Assessment of the Contribution of Wildfires to Ozone Concentrations in the Central US-Mexico Border Region

The annual trends and spatiotemporal patterns of monthly 8-hour maximum ozone (8-hr max O3) concentrations in the Paso del Norte region were analyzed, and their associations with fires were examined for the 2001–2010 period. Hourly O3 measurements were retrieved from the Environmental Protection Agency (EPA) Air Quality System, while the times and locations of fires were acquired from the MODIS fire detection module. The absolute 8-hr max O3 concentrations were comparable in urban, rural and background sites. Time series analysis of deseasonalized monthly 8-hr max O3 levels showed statistically significant declining trends for most of the sites located in populated areas, and high correlation coefficients among these. Conversely, a 0.12 ppbv/yr increase of 8-hr max O3 concentration was computed for Chiricahua, a background site located in a Class I protected area. Strong relationships between the monthly 8-hr max O3 concentrations and categorical variables representing the number of fire detections for each month in six buffer zones were computed. Fire incidents near the sites (within 400 km) in central Arizona, central Texas and western Mexico triggered a decrease in the 8-hr max O3 concentration by 1 to 12 ppbv in urban and rural sites, and an increase of 3 ppbv in Chiricahua. Conversely, fire incidents in southeast US, Cuba and central Mexico contributed from 5 up to 19 ppbv. These findings indicated that regional fire incidents may trigger high O3 episodes, which may exceed air quality standards.


INTRODUCTION
The World Health Organization (WHO), the US EPA and the European Commission enacted the air quality threshold values for atmospheric O 3 to protect human health.In the US, the latest National Ambient Air Quality Standards (NAAQS) revision in 2008 requires that the three-year average of the annual fourth-highest daily 8-hour maximum (8-hr max) average at every ozone monitor is less than or equal to 75 ppbv (from 80 ppbv enacted in 1997) (US EPA, 2008).Non-attainment areas (exceeding the O 3 standard) include metropolitan urban and surrounding communities across the country and rural regions in the vicinity of oil and gas exploration activities (Schnell et al., 2009;EPA, 2011).
The formation of O 3 during transport from upwind natural and anthropogenic sources of NO x and VOCs is an important contributor to elevated ozone concentrations at many urban and continental background sites.Emissions of O 3 precursors from wildfires frequently exceeded emissions from lightduty gasoline vehicle emissions (Junquera et al., 2005).Intercontinental transport of fires emissions in Siberia, Asia and Alaska increased the O 3 levels in mainland US and over Europe (Jaffe et al., 2004;Pfister et al., 2006).An increasing trend in the O 3 concentrations at Class I protected areas in the US was also observed (NPS, 2009).This trend was initially attributed to increased temperatures but recent studies indicated that changes in the spatial distribution and intensity of anthropogenic emissions and increased wildfires emissions may also be responsible (Jaffe and Widger, 2012).
Fire frequency and intensity are extremely sensitive to changes in land use practices and El Nino-Southern Oscillation with increases in the number and sizes of wildfires.Over the past two decades, burned areas in the U.S. and Canada exceeded 22,000 km 2 /year and 60,000 km 2 /year, respectively, which are about three times higher than the 1920-1980 period (Stocks et al., 2002;Gillet et al., 2004;Schoennagel et al., 2004).The increase in the number and magnitude of fires is even more pronounced for forests in the western US (Westerling et al., 2006).Longer summers, which are driven by increasing temperatures, increase the fire ignition risk by 10-to-30% and are expected to increase burned areas by as much as 120% by the end of the century.These projections are worse for higher elevations where wildfire occurrences are more frequent and earlier snowmelts are extending the annual window for wildfires (Flannigan et al., 2004;Running, 2006;Westerling et al., 2006).
The Paso del Norte region encompasses the metropolitan El Paso area in Texas, the greater Las Cruces area in southwestern New Mexico and the Ciudad-Juarez metropolitan area in Mexico.Ambient O 3 concentrations in El Paso, Texas and portions of Doña Ana County (69-76 ppbv) are comparable to the 8-hr max O 3 national ambient air quality standard (NAAQS) (75 ppbv).The objectives of this study were: (i) to determine the long-term spatial and temporal trends of O 3 levels in the binational border region of southwestern New Mexico, El Paso and Ciudad-Juarez area, and; (ii) to assess the effect of wildfires on O 3 levels.It is part of larger effort to characterize sources and trends of air pollution in the region and their potential impacts on human health.

Ozone Data Acquisition and Reduction
Hourly O 3 measurements at 23 sites in the Paso del Norte region during the 1993-2010 period were retrieved from the US EPA's Air Quality System and the Clean Air Status and Trends Network (CASTNET).The boundaries of the study domain and the locations of ozone monitoring sites since 1993 are illustrated in Figs.1(a)-1(c).Table 1 presents the characteristics and monitoring periods for each site.The Chiricahua site in Arizona (ch in Fig. 1(b) is located within the Chiricahua National Monument Class I protected area and it is operated by the US National Park Service.Ten sites are operated by the New Mexico Environment Department, three of them within Las Cruces (nm3, nm4 and nm5 in Fig. 1(c)), four of them in the Sunland Park border region (nm7, nm8, nm9 and nm10 in Fig. 1(c)) and the remaining three in rural communities (nm1, nm2 and nm6 in Figs.1(b)-1(c)).Two sites (nm3 and nm5) were discontinued in 2002 and 2004, and three sites (nm2, nm1 and nm4) started after 2004.There are 7 sites within the El Paso metropolitan area and 4 sites at Ciudad Juarez in Mexico.These sites were operated by the Texas Commission on Environmental Quality.Two of the sites (ep5 and cj3) moved to other locations within a few hundred of meters in 1999 (ep6) and 1998 (cj4), respectively.These sites were treated separately to minimize the effect of differences on O 3 measurements caused by the micro-environmental conditions of the monitoring site (e.g., distance to the nearest street,orientation, prevalence of wind conditions).
The monthly 8-hr max O 3 concentration was computed for months with more than 75 percent valid estimates of daily 8-hr max O 3 concentrations for days with more than 18 valid estimates of the 8-hr running averages.The use of the mean monthly 8-hr max O 3 concentration as an indicator was corroborated by previous evidence that wildfires contribute episodically to high O 3 concentrations at receptor sites several days or weeks following the fire (Jaffe et al., 2004;Junquera et al., 2004;Pfister et al., 2006).For example, Dupont et al. (2012) recently reported that two wildfires events in Asia influenced O 3 levels in the US after 10-12 days.During the analysis, we identified cases in which fire detection was observed after the highest O 3 concentration.These cases were observed in winter (January and February) months when the highest 8-hr max O 3 concentration was low and typically defined by a low number of fire detections.These cases actually have increased the instability of the overall model but they have very little effect on the values of the regression coefficients (described in Statistical Analysis section).The 25 year time series of 8-hr O 3 design (the 4-th highest 8-hr running averages) values and the non-overlapping block five year averages of the percentage of 8-hr averages above the 2008 O 3 NAAQS value for the summer months were previously used by Sather and Cavender (2012) to evaluate the compliance with the NAAQS and to smooth the effects of local meteorology and short-term changes in local emissions.The average monthly O 3 concentration was previously used as an indicator to detect the effect of temperature variation and aloft O 3 in IMPROVE sites along the Pacific Northwest (Jaffe and Ray, 2007).

Wildfires Data Acquisition and Reduction
The locations of fires (detected as temperature abnormalities) for the North America were retrieved from MODIS Active Fire Mapping Program at 1 km spatial resolution for each day (http://activefiremaps.fs.fed.us/gisdata.php).The detections were observed by both TERRA and AQUA instruments of MODIS and processed by the United States Department of Agriculture Forest Service Remote Sensing Applications Center, the NASA-Goddard Space Flight Center and the University of Maryland (Giglio et al., 2003).Fires, as small as 100 m 2 , are detected twice every day (one daytime and one nighttime), allowing to monitor changes in fire frequency, intensity or location over the course of the day(s).Large fires are detected multiple times for every day they are active.
The date and locations of fire detections during the 2001-2010 period were extracted and the number of fires within 160, 160-400, 400-800, 800-1600, 1600-3200 and 3200-4800 km buffer zones from each site was estimated.

Statistical Analysis
The absolute (ΔC) and the relative (%ΔC/Ref) differences of monthly 8-hr max O 3 concentrations between two sites were computed to evaluate concentration gradients in the region.The O 3 monitor at Chiricahua National Monument (ch) was the reference site because of its upwind location with respect to the other sites in the study domain.The relative concentration differences were computed as the percentage of the absolute concentration difference to the reference site concentration.Positive values indicate that O 3 concentrations at the site were higher than those measured at Chiricahua NM.The coefficient of divergence (COD) was used to assess the spatial uniformity of measurements with respect to the concentration levels.
where p is the total number of paired measurements, and C ij and C ik are the measured concentrations at the reference and comparison sites on the i-th month, respectively (Pinto et al., 2004;Lianou et al., 2007).COD values vary from 0 to 1, with COD values close to unity being suggestive of strong spatial variation.
(a)  Deseasonalized concentrations were calculated using the "Census I" method integrated in SPSS.Ordinary least squares regression analysis of deseasonalized monthly 8-hr max O 3 concentrations was used to determine the trends of O 3 without the seasonal component (Jaffe and Ray, 2007).
The contribution of fires on O 3 concentrations was evaluated by multivariate regression analysis.The dependent variable was the monthly 8-hr max O 3 concentration for imonth and the independent variables were the number of fires detections (X ij ) for i-th month in j-th buffer zone.
The intercept a was attributed to ozone concentrations from non-fire local and regional sources.The significance threshold value for the regression coefficients was set at the α = 85% level.The assumptions and limitations of this approach are described in details in a later section.

Spatiotemporal and Annual Trends
Table 2 shows the annual trends of deseasonalized monthly 8-hr max O 3 concentrations for each site (ppbv/yr), the COD and the median (and standard deviation (σ)) of absolute (ΔC) and relative (%ΔC/Ref) concentration differences.The values of the COD (from 0.06 to 0.21), absolute (less than 12 ppbv) and relative (less than 15%) concentrations differences for each site using the Chiricahua site as reference indicated the absence of a spatial gradient of the monthly 8-hr max O 3 concentrations.For three sites, the %ΔC/Ref ranged from 17% to 21% (up to 14 pbbv); however, similar values were observed for sites within an urban area (Kavouras et al., 2012).The highest absolute and relative concentration differences were estimated for the site in the University of Texas, El Paso which is located downwind (north) of the El Paso-Ciudad Juarez downtown complex.The site-to-site variation of monthly 8-hr max O 3 concentrations (described by the standard deviations of ΔC and %ΔC/Ref) suggested common characteristics for El Paso-Ciudad Juarez sites.
The observed trends were statistically significant at α = 99% in fourteen sites.Most of the sites (18 of 23) showed trends from -0.07 to -1.85 ppbv/yr.The three sites within the Las Cruces urban area (nm3, nm4 and nm5), demonstrated a statistically insignificant trend from -0.07 to -0.27 ppbv/yr.For sites located on the southeast part of the Dona Ana county (nm6-nm10), the decreasing trends were higher (from -0.37 to -0.77 ppbv/yr) and statistically significant at α = 99% level.All sites at Ciudad Juarez in Mexico (cj1-cj4) experienced a decrease of more than 1 ppbv/yr on the monthly 8-hr max O 3 concentrations.In El Paso, annual trends from -0.15 to -0.51 ppbv/yr were estimated in four sites (two of them were statistically significant).Positive trends (but statistically insignificant) of +0.07 and +0.13  24) -1.00 b a Values were calculated for all data in the monitoring periods in each site (see Table 1).b significant at p < 0.01.ppbv/yr were computed for ep1 and ep6 site.Two sites showed statistically significant trends of -2.77 (ep2) and +2.38 (ep5) ppbv/yr; however these sites were decommissioned in 2000 and 1999, respectively.The annual trends of their replacement sites (ep3 and ep6, respectively) were -0.50 (a = 99.9%) and +0.07 ppbv/yr.For the Chiricahua site, a statistically significant positive trend (+0.12 ppbv/yr) was observed, in agreement with the observed increase of the monthly mean O 3 concentrations in other Class I protected areas in western United States (Jaffe and Ray, 2007).Table 3 also presents the annual trends of monthly 8-hr max O 3 concentrations in May, June, July, August, September and October, separately for four sites.In Chiricahua, insignificant trends were computed for the months of May, June, July and September (from -0.08 to +0.08 ppbv/yr).On the other hand, an increase of 0.4 ppbv/yr was estimated for August and a decline of -0.15 ppbv/yr for October.For the nm10 site, O 3 concentrations declined by -0.35 to -1.21 ppbv/yr for the May-June period.For the ep1 and cj2 sites, O3 levels followed an increasing trend on May-July (up to 1.9 ppbv/yr) and a declining pattern in August and September (up to -1.01 ppbv/yr).
The observed positive trends on both monthly mean (Jaffe and Ray, 2007) and monthly 8-hr max O 3 concentration (this study) in continental background sites are in contrast with the decreasing trends of O 3 concentrations in urban areas (Sather and Cavender, 2012) highlighting the complexity of reducing O 3 concentrations over large geographic areas.
Adjustments for meteorological conditions showed very little effect on O 3 trends in urban and continental background locations in the western US (EPA, 2012).Other factors including increasing global background O 3 concentrations due to increased emissions from developing countries in Asia and increased emissions from more frequent and intense wildfires may be important determinants of this variability.In addition, spatiotemporal changes of NO x and VOCs emissions (e.g., due to expansion of small size cities and rural communities) may also be an important determinant of the observed positive trends in background sites.In urban areas, advected aloft O 3 will be partly destroyed due to the primary emissions but overall primary emissions will increase the surface ozone concentration.
Overall, these findings indicated important qualitative differences on the annual trends and timings of high 8-hr max O 3 episodes throughout the study region and among different types of sites (urban vs. background); however, the absolute 8-hr max O 3 concentrations did not vary a lot.

Effect of Wildfires on Ozone Levels
This section attempts to characterize the contributions of wildfires on monthly 8-hr max O 3 concentrations.First, monthly 8-hr max O 3 concentrations for periods with different fire frequencies were compared to qualitatively assess the effect of wildfires.Secondly, a multivariate regression model was applied to estimate the contributions of wildfires on O 3 concentrations.The analysis was done for the 2001- The site codes refer to monitoring sites labels in Fig. 1. 2010 period.For sites with O 3 measurements less than 10 years during the aforementioned period (Table 1), analysis is limited to the period in which O 3 data were available.
Initially, each calendar month was binned into one of four categories based on the number of fires detections as compared to the total number of fires in each buffer zone.The 25-th, 50-th and 75-th percentiles were applied as cutpoints for months with: (i) low number of fires detections (less than the 25-th percentile); (ii) moderate number of fires detections (between 25-th and 50-th percentiles); (iii) high number of fires detections (between 50-th and 75-th percentiles) and; (iv) extreme number of fires detections (higher than the 75-th percentiles).Fig. 2 illustrates the monthly 8-hr max O 3 concentrations for months with low, moderate, high and extreme number of fire detections in 0-400 km (a) and more than 400 km (b) for Chiricahua (ch), and urban/rural sites.In Chiricahua, the 8-hr max O 3 concentration increased as the fire intensity increased ).The same trend was also observed for urban and rural sites in the study domain for fires more than 400 km (Fig. 2(b)).The increase may be attributed to mixing of wildfire plumes with locally-emitted NO x (Singh et al., 2010) and/or the decomposition of peroxyacetylnitrates (PAN) that results in the production of O 3 in excess far downwind from the fires.PANs are formed from the photolysis of wildfire-emitted NO within a few hours (Alvarado et al., 2010) and are present in plumes for up to 15 days (Val Martin et al., 2006).Conversely, for fires within 400 km (Fig. 2(a)), the 8-hr max O 3 concentrations measured for months with low fire frequency in urban/rural sites was higher (~3-4 ppbv) than the 8-hr max O 3 concentrations measured for months with moderate and high fire frequencies (Fig. 3(a)) due to NO titration of O 3 .
The average levels of NO, iso-pentane, isoprene, benzene and toluene measured at ep3 on August 2008 (month with low fire frequency within 400 km) were 3.3 ppbv (maximum: 61.1 ppbv), 5.6 ppbC (maximum: 44.0 ppbC), 0.1 ppbC (maximum: 1.7 ppbC), 0.9 ppbC (maximum: 8.3 ppbC) and 4.7 ppbC (maximum: 39.7 ppbC), respectively.They were lower than those measured on months with high fire frequency during the same period (July and September) (NO: 5.5 ppbv (maximum: 55.0 ppbv); iso-pentane 6.2 ppbC (maximum: 64.3 ppbC); isoprene: 0.2 ppbC (maximum: 3.1 ppbC); benzene: 1.4 ppbC (maximum: 14.6 ppbC) and; toluene: 7.8 ppbC (maximum: 89.0 ppbC)).The reduction of NO 2 and O 3 photolysis rates due to smoke plumes is also possible; however, this may lead to both reductions and enhancements in O 3 production (Real et al., 2007).Table 4 shows the contributions of fires on monthly 8-hr max O 3 concentrations in each buffer zone.For all the sites (with the exception of Chiricahua), monthly 8 hr-max O 3 concentrations decreased by as much as 8 ppbv for fires within 400 km of the sites (estimates were statistically significant at a = 85%).A net positive contribution of 2 ppbv was estimated for the background site in Chiricahua for fires within 400 km.For fires in the 400-800 km buffer zone and more than 1600 km, positive contributions ranged from 1 up to 13 ppbv (statistically significant at a = 85%) in all sites.However, statistically insignificant contributions were computed for fires in the 800-1600 km buffer zone.Previous studies clearly demonstrated that fire plumes in the Canadian Rockies, Alaska and Siberia (from 2000-5000 km) influence O 3 concentrations in the US and Europe (Forster et al., 2001;Colarco et al., 2004;Hudman et al., 2004;Bertchi and Jaffe, 2005;Oltmans et al., 2010) and attributed to O 3 production due to long-range transport of PANs (Fischer et al., 2010).The influence of PANs on O 3 Fig. 3. Monthly variation of MODIS fires detections in the 400-800 and 1600-4800 km buffer zones (a) and spatial variation of MODIS fire detections within the 400-800 km (b) and 1600-4800 km (c) buffer zones for sites in Sunland Park, El Paso and Cuidad Juarez region.
-5 -8 12 0 3 4 a Significant estimates (at p-value < 0.15) are in bold.may be variable in different locations because of the dependence of PAN decomposition to ambient temperature and transport time.Overall, wildfires contributed from 1 to 13 ppbv on the monthly 8-hr max O 3 concentration.The estimated contributions represented from 9 to 22 percent of measured monthly 8-hr max concentrations, which were comparable to those computed by Pfister et al. (2006).
The monthly variation of fire detections demonstrated a prevalence of fires in spring for the 400-800 km buffer zone, and a bimodal distribution in early spring (March-April) and summer (July-August) for the 1600-4800 km buffer zone (Fig. 3(a)).The spatial variations of normalized fires detection (number of fires detection in the cell /total number of fires detection in the buffer zone) for 0.25° × 0.25° cells in the 400-800 km (Fig. 3(b)) and 1600-4800 km (Fig. 3(c)) buffer zones for sites in Las Cruces-El Paso-Ciudad Juarez region are also depicted.Note that we combined the data for all sites; thus, the buffer zones have a rather irregular (mostly oval) shape.Three areas in the 400-800 km buffer zone, central Arizona, central Texas and south Sonora/north Sinaloa in Mexico, experienced the highest numbers of fire detections (Fig. 3(a)).For the combined 1600-4800 km buffer zones, the highest number of fire detections were observed along the Colorado Rockies in Idaho and British Columbia, in southeast US (Alabama, Georgia and Florida), central Mexico, Cuba and Jamaica.As part of the assessment of the causes of haze in Class I protected areas in the western regional air partnership domain, the analysis of backward trajectories residence times for Class I sites (Chiricahua, Salt Creek wilderness area 150 km north of Las Cruces, New Mexico, Carlsbad Caverns and Guadelupe Mountain wilderness area 120 km east of El Paso, Texas) showed that air masses in the period of April-June originated from northern and western Mexico, while for summer months, air masses came from the Gulf of Mexico and Cuba (DRI, 2006).Overall, the observed patterns of fires and air masses origin indicated that regional fire events in northwestern Mexico in spring, and southeast US and Caribbean islands in the summer may contribute to elevated O 3 concentrations in southern New Mexico and El Paso regions.
Weather conditions play an important role on both O 3 formation and the increased number of wildfires at different time scales.O 3 formation is promoted in calm, cloudless conditions by reduced horizontal and vertical mixing and increased direct solar radiation.On the other hand, longterm trends demonstrating short snow cover periods in the winter and long lasting warmer temperatures during summer in forest areas promote the accumulation of biomass fuel and increase the fire risk.Wildfires induced by humans or lightning typically initiate in windy and stormy conditions.

Limitations of Methodology
This approach relied on the ability to identify a statistical association between monthly 8-hr max O 3 concentrations and fire detections in predefined buffer zones.There were several assumptions that applied to the outcomes of this analysis.First, the regression model in Eq. (2) assumed that there is a linear relationship between the numbers of fire detects and O 3 levels and those regression coefficients are invariant with time.Emissions of O 3 precursors (VOCs and NO x ) from wildfires are highly variable and depend on the fuel and fire conditions with average molar VOCs/NO x ratios from 15 for savannahs to 150 for boreal forests (Akagi et al., 2011).Furthermore, the chemistry of VOCs/ NO x irradiated mixtures is non-linear and challenging to describe due to large number of VOCs and high emissions uncertainties (Roy et al., 2006).However, the variability in net O 3 production caused by fire emissions and photochemistry is reduced as plumes travel away from the fire due to transformations of NO x to PAN and VOCs to alkyl (R), alkoxy (RO) and alkylperoxy radicals (RO 2 ) within a few hours and vertical/horizontal mixing.The net O 3 production within the plume is, then, related to the decomposition of PAN over time.The outcome of using the number of fire detections as a surrogate for fire emissions would be to weaken the confidence in the coefficients in Eq. ( 2) because periods of high fire detection frequency and high ozone (e.g., wildfires in summer) would be counteracted by periods of high fire detection frequency (e.g., prescribed fires in winter and early spring) and low/moderate ozone.It is worth noting that recent efforts to obtain more precise estimates on fire emissions rely heavily on the satellite observations of fires locations (Pouliot et al., 2008;Zhang et al., 2008).
Second, we assume that the direction and characteristics of smoke plumes for fires within the same buffer zone did not influence the overall contribution of the buffer zone on ozone concentrations.Plumes near the fire are aligned with the prevailing wind conditions and thus, the use of circular buffer zones may overestimate the cases in which wildfire smoke influences the receptor site because for some cases smoke would move, at least initially, away from the receptor site.However, during transport, smoke plumes are mixed and follow the clockwise/counterclockwise circulation of synoptic scale weather systems.This pattern was demonstrated by the air mass trajectories residence time (DRI, 2006) and in previous studies showing that fires plumes in California initially spread over Pacific Ocean and covered areas in the Pacific Northwest and central Canada (Forster et al., 2001;Collarco et al., 2004;Bertchi and Jaffe, 2005;Oltmans et al., 2010).The error introduced by this assumption would be higher for the outer buffer zones resulting in overall statistically insignificant results.
By setting the significance criterion at the α = 85% level, the outcomes of this analysis represent a conservative but robust estimate of wildfire contributions of relatively large buffer zones on O 3 levels.If a larger number of buffer zones was selected (increase the number of independent variables in the regression model), it would have a detrimental impact on extracting statistically significant results.As the availability of fire detection data is increasing, these limitations may be addressed by running the regression model using a larger number of independent variables.

CONCLUSIONS
Hourly O 3 concentrations for 23 sites in southeastern Arizona, south New Mexico, El Paso and Ciudad Juarez for the 1993-2010 period were retrieved from the EPA Air Quality System.The monthly 8-hr max concentrations were calculated to evaluate the annual trends and the impact of wildfires on O 3 pollution.For monitoring sites located in urban and rural populated areas, the monthly 8-hr max O 3 concentrations have been declining from 0.15 up to 1.22 ppbv/yr due to the reduction in emissions of O 3 precursors in urban communities.On the contrary, O 3 concentrations in Chiricahua demonstrated a statistically significant increasing trend.The moderate correlations between urban sites and Chiricahua site indicated that there were substantial differences between urban and background sites.These discrepancies may be partially attributed to the increasing frequencies and intensities of wildfires.
We identified a quantitative relationship between monthly 8-hr max O 3 concentrations and the number of fire detections by satellites in six buffer zones (< 160, 160-400, 400-800, 800-1600, 1600-3200 and 3200-4800 km).Fires within 400 km triggered a reduction of monthly 8-hr max O 3 concentrations as high as 12 ppbv in urban sites.This was attributed to NO titration of O 3 and the reduction of NO 2 and O 3 photolysis rates.Strong positive contributions (up to 12 ppbv) were computed for fires incidents at more than 800 km.This was attributed to the formation of aloft PAN and O 3 in smoke plume during transport and its mixing with ground-level air.Fires in western Mexico, southeast US and Caribbean islands may be responsible for a fraction of the observed contributions.The use of fire metrics (i.e., number of fire detections) addresses a major limitation of traditional approaches that use fire emissions because of the large uncertainties of emissions factors, characteristics of the fuel composition and incomplete understanding and/or simulation of physical and chemical aging during transport to the receptor sites, and the contributions of other sources.

Fig. 1 .
Fig. 1.Map showing the boundaries of the study domain (a), the locations of the O 3 monitoring sites in Arizona and western New Mexico (b) and El Paso, Las Cruces and Cuidad Juarez (c).

Fig. 2 .
Fig. 2. Monthly 8-hr max O 3 concentration for months with low, moderate, high and extreme number of fire detection in 0-400 km (a) and > 400 km zones (b).

Table 1 .
Site characteristics and O 3 monitoring periods.
aThe site codes refer to monitoring sites labels in Fig.1.b Aerometric Information Retrieval System.n.a.not available.

Table 2 .
The COD ratio, absolute (ΔC) and relative (%ΔC/Ref) concentration differences (compared to the "ch" site) and annual trends of deseasonalized O 3 concentrations.

Table 3 .
Annual trends of monthly 8-hour max O 3 concentrations for individual months.

Table 4 .
Buffer contributions of wildfires on monthly 8-hr max O 3 concentrations.