daytime, when fires occur close to the air quality measurement, and with corresponding measurements of PM2.5 > 25 μg/m3. When an upper‐limit correction for the bias in UV photometric detection of ozone with MnCl2 scrubbers is applied, the analysis still yields a persistent increase in O3 during wildfires except for the highest PM2.5 levels. However, a more accurate correction to the potential bias is needed to fully understand the magnitude of the impact of wildfires on O3.


Introduction
Wildfires present a serious environmental concern, heightened by the lengthening burning season and by their increasing frequency and size (Abatzoglou & Williams, 2016;Abatzoglou et al., 2019;Kirchmeier-Young et al., 2019;Williams & Abatzoglou, 2016).The environmental impacts of wildfires are not contained to the area burned, with pollutants emitted into the atmosphere being transported thousands of kilometers from the burn site.The transported wildfire plume includes commonly monitored air pollutants, including carbon monoxide (CO), volatile organic compounds (VOCs), nitrogen dioxide (NO 2 ) and particulate matter smaller than 2.5 μm in size (PM 2.5 ), which lead to deteriorated air quality for populations downwind of the fire (Jaffe et al., 2020).As the wildfire plume is transported, chemistry within it leads to the formation of secondary pollutants, such as ozone (O 3 ).
The net formation of O 3 depends on many factors, making it complex to understand.In particular, the emission of O 3 precursors, such as VOCs and NO x (NO + NO 2 ) is dependent on fuel type, fuel water content, and burn temperature (Andreae, 2019;Xu et al., 2021).NO x can be sequestered into reservoir compounds such as peroxynitrates, which can decompose to reform NO x and drive O 3 formation further downwind.O 3 can also be irreversibly lost through heterogeneous reactions involving both dry deposition and reactive loss to particles (Konovalov et al., 2012).Lastly, environmental factors such as the actinic flux within the plume, plume injection height, wind speed, and temperature can also impact wildfire O 3 formation (Jaffe & Wigder, 2012).
Ground level O 3 is associated with negative health outcomes, and thus is regulated and monitored.Unfortunately, evidence of short-term O 3 toxicity persists at mixing ratios lower than the Canadian and National (United States of America) Ambient Air Quality Standards (Di et al., 2017).A study examining the long-term impacts of O 3 has shown that a small increase in the annual average of the 8-hr maximum daily O 3 (MDA8) of 5 ppbv led to a 0.25 year decrease in life expectancy (Li et al., 2016).So, while it is challenging to understand the impact of wildfires on O 3 , the health impacts from increased O 3 cannot be ignored.
Previous studies determining wildfire influence on O 3 mixing ratios have relied on a wide variety of groundbased measurements, satellite measurements, models, and emission estimates (Buysse et al., 2019;Di Carlo et al., 2015;Dreessen et al., 2016;Jaffe & Wigder, 2012;Jaffe et al., 2013;Kalashnikov et al., 2022;Lindaas et al., 2017;McClure & Jaffe, 2018;Moeini et al., 2020;Ninneman & Jaffe, 2021;Pollack et al., 2021;Rickly et al., 2023;Robinson et al., 2021;Sharma et al., 2022).In particular, they can be largely grouped into studies which are based on O 3 observations and some measure of wildfire influence, and others that mechanistically probe the chemistry of individual fire plumes.Determining wildfire influence is challenging, however, and the increases in O 3 mixing ratios relative to natural variability can be hard to identify.Previously, we developed the Trajectory Fire Interception Method (TFIM), which assesses wildfire influence by quantifying the interceptions of air mass trajectories with upwind wildfires.TFIM is fully automated, can be applied to any location, and does not rely on emission databases (Schneider et al., 2021).Thus, it can probe for air pollution impacts within routinely monitored air quality data across many data sets and locations.The goal for this study is to use the TFIM to assess the effects of wildfires on measured ground-level O 3 .We demonstrate the relative influence of factors such as time of day, fire event severity, location, and proximity on wildfire driven O 3 enhancements.In comparison to previous work which has tended to focus on individual fires or seasons, this study uses a very large data set in a heavily impacted area collected over 19 years at 73 locations with many fire events.It focuses on air quality in Western Canada, whereas past studies have largely been focused on the United States.In general, the TFIM approach can help policy regulators and scientists identify time periods where measured O 3 is affected by wildfires.

Determining Wildfire Influence
Wildfire influence was determined using the trajectory-fire interception method (TFIM) developed previously (Schneider et al., 2021).Briefly, the method spatially compares 72-hr HYSPLIT back-trajectories (NARR meteorology) to the location of wildfire "hotspots" as detected by the MODIS satellite-based instruments.The MODIS Aqua and Terra satellites provide hotspots at a resolution of 1 km × 1 km, which is matched to the NARR HYSPLIT back trajectory with a spatial resolution of 50 km.Thus, we identify an interception if a HYSPLIT trajectory and a hotspot are within 0.5°of each other.Interceptions are calculated over a 6-hr period, where one back trajectory is initialized at measurement sites for every hour of the averaging time (i.e., 6 back trajectories).As validated in our earlier work by examining relationships to both CO and PM 2.5 , a wildfire period is defined as a 6-hr period with at least 20 interceptions.O 3 , PM 2.5 , NO 2 , NO, and CO measurements in the National Air Pollution Surveillance (NAPS) program are made hourly and averaged to 6-hr periods.The 6-hr time periods are set in UTC.The "morning" period includes the measurements between 12 and 17 UTC (6-11 MDT: Alberta stations or 5-10 PT: British Columbia stations).The "afternoon" period includes the measurements between 18 and 23 UTC (12-17 MDT: Alberta stations or 11-16 PT: British Columbia stations).The "evening" period includes the measurements between 24 and 5 UTC (18-23 MDT: Alberta stations or 17-22 PT: British Columbia stations).The "night" period includes the measurements between 6 and 11 UTC (24-5 MDT: Alberta stations or 23-4 PT: British Columbia stations).
Changes were made to the original TFIM analysis to extract the time in the HYSPLIT back-trajectory which intercepted the wildfire hotspots detected by MODIS.The average of all the interceptions was taken as a measure of the lifetime of the wildfire plumes in the atmosphere, from emission to detection.This average interception time could represent the mixture of freshly emitted and aged smoke if both events occurred along the backtrajectory.The latitude and longitude of the HYSPLIT back-trajectories were screened for marine influence (i.e., whether they were originating from over the ocean).If over half the HYSPLIT back-trajectories points were ocean influenced, they were identified as "ocean influenced," that is, O 3 increases were only evaluated by comparing wildfire impacted marine data points to non-wildfire impacted marine data points.

Applying the TFIM
The TFIM for determining wildfire influence was applied to ambient air quality measurements at 73 different locations across Alberta and British Columbia (Canada) during the wildfire season (May-October) with data from 2001 to 2019.Data were downloaded from the NAPS program (https://data-donnees.ec.gc.ca/data/air/monitor/ national-air-pollution-surveillance-naps-program/Data-Donnees/?lang=en, last accessed April 2022).Only years and stations with >75% data completeness were included in the analysis.The air quality stations included in the NAPS program are characterized in terms of source influence.For this study, we only looked at stations designated as "PE" sites which represent general population exposure without transportation or point sources nearby and "RB" sites which are regional background sites outside of urban areas.A full description of each site type, the pollutants measured, and the years included are presented in Table S1 of the Supporting Information S1.A map of the sites is included in Figure 1.More details on the TFIM analysis are illustrated in Figure S1 of the Supporting Information S1.
Wind direction measurements were obtained using the "worldmet" R package (v0.9.8, https://CRAN.R-project.org/package=worldmet, last accessed September 2023), which accesses NOAA's Integrated Surface Database.The closest meteorological measurement to the air quality station is taken and averaged to 6-hr time periods which match with the TFIM time periods.

Using Effect Size (Cohen's d)
A statistical challenge to interpreting the results arises from our large sample size so that the calculated O 3 increases during defined wildfire periods are always significantly different using Student's t-test (p ≤ 0.0001) from the non-wildfire periods due to the large sample size of each population (Equation S1 in Supporting Information S1).The qualitative increase in O 3 from wildfires is well documented, so instead of discussing whether there is a statistical difference during wildfire-influenced periods versus non-wildfire periods, the effect size is presented.This offers a different, meaningful interpretation of the wildfire effect beyond statistical significance of p < 0.05 (Wasserstein et al., 2019).
The effect size quantifies the magnitude of the effect (in this case, wildfires as defined by the TFIM) on a population relative to the variability within each population.Cohen's d generates an effect size by comparing the difference between the means of two populations (e.g., O 3 with and without a TFIM-defined wildfire influence) relative to the standard deviation within those populations.Quantitatively, the effect size represents the degree of overlap between the two populations, in this case time periods with and without wildfires.In this context, a large, positive effect size represents a higher likelihood of O 3 increasing above the range of what is typically observed without wildfires.The calculated effect sizes (here, Cohen's d, ES2) are often discussed in qualitative terms, that is, effect sizes are described as "negligible" if Cohen's d <0.20, "small" for Cohen's d ≥0.20 and <0.5, "moderate" for Cohen's d ≥0.50 and <0.8, and "large" for values ≥0.8 (Sawilowsky, 2009).These definitions apply only to the effect size value and not the impact of the effect.For example, in urban areas, the mean increase of O 3 during all wildfire-influenced periods is 2 ppbv and it has a "small" effect size of 0.30.This should not be interpreted to mean that the environmental impact of the 2-ppbv increase of O 3 is "small."For this study, the magnitude of the effect size is used to understand the effect of different types of wildfire events on measured O 3 .The calculation of the effect size is described in Supporting Information S1 (Text S2).

Correcting for O 3 Measurement Bias During Wildfires
As noted by Long et al. and Bernays et al., interferences in UV-absorption O 3 measurements during wildfire periods can be prone to interferences during severe wildfires, as defined by elevated CO > 200 ppbv (Bernays et al., 2022;Long et al., 2021).To address this, the background mixing ratio of CO is calculated for each station and for each year of measurement by taking the average of all non-wildfire influenced periods.During wildfire periods, the difference between the measured CO and the background CO is calculated.Thus, the correction is only applied in this data set if there is an increase in CO larger than 200 ppbv during wildfire periods relative to the background.The correction is the lower limit presented in the Long et al. paper (but higher than the correction in Bernays et al.), that is, a reduction of 16 ppbv O 3 /ppmv CO applied to the measurement of O 3 (Long et al., 2021).

Determination of O 3 Background
Figure 2a shows the application of the TFIM to the PM 2.5 measurements at one rural site, Caroline (NAPS ID = 91909), in 2018.At that site in 2018, there was a large wildfire event in late August which increased the PM 2.5 concentration well above the average PM 2.5 without fire influence, in addition to smaller fire events closer to the average PM 2.5 measurements without wildfire influence earlier in the season.In the case of PM 2.5 , it is clear from the TFIM and the raw time series data when there is a large wildfire influence but is less clear when there is a small one from the PM 2.5 .In contrast, Figure 2b, representing the same period as Figure 2a, demonstrates that the effect of even the largest wildfire on the measured O 3 mixing ratio is less obvious.
The dashed gray line in Figure 2b shows the mean of O 3 for periods from May to September without wildfires in Caroline in 2018.In our previous work, the dashed gray line was the value that was used to compare the measurements of O 3 during wildfire-impacted periods (Schneider et al., 2021).From Figure 2, it is apparent that this average does not capture the dependence of O 3 on seasonality and time of day, which are related to a number of factors including meteorology and regional background values.To better represent the baseline of O 3 without wildfires over a season, a simple statistical regression model was used.In particular, the linear regression of the measured O 3 for time periods without wildfires was calculated every year and measurement time for each air quality station, and whether the air mass was ocean influenced or not.The linear regression for Caroline in 2018 is shown as the yellow line for the "daytime" period (18-24 UTC, 12-18 MDT, local time) and the black line for the "nighttime" period (6-12 UTC, 24-6 MDT, local time).For each wildfire-influenced period, the difference in the measured O 3 value relative to the calculated O 3 baseline from the linear regression is obtained.We note that this statistical modeling approach does not control for all factors which affect O 3 , but we assume that these factors occur in both the wildfire and non-wildfire periods equally.The decreasing ozone baseline during the wildfire season is consistent with the ozone climatology in Western Canada (see Figure 3 of https://www.alberta.ca/airindicators-ground-level-ozone,accessed October 2023).
Statistical modeling has been previously used to understand the effect of wildfires on O 3 and can account for variability arising from changes in temperature, solar irradiance, relative humidity and more (Brey & Fischer, 2016;Cisneros et al., 2020;Gong et al., 2017).Briefly, these statistical methods relate the measured O 3 to factors which affect the O 3 mixing ratio to determine an unbiased baseline to compare to the measurements during wildfires.However, defining a wildfire period remains a challenge.In this work, only factors which can be extracted from the available air quality data and the TFIM are used to make the method as consistent and as broadly applicable as possible.

Interpreting the Impact of Wildfires on O 3
From Figure 3, the TFIM definition of wildfire influence shows an increase in measured O 3 above the baseline without fires of 2 ppbv, with a small effect size (0.27).Overall, wildfire-influenced times represent ∼6.4% of the total time periods studied.The heterogeneity of wildfire events makes interpretation of O 3 changes difficult.To combat this, wildfire events are categorized more precisely than "wildfire influence" and "no wildfire influence" by looking at additional factors that may affect ozone.From the data produced by the TFIM, the impact of fire event severity, fire location, and measurement time on the calculated increase of O 3 at air quality stations across Alberta and British Columbia is assessed to identify the extent to which each factor perturbs the measured O 3. This aggregate method of identifying and classifying wildfire events increases the statistical power of the study, however other, unspecified factors which affect O 3 may be hidden with this approach.Additionally, the averaging of O 3 measurements to 6-hr periods could mask shorter episodic effects of wildfires.
To illustrate these effects in Figure 3, if all wildfire-influenced periods are sampled for only evening measurements, a larger increase of 3 ppbv is observed (small effect size 0.41).To further sample evening measurements for wildfire-influenced periods with an average atmospheric lifetime of 12 hr, the increase rises to 4 ppbv, a moderate effect size (0.51).Lastly, this population is sampled for PM 2.5 measurements greater than 50 μg/m 3 and below 100 μg/m 3 (PM 2.5 filter), to give an increase of 11 ppbv which represents a large effect (1.49).Each effect is explored and defined in depth below.
To explore the effect of measurement time in depth, four time periods (introduced above) are defined to match with the TFIM 6-hr average periods.The increase in O 3 measured during wildfire periods at the different measurement times is summarized in Figure S2 of the Supporting Information S1.The afternoon and evening times show small effect size increases, with average O 3 increases of 2 and 3 pbbv, respectively (effect sizes 0.28 and 0.41).Night and morning periods show small and negligible effects, with average increases of 2 and 1 ppbv (effect sizes 0.25 and 0.13).These results are expected, considering O 3 is produced photochemically.
The TFIM can quantify the average atmospheric lifetime of the wildfire plume which is influencing measurements by examining when the interceptions between the "hotspots" and the HYSPLIT back-trajectories are occurring.The dependence of atmospheric lifetime could be as simple as the effect of dilution from transport, however it is also understood that maximum O 3 production occurs a few hours after emission for daytime plumes (Langford et al., 2020;Robinson et al., 2021).To investigate the effect of atmospheric lifetime on measured increases to O 3 during wildfire periods, smoke plumes with 12-hr average lifetimes (15 > average interception time >9), 24-hr average lifetimes (27 > average interception time >21), 36-hr average lifetimes (39 > average interception time >33), 48-hr average lifetimes (51 > average interception time >45) and 60-hr average lifetimes (63 > average interception time >60) were examined.The 6-hr interception windows correspond to the 6-hr time periods defined by the TFIM to give the average atmospheric lifetime.Average atmospheric lifetimes of 12 hr show a small increase of 3 ppbv of O 3 (effect size 0.22), which is the highest average increase.Higher O 3 increases with less atmospheric transport correspond to less dilution of the plume as it is transported toward the monitor.Interestingly, atmospheric transport between 24 and 60 hr consistently shows increases of 2 ppbv, with no further observed effects from dilution or other factors related to atmospheric lifetime (negligible effect sizes of 0.1).These results are summarized in Figure S3 of the Supporting Information S1.
Lastly, O 3 measurements are often co-located with measurements of PM 2.5 .PM 2.5 can serve as an indicator for the severity of the fires since it is a primary wildfire pollutant.Previous studies have shown that O 3 increases linearly with PM 2.5 until a value of ∼60 μg/m 3 , after which it is independent of PM 2.5 (McClure & Jaffe, 2018).O 3 production during periods with high PM 2.5 can be limited through reduced irradiance or increased PM 2.5 surface area which acts as a heterogeneous sink (Baylon et al., 2018;Buysse et al., 2019;Konovalov et al., 2012;McClure & Jaffe, 2018).More recently, it was shown that high PM 2.5 can also be a loss pathway for HO 2 , an important intermediate for photochemical O 3 production (Ivatt et al., 2022).In this data set, a similar relationship between increased O 3 and measured PM 2.5 was observed and is presented in Figure S4 of the Supporting Information S1.The largest average increases in O 3 of 5 and 6 ppbv were observed when PM 2.5 was above 75.5 μg/m 3 and below 105 μg/m 3 .This corresponds to a large effect size (0.69-0.79) on increasing O 3. O 3 increases were lower with both higher and lower PM 2.5 , with effect sizes ranging from negligible to moderate.Recently, measurements of O 3 at high PM 2.5 are found to be prone to interference, depending on the method of O 3 detection (Bernays et al., 2022;Long et al., 2021).This important issue is discussed in more detail in Section 3.5.As shown in Figure S4 of the Supporting Information S1, the most consistent predictor of O 3 increase is fire event severity.The TFIM is more sensitive to small wildfire events compared to Canadian wildfire emission models for PM 2.5 , even if the impacts on air quality are small or even negligible (see Figure 4, Schneider et al., 2021).However, even for wildfire events where the measured PM 2.5 is less than the Canadian Ambient Air Quality Standard (CAAQS) guideline of 25 μg/m 3 , there is still an increase in O 3 between 1 and 3 ppbv.In Figure 3 we illustrate the additive effect of all these factors, by identifying the collection of factors which causes the largest increase in measured O 3 .
Comparison to other studies is not straightforward, since our value of 2 ppbv represents the average of over 8,000 days of wildfire observation from across Western Canada in two decades, across all time periods.This is a different approach than most studies, which compare the maximum daily average 8 hr (MDA8) for wildfire and non-wildfire periods.The MDA8 captures the photochemical maximum of O 3 , however it does not consider any additional contribution of wildfires to O 3 outside of those 8 hr.The MDA8 is what is reported for NAAQS and CAAQS air quality targets for O 3 .One study investigating the effect of wildfires on the MDA8 used chemical and statistical models to compare measurements of O 3 during two smoke events to quantify an increase between 19 and 60 ppbv from these specific events.A similar approach using statistical models investigated a larger data set with 8 U.S. cities over 7 years to estimate average contributions between 3 and 8 ppbv to the MDA8 from wildfires (Gong et al., 2017).Identification of wildfire-impacted days using NOAA's Hazard Mapping System (HMS) reported increases in MDA8 values in the U.S. during wildfire periods ranging from 3 to 36 ppbv (Brey & Fischer, 2016;Pan & Faloona, 2022).The most similar study to this one measured the increase in O 3 in California across all times (i.e., not just the MDA8) between 2006 and 2016 from wildfires identified using the HMS (Cisneros et al., 2020).This study noted a median increase of 1.5 ppbv in O 3 for all years (except 2008, which increased a median amount of 7.8 ppbv) which is very comparable to our study with median values of 2 ppbv for all wildfire periods, and median values of 8 ppbv for O 3 measurements in the evening of smoke plumes transported an average of 12 hr, with PM 2.5 measurements above 50 μg/m 3 and below 100 μg/m 3 .Modeling of a 2019 Washington State wildfire with WRF-Chem/DART which used measurements from British Columbian and Albertan NAPS stations concluded O 3 increases of 3-5 ppbv near the wildfire, and 2-3 ppbv further from the fire (Pouyaei et al., 2023).

Effect of NO x (NO + NO 2 )
Wildfires are a source of two O 3 precursors, NO x and volatile organic compounds (VOCs).As the plumes are transported, it is understood that they quickly become NO x limited, with no increase in NO 2 after ∼3 hr (Jin et al., 2023;Peng et al., 2021).And so, the interaction of the NO x -limited plume with a NO x -rich urban area has the potential to cause O 3 formation and lower urban air quality (Apel et al., 2015;Brey & Fischer, 2016;McClure & Jaffe, 2018;Ninneman & Jaffe, 2021;Rickly et al., 2023;Xu et al., 2021).In particular, in NO x -rich urban areas wildfire plumes could increase local production of O 3 on top of the O 3 formed via the transported plume chemistry (Lindaas et al., 2017).As the wildfire plume is transported, it interacts with the local photochemical O 3 production regime of the air quality monitoring site, which could be either VOC or NO x limited.A recent study in Boulder, Colorado concluded that in NO x -limited urban environments, observed increases in O 3 from wildfires arose solely from transported O 3 produced from plume chemistry and not from additional photochemical local formation of O 3 (Rickly et al., 2023).Conversely, cities in California were predominantly VOC-limited, and aged biomass burning plumes that were low in NO 2 but high in VOCs enhanced local O 3 production (Jin et al., 2023).
At ambient air quality stations in Western Canada, VOCs are not regularly measured as opposed to NO x, which is routinely monitored.Thus, we compared the increase in O 3 as a function of measured NO x during non-wildfire periods and wildfire periods in Figure 4. Since NO x was not a factor that was used to calculate the background of O 3 , the change in O 3 from the background during non-wildfire periods can be investigated to understand the average photochemical regime in Western Canada.During the non-wildfire periods, the O 3 is decreasing relative to the calculated background with increasing NO x , which suggests that the average air-quality monitoring station is in a VOC-limited O 3 production regime.Previously, NO x -limited O 3 production regimes have been identified by comparing weekday-weekend trends at the same air quality monitoring station, based on decreased O 3 with decreased NO x (Pollack et al., 2021).This approach is not as precise as measuring the VOC/NO x ratio directly, but without measurements of VOCs it can provide some insight into the average sensitivities of the O 3 production regimes.From Figure 4, the mean increase in O 3 for all wildfire periods is higher compared to average nonwildfire periods, indicating that wildfires are enhancing the measured O 3 at all locations, and the magnitude of the dependence has no clear relationship with NO x .

Case Study: Exploring Urban Plumes Interacting With Wildfire Plumes
It is hypothesized that urban emissions mixing with wildfire plumes cause the largest increases to O 3 within and downwind of urban areas since the generation of O 3 within a plum can be increased with urban emissions (Lindaas et al., 2017).To explore the relationship between urban centers and nearby rural areas, the Edmonton area in 2017 and 2018 is investigated.This area was chosen because of the high concentration of O 3 monitors within the city, and both east and west of the city.2017 and 2018 were years with many wildfire-influenced periods, thus there is the most data available.Time periods were filtered for westerly winds (between 200 and 320°) to ensure that the measurements were of air masses occurring from the west (the location of the fires), before interacting with the Edmonton air and heading east.The results of this comparison are shown for the daytime measurements in Figure 5.The stations are shown on a map in Figure S5 of the Supporting Information S1.
From the top panel in Figure 5, it is clear that there is a dependence on the background O 3 measurement as a function of wind direction, since the mean is not 0 ppbv.However, the wildfire-influenced results can be interpreted relative to the specific non-wildfire period measurements to explore the difference more precisely.For both non-wildfire and wildfire time periods, the calculated O 3 decreases as the air mass moves from the westward stations, through the city to the eastward stations in 2017, and for the non-wildfire periods in 2018.During the wildfire periods in 2018, there is a calculated decrease in O 3 from the upwind stations to the urban stations, however the O 3 increases slightly as the air mass moves downwind.
Measurements of NO 2 (Figure S6 in Supporting Information S1) and NO (Figure S7 in Supporting Information S1) are higher in the urban stations relative to the upwind stations (increasing 2.9 and 1.9 ppbv, respectively), and are slightly increased downwind of the urban center compared to the upwind stations (increasing 0.55 and 0.15 ppbv, respectively) during non-wildfire periods.This highlights that urban plumes are higher in NO x than rural plumes upwind of cities, and increased NO x is observed downwind of the city, but is diluted compared to urban measurements.During the 2018 wildfires, NO 2 and NO increases at all the stations relative to the background without wildfires, which suggests that the wildfire plume itself was rich in NO x .This was not observed in 2017, with the NO x decreasing slightly upwind of the urban stations during wildfire periods and elevating in the urban and downwind areas.From Figure 4, there is not a strong dependence of O 3 increase as a function of measured NO x , which may be what is being observed in the Edmonton case study as well.
Overall, we do not see strong evidence for O 3 increase within or downwind of an urban center during wildfire periods.We can not discern whether increased production is offset to some degree by additional loss processes.In particular, the possibility of decreasing O 3 as a result of wildfire plumes has been previously described, potentially driven by reduced solar flux, additional loss of O 3 onto particulate matter, or changes to O 3 photochemical regimes (Baylon et al., 2018;Ivatt et al., 2022;Konovalov et al., 2012;McClure & Jaffe, 2018).

O 3 Measurement Uncertainties and Limitations
Recently, a study by Long et al. compared the bias for different UV photometric methods to measure O 3 during wildfire periods (Long et al., 2021).We that a scrubber type was misidentified as MnO 2 when it was in fact MnCl 2 in some sensors (Bernays et al., 2022).Ultimately, the results appear to indicate that MnCl 2 scrubbers show positive bias during wildfire periods with CO > 200 ppbv, for reasons that are not fully understood but may be due to VOC absorption and scrubbing within the unit (Bernays et al., 2022;Long et al., 2021).The authors from both studies suggested ranges of O 3 reduction corrections based on measurements of CO, ranging from 12 ppbv O 3 /ppm CO to 24 ppbv O 3 /ppm CO.In Alberta and British Columbia, the approved instruments for measuring O 3 include Thermo 49, 49C, and 49i, the Teledyne API T400 and the Serinus Ecotech 10 instruments (Alberta Environment Air Monitoring and Audit Centre, 2011; Province of British Columbia Ministry of Environment and Climate Change Strategy, 2020).The Thermo 49C and 49i were explicitly tested in the previous publications to show a bias.The Serinus Ecotech 10 was also tested to show no bias.From the user manual, the Teledyne API T400 uses a MnO 2 scrubber, so we assume it is free from bias (Teledyne API, 2018).Unfortunately, the information for the exact instruments which are available at each monitoring station over the two decades that measurements were made is not freely available.Thus, to calculate the maximum correction we assume that every data point has a potential bias, which we correct with the moderate reduction of 16 ppbv O 3 / ppm CO in Figure 6.The bias is corrected to the wildfire CO specifically, by calculating the difference in CO measured during the wildfire periods relative to the non-wildfire background of CO at each station (see Section 2).
Because not every instrument may exhibit the bias, the results of the correction represent an upper limit for the bias in O 3 measurements during wildfire periods described above.The increase in O 3 remains for all the wildfire periods except for more severe events where PM 2.5 is between 50 and 100 μg/m 3 .Figure S8 in Supporting Information S1 illustrates the application of the correction as a function of PM 2.5.As both PM 2.5 and CO are considered primary pollutants from wildfires, the correction is largest at higher measurements of PM 2.5 .Since the strongest effect sizes of wildfires are observed during periods of high PM 2.5 , or during more severe wildfire events, the appropriate correction for the bias is needed to fully understand the magnitude of the largest measured impacts of wildfires on O 3 .
More work is needed to fully characterize and develop corrections for bias in the instruments used here, as to best utilize the rich O 3 data set collected over the last two decades.We note that only 43% of O 3 monitors are colocated with CO monitors, and only 6% of those sites are rural background sites.All but two urban sites are co-located with a PM 2.5 monitor, which allows comparison to observations of PM 2.5 .

Conclusions
Using a data-driven approach coupled to the TFIM, this paper assesses the factors that drive the response of O 3 mixing ratios to wildfires in 73 urban and rural locations in Western Canada from 2001 to 2019.The average wildfire impact across all sites and time periods is an increase of approximately 2 ppbv.There are many environmental factors which affect the average increase in O 3 , including measurement time, atmospheric lifetime, and fire event severity (approximated by PM 2.5 measurements).O 3 measurements made in the evening (18-23 MDT: Alberta stations or 17-22 PT: British Columbia stations) of wildfire plumes with an average lifetime of 12 hr in the atmosphere and with PM 2.5 increases between 50 and 100 μg/m 3 show the largest average increase of 11.4 ppbv without accounting for potential interferences.While these average increases are potentially important, they are generally low compared to prior studies that focused upon maximum ozone increases during intense wildfire episodes.Our assessment is across the whole wildfire season.The response of O 3 mixing ratios via wildfire influence can depend on the interactions between wildfire plumes and local chemical environments, such as NO x -rich urban regions.From this data set, the average O 3 consistently increased with NO x , independent of the NO x mixing ratio.However, without measurement of VOCs, it is difficult to identify the specific photochemical regime of each air monitoring station to evaluate the relative impacts of wildfires for different regimes.We also explored the effect of urban centers on regional O 3 changes during wildfire periods using observations from stations upwind (west), downwind (east), and within Edmonton.From these observations during two wildfire seasons with high frequencies of wildfire events, unique changes to O 3 were observed for each group.No specific downwind effect is observed for this case study.
An emerging issue in monitoring O 3 is the bias in UV photometric detection methods which use MnCl 2 scrubbers (Bernays et al., 2022;Long et al., 2021).Here, we provide an upper-estimate correction to this data set by assuming all the instruments are subject to a bias of 16 ppbv O 3 /ppm CO.Even with this correction, the increase in O 3 persists, except for severe wildfire events with PM 2.5 measurements between 50 and 100 μg/m 3 .This result highlights the need for transparency in measurement methods and appropriate corrections to make use of the last two decades of O 3 monitoring.The corrections here are limited by specific instrument information, the lack of corrections for certain instruments, and the lack of CO measurements in tandem with O 3 measurements.When taking the bias into consideration, while there is still an influence of wildfires on O 3 , the average effect may not be that large as previously thought.

Data Availability Statement
Detailed air quality station data information is described in Table S1 of the Supporting Information S1.Additional details on the TFIM are available (Figure S1 in Supporting Information S1).The change in O 3 as a function of measurement time (Figure S2 in Supporting Information S1), atmospheric lifetime (Figure S3 in Supporting Information S1), and fire severity (Figure S4 in Supporting Information S1) are also presented.Supporting Information S1 for the Edmonton case study includes the station map (Figure S5 in Supporting Information S1), the measured NO 2 at the stations (Figure S6 in Supporting Information S1), and the measured NO at the stations (Figure S7 in Supporting Information S1).The O 3 bias correction as a function of PM 2.5 is included as Figure S8 in Supporting Information S1.The Student's t-test and Cohen's d equations, as well as how they are calculated in R, are presented in Text S2 of the Supporting Information S1.The data used to make the Figures and Tables is available online as a .csvfile (Schneider, 2023).

Figure 1 .
Figure 1.A map showing the locations of all the population exposure (PE) and rural background (RB) air quality stations across Alberta and British Columbia, Canada.The red points represent urban sites, and the blue points represents rural background sites.The map was generated using Stadia Design, under Creative Commons Attribution (CC BY 3.0) license.

Figure 2 .
Figure 2. Panel (a) shows the measured PM 2.5 concentration in Caroline (NAPS ID = 91909) from May to October 2018, colored red when there is a TFIM fire influence and black for measurements without TFIM fire influence.Panel (b) is showing the O 3 measurement at the same site for the same time period, colored for the hour range measurements were made.The large points represent wildfire influenced measurements.Black points represent 6-12hr UTC (24-6hr local time, UTC-6 "nighttime") and yellow points represent 18-23hr UTC (12-18hr local time, UTC-6 "daytime").The gray dashed line represents the average O 3 concentration without fires for the season.The solid yellow and black lines represent the moving average of O 3 without wildfires over the season, for each respective time period.We note that the evening and morning time periods discussed below are not illustrated here.

Figure 3 .
Figure 3. Violin plot showing the distribution of increase of O 3 relative to the calculated background of O 3 with increasing specificity of wildfire event type.The width of the violin plot indicates the frequency of the value, and the mean and standard deviation are represented by the boxplot inside the violin.The numbers above the violins represent the average O 3 increase in ppbv (top number) and the total number of data points (n, bottom number), where one data point is a 6-hr average.The meanings behind the terms on the horizontal axis are given in the text.The populations chosen here represent the measurement time, atmospheric lifetime, and fire severity which give rise to the highest effect size.

Figure 4 .
Figure 4. Calculated increase in O 3 as a function of measured NO x (NO + NO 2 ), for wildfire (red) and non-wildfire (gray) periods for stations with NO x measurements.The center line of the boxplot represents the median, the outer box represents the standard deviation, and the whiskers represent the 95th and 5th percentile values.The top number above the boxplots represents the mean increase of O 3 in ppbv and the bottom number represents the number of observations (n), where one observation is a 6-hr average.The effect size is included under the box plot.The difference between the means of the wildfire influenced periods and the non-wildfire periods is represented by the black circles.The rounded parentheses indicate the value is included in the range, and the square parentheses indicate that the value is not included.

Figure 5 .
Figure 5. Violin plots showing the distribution of increase of O 3 relative to the calculated background of O 3 for rural stations west of Edmonton (red), urban stations in Edmonton (green) and rural stations east of Edmonton (blue) during the daytime.The top plot indicates the distribution of O 3 relative to the background for time periods without wildfires, and the bottom plot indicates the distribution of O 3 relative to the background for wildfire-influenced periods.This comparison is shown for 2 years, 2017 and 2018.

Figure 6 .
Figure 6.Violin plot showing the distribution of increase of O 3 relative to the calculated background of O 3 with increasing specificity of wildfire event type.The width of the violin plot indicates the frequency of the value, and the mean and standard deviation are represented by the boxplot inside the violin.The numbers above the violins represent the average O 3 increase in ppbv (top number) and the total number of data points (n, bottom number), where one data point is a 6-hr average.The green violins represent the calculated increase for only stations with CO measurements, but without applying the correction.The pink violins represent the calculated increase with a reduction of 16 ppbv O 3 /ppm CO to account for instrument bias.