Background ozone over the United States in summer: Origin, trend, and contribution to pollution episodes

[ 1 ] Observations indicate that ozone (O 3 ) concentrations in surface air over the United States in summer contain a 20–45 ppbv background contribution, presumably reflecting transport from outside the North American boundary layer. We use a three-dimensional global model of tropospheric chemistry driven by assimilated meteorological observations to investigate the origin of this background and to quantify its contribution to total surface O 3 on both average and highly polluted summer days. The model simulation is evaluated with a suite of surface and aircraft observations over the United States from the summer of 1995. The model reproduces the principal features in the observed distributions of O 3 and its precursors, including frequency distributions of O 3 concentrations and the development of regional high-O 3 episodes in the eastern United States. Comparison of simulations with 1995 versus 1980 global fossil fuel emissions indicates that the model captures the previously observed decrease in the high end of the O 3 probability distribution in surface air over the United States (reflecting reduction of domestic hydrocarbon emissions) and the increase in the low end (reflecting, at least in the model, rising Asian emissions). In the model, background O 3 produced outside of the North American boundary layer contributes an average 25–35 ppbv to afternoon O 3 concentrations in surface air in the western United States. and 15–30 ppbv in the eastern United States during the summer of 1995. This background generally decays to below 15 ppbv during the stagnation conditions conducive to exceedances of the 8-hour 0.08 ppmv (80 ppbv) National Ambient Air Quality Standard (NAAQS) for O 3 . A high background contribution of 25–40 ppbv is found during 9% of these exceedances, reflecting convective mixing of free tropospheric O 3 from aloft, followed by rapid production within the U.S. boundary layer. Anthropogenic emissions in Asia and Europe are found to increase afternoon O 3 concentrations in surface air over the United States by typically 4–7 ppbv, under both average and highly polluted conditions. This enhancement is particularly large (up to 14 ppbv) for O 3 concentrations in the 50–70 ppbv range, and would represent a major concern if the NAAQS were to be tightened.


Introduction
[2] Many regions of the United States are plagued by chronic high summertime ozone (O 3 ) levels produced from rapid photochemical oxidation of carbon monoxide (CO) and hydrocarbons in the presence of nitrogen oxides (NO x = NO + NO 2 ) [National Research Council, 1991]. In an effort to better protect public health and vegetation from the adverse effects of O 3 pollution, the Environmental Protection Agency (EPA) revised its standard for O 3 in 1997 to a 0.08 ppmv (80 ppbv) annual fourth-highest daily maximum 8-hour concentration averaged over three years [Environmental Protection Agency, 2000]. Implementation of this new standard will pose a substantial challenge to the design of effective air pollution control strategies [Lefohn et al., 1998]. Saylor et al. [1998] found that 30-50% of rural sites measuring O 3 in the eastern United States exceeded the 80 ppbv standard during 1993 -1995, while only 2 -12% did not attain compliance with the prior 120 ppbv 1-hour standard. According to the Environmental Protection Agency [2001], 123 million Americans lived in regions failing to achieve compliance with the new standard in 1999.
[3] A substantial fraction of O 3 in surface air over the United States in summer, about 20-45 ppbv, appears to represent a background that either subsides or is horizontally advected into the U.S. boundary layer. The background has been estimated from various techniques, including correlations with reactive nitrogen oxides (NO y = NO x + HNO 3 + HNO 4 + PAN + PMN + PPN + alkyl nitrates + 2N 2 O 5 ), measurements at remote surface sites, and analyses of probability distributions of O 3 concentrations, all of which yield similar values [Trainer et al., 1993;Hirsch et al., 1996;Lin et al., 2000]. Some authors have noted a much larger range of hourly average concentrations at sites considered to measure background levels of O 3 [Altshuller and Lefohn, 1996;Lefohn et al., 1998Lefohn et al., , 2001. A component of the O 3 background is produced from natural precursor sources, which include NO x emissions from soil and lightning, and hydrocarbon emissions from vegetation. Nineteenth century observations in Europe suggest that the natural O 3 background is only 5-10 ppbv [e.g., Anfossi et al., 1991;Marenco et al., 1994]. Although these measurements are highly uncertain, they support the contention that present-day background O 3 in Europe as well as in the United States includes an anthropogenic enhancement [Beck and Grennfelt, 1994;Marenco et al., 1994;Kasibhatla et al., 1996].
[4] There is limited evidence that background concentrations have risen by a few ppbv in surface air over the United States during the past 2 decades, from observations at Whiteface Mountain, New York [Oltmans et al., 1998], and from an analysis of the probability distribution of daily maximum 8-hour average concentrations at rural sites in the EPA AIRS network [Lin et al., 2000]. Such a rise in the background may thwart efforts to meet the U.S. air quality standard via domestic emission controls, considering the less-than-linear response of O 3 to reductions of anthropogenic NO x and NMHC emissions Jacob et al., 1999;Sillman, 1999;Ryerson et al., 2001]. In the coming decades, rising emissions from fossil fuel combustion in developing countries have the potential to further increase this background. As the gap between background O 3 levels and the national air quality standard for O 3 narrows, it becomes increasingly important to understand and quantify the anthropogenic enhancement to the background, since it has the potential to be reduced via international agreements on emission controls [NARSTO Synthesis Team, 2000].
[5] In this work, we investigate the origin of the presentday background O 3 in surface air over the continental United States with a three-dimensional global tropospheric chemistry model (GEOS-CHEM) driven by assimilated meteorological data from the Goddard Earth Observing System (GEOS). We focus our analysis on the summer of 1995 because of the wealth of observations of O 3 and related species available from two field campaigns: the North American Research Strategy for Tropospheric Ozone-Northeast (NARSTO-NE) Mueller et al., 1996] and the Southern Oxidant Study (SOS) [Cowling et al., , 2000Hubler et al., 1998;McNider et al., 1998;Meagher et al., 1998]. These campaigns focused on improving the current understanding of regional O 3 pollution over the eastern United States. They allow a regional evaluation of O 3 chemistry in the GEOS-CHEM model, complementing the global evaluations presented elsewhere [Bey et al., 2001a[Bey et al., , 2001bLi et al., 2001;Liu et al., 2001Liu et al., , 2002Martin et al., 2002;, Global simulation of tropospheric ozone chemistry over the North Atlantic, 1, Model evaluation and ozone budget, manuscript in preparation, 2002 (hereinafter referred to as Li et al., manuscript in preparation, 2002)]. Another important data set for our work is the ensemble of continuous O 3 observations at 950 surface sites in the United States, compiled by the EPA in the Aerometric Information Retrieval System (AIRS) database. Five major regional episodes of elevated O 3 in the eastern United States in 1995 enable us to examine the background contribution to total surface O 3 over a wide range of conditions.
[6] We provide a summary in section 2 of the meteorological and chemical environment for the summer of 1995. Section 3 presents a brief description of the GEOS-CHEM model as applied to our study. The ability of the model to reproduce the observations from AIRS, SOS, NARSTO-NE, and other programs (Table 1) is examined in section 4. In sections 5 and 6, we apply the model to investigate the origin of the background O 3 in surface air over the United States via (1) regional source tagging and (2) sensitivity simulations. Conclusions are given in section 7.

Meteorological and Chemical Environment for the Summer of 1995
[7] Winds and mixing depths are the critical meteorological variables controlling the dispersion of pollutants and the regional buildup of O 3 . The summertime circulation over the United States is governed by the Bermuda and Pacific high pressure systems, as illustrated in Figure 1a with a map of average July 1995 surface winds. Anthropogenic emissions of O 3 precursors in the United States roughly follow population density and are largest in the eastern part of the country and in California. Winds in the southeastern United States are generally weak, promoting the accumulation of pollution. The northeastern United States is usually better ventilated, but weak migratory high-pressure systems episodically traverse the region during summer; the associated high surface temperatures and low wind speeds are conducive to O 3 formation and accumulation [Logan, 1989;NRC, 1991;Vukovich, 1995]. The largest O 3 episode of 1995 occurred in mid-July under these characteristic stagnation conditions. In addition to this major episode, four smaller episodes occurred in the Northeast [Mueller et al., 1996], and one in the Southeast [McNider et al., 1998].
[8] The depth of the mixed layer plays a key role in governing chemical concentrations in surface air. Figure 1b displays the average summer afternoon (1300 -1700 local time (LT)) mixing depths from the 3-hour average GEOS data. These mixing depths are diagnosed by the GEOS-1 Data Assimilation System as the pressure level where turbulent kinetic energy is 10% of the surface layer value [Allen et al., 1996]. Typical values are 1-2 km in the eastern United States and 2-4 km in the west (Figure 1b), consistent with the general climatology presented by Holzworth [1967], and also with observations from the summer 1995 NARSTO-NE and SOS field campaigns [Blumenthal et al., 1997;Banta et al., 1998;McNider et al., 1998;Berman et al., 1999;Seaman and Michelson, 2000]. The latter observations exhibit substantial spatial variations in mixing depth on scales smaller than the 2°Â 2.5°GEOS resolution, particularly at coastal locations [Berman et al., 1999] where steep gradients in surface heating exist. The inability of the model to resolve this gradient yields a poor simulation of chemical concentrations at coastal stations, as discussed in section 4.1.
[9] Table 1 lists the observational data sets used to evaluate the GEOS-CHEM simulation over the United States. The AIRS network provides a comprehensive set of data for evaluating the surface O 3 simulation, but aside from California, stations are sparse over the western United States. Figure 2 shows that mean afternoon (1300 -1700 LT) O 3 concentrations measured by the AIRS sites during the summer of 1995 are greater than 60 ppbv over large areas of the United States, and exceed 80 ppbv in the Los Angeles area.
[10] The summer 1995 NARSTO-NE and SOS field campaigns provide both aircraft and surface measurements for a variety of species relevant to O 3 production. During the summer of 1995, meteorological conditions conducive to rapid photochemical O 3 production occurred more frequently than would be expected from climatological averages [McNider et al., 1998]. Much work has been done to analyze and model observations from these campaigns [e.g., Southern Oxidants Study Nashville/Middle Tennessee Ozone Study compendium; special section of the Journal of Geophysical Research [April 16, 2000 pp. 9075 -9211]); Anderson et al., 1996;Mueller et al., 1996;Zhang et al., 1998;Ryan et al., 1998;Zhang and Rao, 1999]. In particular, these studies have examined the chemical and dynamical processes contributing to episodes of regionally elevated O 3 [e.g., Kasibhatla et al., 1998;McNider et al., 1998;Ryan et al., 1998;Zhang et al., 1998;Seaman and Michelson, 2000]. The largest exceedances of the national O 3 standard in the eastern United States occurred in mid-July during a classic stagnation event (clear skies, high temperatures, and reduced wind speeds) that ended when a cold front moved through the region [Mueller et al., 1996;McNider et al., 1998;Olszyna et al., 1998;Seaman and Michelson, 2000].
[11] An unexpected finding based on the analysis of summer 1995 observations was that Canadian fires in late June and early July produced episodic enhancements of over 100 ppbv to CO levels in the southeastern United States during northerly flow following cold fronts . McKeen et al. [2000McKeen et al. [ , 2001] used a regional 3-D photochemical transport model to show that surface O 3 was enhanced by 10 -30 ppbv over the eastern United States during the 29 June to 2 July time period when the Canadian fires had their largest impact. They find CO oxidation to be an important source for this O 3 enhancement, particularly over regions with high NO x emissions where O 3 chemistry is hydrocarbon-limited. Emissions from Canadian forest fires are found to elevate background levels of O 3 over the eastern United States on days when northerly flow would otherwise bring clean air from Canada, but these emissions have little influence on pollution episodes which occur under stagnant meteorological conditions [McKeen et al., 2001].

Model Description
[12] The GEOS-CHEM model is a global three-dimensional tropospheric chemistry model driven by assimilated meteorological observations from the Goddard Earth Observing System (GEOS) of the NASA Data Assimilation Office. A detailed description of the model as applied to a global ozone-NO x -hydrocarbon simulation is given by Bey et al. [2001a], and we provide a summary below. The model has previously been applied to a number of tropospheric chemistry problems including ozone over the Pacific [Bey et al., 2001b] and in the Middle East , aerosol tracers [Liu et al., 2001], acetone , HCN [Li et al., 2000], and CH 3 I [Bell et al., 2002]. Vertical profiles from GEOS-CHEM have been used to constrain retrievals of CH 2 O and NO 2 column concentrations from the Global Ozone Monitoring Experiment (GOME) satellite instrument, and the observed columns have been used in turn to evaluate the model [Palmer et al., 2001;Martin et al., 2002, P.I. Palmer et al., Mapping isoprene emissions over North America using formaldehyde column observations from space, submitted to Journal of Geophysical Research, 2002 (hereinafter referred to as Palmer et al., submitted manuscript, 2002)].
[13] Several improvements targeted at the present study have been made to the original Bey et al.  mechanism, the radiative transfer code, and biogenic emissions.

General Description of GEOS-CHEM
[14] The 1995 GEOS meteorological data used as input to GEOS-CHEM [Bey et al., 2001a] include 3-D fields updated every 6 hours for winds, temperatures, specific humidities, wet convective mass fluxes and detrainment rates, water condensation rates, and cloud optical depths; 2-D fields updated every 6 hours for surface pressures and surface albedos; and 2-D fields updated every 3 hours for column cloud fractions, roughness heights, friction velocities, surface winds at 10 m altitude, surface temperatures, solar radiation flux at the surface, sensible heat flux, total precipitation at the ground, convective precipitation at the ground, and mixing depths. These data serve as input to different GEOS-CHEM modules that simulate tranport, emissions, photochemistry, and deposition (see Table 1 of Bey et al. [2001a] for more information). Rasch et al. [1997] have previously shown that the temporal resolution used for the meteorological fields is adequate for the scales of interest in our simulation. The GEOS data are provided on a 2°latitude by 2.5°longitude (2°Â 2.5°) horizontal grid with 20 vertical sigma layers, including five located below 2 km altitude (for a column based at sea level, these levels are centered at 50, 250, 600, 1100, and 1750 m). We use this resolution in our standard simulation. A faster version of the model [Bey et al., 2001a[Bey et al., , 2001b] uses a 4°Â 5°resolution, with GEOS fields regridded accordingly. We conduct our spin-up and sensitivity simulations at this coarser resolution, which is compared with the finer resolved version in section 4. Initial conditions for January 1995 are taken from Bey et al. [2001aBey et al. [ , 2001b, who had previously run the model from July 1993 through December 1994. We continue to spin up the model until June 1995, and present results from June through August 1995.
[15] Transport of O 3 from the stratosphere is simulated with the Synoz (synthetic ozone) method of McLinden et al.
[2000] to yield a cross-tropopause flux of 475 Tg O 3 yr À1 . Dry deposition velocities are calculated with a resistance-inseries model [Wesely, 1989]. The wet deposition scheme is that developed by Liu et al. [2001] and is applied here to the soluble gases HNO 3 and H 2 O 2 .

Emissions 3.3.1. Anthropogenic
[17] Global anthropogenic emissions based upon 1985 inventories of NO x [Benkovitz et al., 1996], nonmethane hydrocarbons [Piccot et al., 1992] and CO [Wang et al., 1998a] are scaled to 1995 levels as described by Bey et al. [2001a]. A mean diel cycle [Jacob et al., 1993a] is applied to these emissions. For the western states (west of 98.75°W), the NAPAP inventory for 1985 (described by Wang et al. [1998a]) is scaled to 1995 levels according to EPA [1997]. For the Ozone Transport Assessment Group (OTAG) region in the eastern United States (68.75°W-98.75°W and 25°N -49°N), we have implemented the Southern Appalachian Mountain Initiative (SAMI) emissions inventory for NO x , CO and hydrocarbons on 11 July 1995 [E. H. Pechan and Associates, 1999;Ogburn et al., 2000], and extrapolated these emissions to the rest of the summer. Table 2 gives daily emission totals for the eastern and western United States and for the northern midlatitudes.

Natural
[19] Natural sources of NO x in GEOS-CHEM include lightning (3 Tg N yr À1 ) and soil (7 Tg N yr À1 ) [Bey et al., 2001a]. Isoprene is the dominant biogenic hydrocarbon contributing to O 3 production in the boundary layer in the United States [Trainer et al., 1987;Goldan et al., 2000]. Global emission of isoprene in the model is 400 Tg C yr À1 , including 30 Tg C yr À1 in the United States [Bey et al., 2001a]. Isoprene is oxidized according to the mechanism of Horowitz et al. [1998], with updates by Bey et al. [2001a]. We have also added a global biogenic source of 8.4 Tg yr À1 CH 2 O from oxidation of monoterpenes in order to better match CH 2 O observations over the United States [Palmer et al., 2001]. This source is based on the monoterpene global emissions inventory of Guenther et al. [1995] (120 Tg C yr À1 ) and a 0.28 CH 2 O molar yield from Orlando et al.

Radiative Transfer
[20] Photolysis rate constants in GEOS-CHEM are calculated with the Fast-J radiative transfer code , which accounts for Mie scattering by clouds and aerosols. Radiative variables include altitude-dependent cloud optical depths from the GEOS archive [Bey et al., 2001a], ultraviolet surface albedos from the TOMS satellite [Herman and Celarier, 1997], climatological O 3 columns from Logan et al. [1999], and aerosol scattering as described below.
[21] Recent work by Kondragunta [1997], Dickerson et al. [1997] and Park et al. [2002] suggests that O 3 concentrations in the U.S. boundary layer during pollution episodes are enhanced by aerosol radiative scattering that increases the NO 2 photolysis frequency (JNO 2 ). The effect on photolysis rates and O 3 will vary with the scattering and absorption properties of the aerosols. Over the eastern United States, aerosol optical depths (AOD) at 380 nm are typically less than 0.6 in summer [Russell et al., 1999], but may exceed two during pollution episodes Park et al., 2002]. We compute the local AOD in GEOS-CHEM by using a linear relationship between AOD at 380 nm and surface O 3 (ppbv) observed at a nonurban Maryland site by Kondragunta [1997] in the summer of 1995: AOD = max(0, ([O 3 ] À 62)/30). While this parameterization is crude in that it neglects any explicit dependence of AOD on relative humidity, its impact on our simulation is small, as discussed further below. We include the AOD in the radiative transfer code as a submicron aerosol distributed uniformly over the local mixing depth with a gamma size distribution, a modal diameter of 0.1 mm, and a single scattering albedo of 0.96; these parameters are representative of aerosols over the eastern United States [Dickerson et al., 1997]. The corresponding power law dependence of AOD on wavelength has an inverse exponent of 1.7, consistent with the measurements of Kondragunta [1997]. For an extreme pollution event with an AOD of 2.0 distributed in this manner under a clear sky, we find that JNO 2 decreases by up to 20% at the surface and increases by 10-30% in the upper part of the mixed layer, consistent with the findings of Kondragunta [1997], Dickerson et al. [1997], and Park et al. [2002].
[22] In our standard simulation, inclusion of scattering aerosols in the mixed layer decreases monthly mean JNO 2 by up to 15% at the surface, and increases it by up to 10% at higher altitudes. The effect on monthly mean O 3 is less than 0.2 ppbv anywhere. Given this small influence on our O 3 simulation, we did not pursue a more complicated representation of AOD in the model. We conducted a sensitivity simulation with an artificially high AOD of 2 to determine the maximal radiative effect of aerosols on O 3 ; even then, O 3 in the boundary layer increased by less than 2 ppbv over the northeastern United States. Although the addition of scattering aerosols in the GEOS-CHEM mixed layer had little effect on the surface O 3 simulation, aerosols may have a larger impact on O 3 levels in urban airsheds which are not resolved by the model.

Model Evaluation
[23] Detailed evaluation of model results with observations, not only for O 3 , but also for its precursors and related species, is critical for assessing the ability of the model to describe the processes controlling O 3 [Sillman, 1999]. In this work, we focus our evaluation on the United States in the summer of 1995 to diagnose strengths and weaknesses in the model through comparison with the observations in Table 1. The vertical resolution of the model is too coarse to simulate properly the diurnal variation of surface air concentrations driven by mixed layer growth and decay. We focus our analysis on the afternoon hours (1300-1700 LT) when O 3 concentrations are usually highest and when surface observations are representative of a mixed layer sufficiently deep (>100 m) to be resolved by the model.
[24] We begin with a summary of previous evaluations of GEOS-CHEM at northern midlatitudes. Measurements from surface sites, NASA/GTE aircraft campaigns and ozonesondes have been employed by Bey et al. [2001a] to compare GEOS-CHEM simulations with seasonal statistics of O 3 and other chemical species affecting the O 3 budget (NO x , PAN, HNO 3 , CO, hydrocarbons, peroxides, and carbonyls). Simulated monthly mean ozone concentrations are typically within 10 ppbv of measured values throughout the troposphere with no systematic bias. The O 3 seasonal cycle as determined from ozonesonde observations at northern midlatitudes is reproduced, with a slight tendency to underestimate the full amplitude of the observed seasonal variation in the middle and upper troposphere, which may indicate that seasonal variation in cross-tropopause transport is too weak in the model. The model generally captures the observed vertical gradients of O 3 at northern mid and high latitudes. Bey et al. [2001a] do not find any systematic bias in simulated NO or PAN concentrations in the northern midlatitudes; monthly mean concentrations for these species are generally within a factor of 2 of aircraft observations. As is typical of global models [e.g., Hauglustaine et al., 1998;Wang et al., 1998b;Lawrence et al., 1999;Mickley et al., 1999], HNO 3 is generally overestimated by a factor of 2 -3, possibly due to a combination of inaccurate representation of precipitation scavenging, especially in the upper troposphere, and the absence of any partitioning of HNO 3 into the aerosol phase. The model underestimates CO concentrations at northern midlatitudes by an average of 10-20 ppbv. In a more recent version of the model [Duncan and Logan, 2001;B.N. Duncan et al., Interannual and seasonal variability of biomass burning emissions constrained by  satellite observations, submitted to Journal of Geophysical Research, 2002 (hereinafter referred to as Duncan et al., submitted manuscript, 2002)], this flaw has been corrected with an additional source of CO from biogenic methanol and updated estimates of nonmethane hydrocarbon emissions from biofuels, fossil fuels, and biomass burning; the present simulation does not include this update. In addition, Palmer et al. [2001Palmer et al. [ , submitted manuscript, 2002 have recently evaluated the GEOS-CHEM simulation over the United States with observations of CH 2 O from surface sites and find no systematic bias.  Figure 2 shows the difference (model bias). Table 3 shows that the spatial variance in the mean O 3 field decreases by 20% when the AIRS observations are averaged over the 2°Â 2.5°model grid, although the mean concentrations are similar. The maximum concentrations are over Los Angeles, where there is a sharp transition from low O 3 concentrations at clean coastal monitoring stations to highly polluted sites in the Los Angeles basin (upper panel of Figure 2). Averaging over the larger model grid smooths out the gradient in O 3 concentrations, yielding the 30 ppbv decrease in the maximum concentration (Table 3).

Surface O 3 Over the United States
[26] The model generally reproduces the mean O 3 concentrations, and about half of the variance in the spatial distribution (r 2 = 0.44), consistent with simulations from other current global models of atmospheric chemistry and transport [e.g. Jacob et al., 1993a;Kasibhatla et al., 1996;Horowitz et al., 1998;Lawrence et al., 1999]. An overestimate of O 3 concentrations in the southeastern United States is responsible for much of the 5 ppbv bias (Table 3), which decreases to 3 ppbv (r 2 = 0.52) when the southeastern quadrant is removed from the analysis. We show below that the model adequately simulates the photochemical relationship governing production of O 3 from its precursor emissions as evidenced by comparison with observed distributions of precursors, O 3 versus NO y -NO x correlations, and the response of O 3 to changes in fossil fuel emissions between 1980 and 1995. In a companion study, Li et al. (manuscript in preparation, 2002) find that GEOS-CHEM reproduces observed relationships between O 3 and its precursors over eastern Canada and the western North Atlantic.
[27] The largest discrepancies between the standard simulation and the observations occur for urban coastal grid boxes where the model values are too high, due to steep subgrid land-to-sea gradients in mixing depth that are not resolved in the model. Ventilation of coastal emissions in the model is restricted by shallow mixing depths imposed by the cold ocean surface. The simulation over California is further compromised by the inability of the model to resolve the complex topography controlling airflow and ventilation in that region.
[28] Compared to our standard 2°Â 2.5°resolution model, the 4°Â 5°resolution model which is applied to our sensitivity simulations in section 6, shows a smaller bias with the observations (+3 ppbv) and a higher coefficient of determination (r 2 = 0.70), but smooths out the highest concentrations (Table 3). Liang and Jacobson [2000] previously showed that degrading the resolution of a 3-D model can either increase or decrease O 3 depending on local emissions and the photochemical regime. The apparent improvements in the bias and r 2 in our 4°Â 5°model relative to the 2°Â 2.5°version stem from smoothing out the data, and do not imply that the 4°Â 5°resolution model offers a better simulation.
[29] Figure 3 shows the frequency distribution of summer afternoon O 3 concentrations for three representative 2°Â 2.5°grid squares in the United States. Observed frequency distributions were constructed using (1) spatial averages of data for the ensemble of the AIRS sites in the grid square and (2) one selected non-AIRS rural site. The model reproduces the frequency distribution observed over the Grand Canyon. It also simulates the high tail of the distribution in the eastern United States, including events with concentrations in excess of 80 ppbv. The lower tail of the distribution is substantially overestimated in the eastern United States, particularly in the southeastern United States (Figure 3). Low O 3 concentrations observed in the southeastern United States are associated with southerly flow from the Gulf of Mexico and the Caribbean, but in the model this southerly flow carries higher O 3 concentrations due to excessive convection over the Gulf of Mexico and the Caribbean, a weakness in the GEOS-1 data previously identified by Allen et al. [1997]. Li et al. [2002] have shown that this problem is also responsible for a model overestimate of O 3 concentrations in surface air at Bermuda in July.
[30] The overestimate of the low tail of the frequency distribution in the northeastern United States is less severe than in the southeastern United States and stems from different factors. Lin et al. [2000] have previously shown that the background O 3 (diagnosed by correlation with low CO and NO y measurements) at Harvard Forest corresponds to the 25th percentile of the probability distribution in summer. Points below this background level reflect depletion of O 3 from deposition or titration by NO. Since the AIRS sites are predominantly clustered in urban areas, the low values could be indicative of titration by NO in urban plumes that the model does not resolve. The low tail in the GEOS-CHEM model compares well with the observations at Harvard Forest (squares in the top panel of Figure 3), with the exception of the two lowest points; the lowest occurred during tropical storm activity, while the second lowest was associated with northerly flow from Canada.

Surface Concentrations of Related Species
[31] Surface observations in 1995 for species other than O 3 are limited. Hourly average concentrations of O 3 , NO, NO y , and CO were recorded at seven surface sites as part of SOS (Table 1) Olszyna et al., 1998]. We sampled the model grid square centered at 36°N and 87.5°W over Tennessee and compared its O 3 , NO, NO y , and CO concentrations with those recorded at two suburban (Youth Incorporated and Dickson County in Tennessee) and two rural (Giles County, Tennessee, and Land Between the Lakes, Kentucky) sites within the grid square. Figure 4 shows the time series of median O 3 , NO, NO y , and CO observations for the four sites (solid lines) along with the extreme values (dotted lines). Local pollution plumes are responsible for the large envelope between the two extremes, particularly for NO and NO y .
[32] The modeled time series for NO, NO y , and O 3 (dashed lines) usually fall within the observed ranges. The model simulates summer afternoon median values of 0.11 ppbv NO and 3.21 ppbv NO y as compared to the observed 0.10 ppbv NO and 3.8 ppbv NO y , indicating biases of less than 20%. Poor correlations between the median observed and simulated time series (r 2 = 0.09 for NO, and r 2 = 0.22 for NO y ) may reflect the influence of subgrid plumes. Summer afternoon median concentrations of O 3 are 65 ppbv in the model and 62 ppbv observed, with an r 2 = 0.54 for the time series. The peaks in O 3 and NO y concentrations during the episode in mid-July (14 -15 July) are not captured properly; it rains in the model on these days, unlike in the observations.
[33] Concentrations of CO in the model are severely underestimated, by 100 ppbv or more (summer median is 110 ppbv in the model and 207 ppbv in the observations; r 2 = 0.38), with the exception of early July when the influence of the Canadian fires reaches Tennessee, and the model predicts values consistent with the observations (above 270 ppbv on 1 and 2 July). As discussed previously, missing background sources of CO may improve the model simulation [Duncan et al., 2001;submitted manuscript, 2002] but cannot provide the whole explanation. The model tends to capture CO concentrations at the Dickson County site, where surface measurements of CO were consistently lower than those at the other three sites. While the instrument at Dickson passed QA/QC protocols, aircraft measurements above Dickson are consistent with the observations at the other surface sites, leaving the discrepancy in surface CO concentrations unexplained [Luke et al., 1998].
[34] We conducted a more general evaluation of the model simulation over the United States by comparing the GEOS-CHEM simulation for 1995 with median or mean June -August observations available from earlier years  for a number of species at nonurban U.S. locations, as previously done by Horowitz et al. [1998] for evaluating an earlier version of the Harvard model. Results are shown in Figure 5. The locations of the measurement sites, length of data record (the entire set of observations spans 1975 -1995), and statistic used in the comparison are listed in Table 3 of Horowitz et al. [1998]. Modeled NO y and NO x are typically within 50% of the observations, without systematic bias. The model underestimates PAN by 30% on average. The sparsity of data (three sites) prohibits a conclusive discussion of the HNO 3 simulation. From the three available measurements, the model appears to overestimate H 2 O 2 by 1 ppbv, perhaps because it does not account for reaction with SO 2 in clouds. Of note, Bey et al.    Table 3 of Horowitz et al. [1998]. The solid line is y = x. The dashed line is the linear regression. production efficiency (number of O 3 molecules produced per molecule of NO x oxidized) [Trainer et al., 1993;Olszyna et al., 1994;Hirsch et al., 1996;St. John et al., 1998;Sillman, 1999]. Figure 6 shows the comparison with summer 1995 observations at Harvard Forest. The slope of the regression line is 7.5 mol/mol in the model and 10.6 mol/mol in the observations; the intercept (background O 3 ) is 35 ppbv in the model and 30 ppbv in the observations. These values are typical of previous summertime observations at Harvard Forest (8.0 mol/mol slope and 20-40 ppbv intercept) [Hirsch et al., 1996], and at Scotia, Pennsylvania (8.5 mol/mol slope and 35 ppbv intercept) [Trainer et al., 1993].

Aircraft Observations
[36] Two low -flying aircraft were deployed along the northeastern U.S. urban corridor as part of NARSTO-NE. Figure 7 shows measured O 3 distributions (colored lines) during horizontal afternoon transects between 400 and 800 m altitude on a highly polluted day (14 July 1995) and on a relatively clean day (10 August 1995). On 14 July, O 3 levels are well above 80 ppbv over most of the flight track, with a peak of 193 ppbv. Concentrations are below 55 ppbv over much of the region sampled on 10 August, with the largest values recorded over Long Island and southern Connecticut. The model (numbers in boxes) simulates the contrast in O 3 concentrations observed on these two days, but fails to reproduce the levels over 90 ppbv recorded on 14 July. Grid-scale dispersion of emissions prevents the model from capturing the extreme range of O 3 concentrations.
[37] Since subsidence from the free troposphere accounts for a large fraction of background O 3 in surface air, it is critical to inspect the model simulation of the chemical gradients between the boundary layer and the free troposphere. We previously summarized the extensive evaluation of model results with ozonesonde vertical profiles at northern midlatitudes [Bey et al., 2001a]. As part of the SOS campaign, the NOAA WP-3D aircraft flew on eighteen days in the boundary layer and free troposphere over Nashville and the surrounding region, measuring a suite of chemicals [Hubler et al., 1998]. In Figure 8, we show a comparison for three days (26 June, 1 July, and 11 July) that best exemplify the different structures in the vertical distribution of O 3 . On both 26 June and 1 July, the aircraft flew over Tennessee and Kentucky, while on 11 July, it headed north into Illinois, Indiana and Ohio. On 26 June, there is little structure in the O 3 profile. Concentrations of O 3 are relatively low in the boundary layer, and higher in the free troposphere where values are typical of the climatology [Logan, 1999]. The O 3 profile on 1 July exhibits more structure, with concentrations reaching above 80 ppbv near the surface, and decreasing to below 60 ppbv above the boundary layer. On the 11 Model results are shown as crosses. The reduced major axis method is used to calculate the regression line [Hirsch and Gilroy, 1984]. Background O 3 originating outside the North American boundary layer in the model is shown as triangles (see section 5 for details).
July flight, the aircraft sampled surface O 3 concentrations above 100 ppbv. Concentrations decrease sharply upwards from the surface to 2 km, where O 3 levels are below 50 ppbv. The vertical profiles of PAN and CO exhibit similar behavior to those of O 3 , with higher surface values measured on 1 July and 11 July. The vertical distributions of NO and CH 2 O show a sharp decrease from the boundary layer to the free troposphere without much variation over the three days, although CH 2 O concentrations are particularly high in the boundary layer on 11 July in association with the high-O 3 episode. Unlike the other species, H 2 O 2 concentrations do not generally decline from the boundary layer to the free troposphere, but instead remain fairly uniform until 4 km, where they begin to decrease.
[38] The model captures much of the spread in the surface observations of O 3 , CH 2 O, and H 2 O 2 , along with the day-to-day variability in the observed vertical distribution of O 3 . The overestimate of O 3 between 2 and 6 km apparent on 1 July and 11 July does not appear to be systematic when observations from the additional flight days are considered; the observations are particularly low above 1 km on 11 July. High CO concentrations simulated on 1 July are due to Canadian fire plumes . At the surface, peak values of NO, CH 2 O and PAN are sometimes underestimated. These measurements were most likely influenced by the Nashville plume or the plumes of neighboring power plants not resolved in the model. Above the surface, CH 2 O appears to be underestimated between 2 and 4 km on 1 July and 11 July. When we consider the entire SOS 1995 set of CH 2 O aircraft data, we find that on average, the model underestimates CH 2 O concentrations by 0.5-1 ppbv on average at all altitudes. The model shows a tendency to underestimate H 2 O 2 between 2 and 6 km, although it reproduces the surface concentrations fairly well. This result contradicts the surface overestimate in H 2 O 2 in the southeastern region noted in section 4.2 (at Metter, Georgia, and Kinterbish, Alabama). From 1980From to 1995 We conduct a simulation with 1980 levels of fossil fuel emissions in order to test that the model adequately captures the trends that have been reported in previous analyses of surface O 3 measurements over the United States from 1980 to 1995 by Lefohn et al. [1998] and Lin et al. [2000]. We examine differences in the O 3 simulation for the month of July in the 2°Â 2.5°resolution model, after a full year of spin-up with 1980 emissions at 4°Â 5°resolution. Any change in the O 3 simulation is fully attributable to the differences in fossil fuel emissions from 1980 to 1995 since all other parameters (e.g., meteorology, all other emissions) were held constant. Global anthropogenic emissions of NO x , CO, and nonmethane hydrocarbons for 1980 are implemented into the model by applying scaling factors (described previously in section 3.3) uniformly to individual nations; changes in CH 4 concentrations are not considered. Global NO x emissions from fossil fuel increased by 20% from 1980 to 1995, primarily due to a doubling of emissions from Asia (emissions increased by 2% over North America and decreased by 13% over Europe). Global fossil fuel emissions of CO and nonmethane hydrocarbons exhibited little change during this time period, with emission controls in Europe (35% decrease) and North America (15% decrease) offset by rising Asian emissions.

Response of Surface O 3 to Anthropogenic Emission Changes
[40] Previous analyses of surface O 3 measurements have shown that the highest levels of O 3 exhibited the largest decreases in response to the reductions in hydrocarbon emissions during 1980. Lefohn et al. [1998 observed what they called a ''piston effect' ': midlevel (60 -90 ppbv) O 3 concentrations at AIRS and CASTNET sites decreased     1980 -1984 versus 1994 -1998. [41] Figure 9 shows the simulated effect on the frequency distributions of July afternoon (1300 -1700 LT) O 3 in surface air for the three grid squares discussed in section 4.1. Over western Massachusetts (top panel), we find that the piston effect is qualitatively reproduced, with the highest O 3 values exhibiting the largest response to the reductions in U.S. hydrocarbon emissions between 1980 and 1995. In the lower half of the distribution, O 3 concentrations show a slight increase from 1980, consistent with the observations shown by Lin et al. [2000]. The simulated increases over western Tennessee (middle panel) and the Grand Canyon (bottom panel) can be attributed to the global rise in anthropogenic emissions in the model from 1980 to 1995. The enhancement to the background from Asian emissions is explored further in section 6 for the standard 1995 simulation. While the piston effect is not clearly manifested in the model over Tennessee, the highest values increased by smaller amounts as compared to the lower half of the distribution. Jacob et al. [1993b] previously showed a greater response of O 3 to hydrocarbon emissions controls in New York and Michigan than in Georgia in their model. They attributed this result to the role of abundant natural isoprene emissions in reducing the sensitivity of O 3 production to changes in anthropogenic hydrocarbon emissions in the southeastern United States.

Summary of Model Evaluation
[42] Our evaluation of the GEOS-CHEM model over the United States has identified four main problems: (1) excessive convective mixing over the Gulf of Mexico and the Caribbean, (2) poor resolution of the gradient in mixed layer depths between land and sea, (3) an inability to resolve topography over California, and (4) a substantial underestimate of CO concentrations. The first problem leads to an overestimate of background O 3 in surface air in the southeastern United States, while the second compromises the model simulation over coastal urban environments. The third problem is manifested over the Central Valley region of California. The fourth problem is difficult to fully attribute, but should be of little consequence for our study since O 3 production is largely NO x -limited.
[43] The model generally captures the day-to-day variability in species concentrations, both at the surface and aloft. Episodes of elevated O 3 over the eastern United States are reproduced although peak levels on scales smaller than the 2°Â 2.5°model resolution are underestimated (Figure 4). Much of the relevant variance in the O 3 distribution is simulated (Figures 2 and 3), with the exception of the background O 3 problem in the southeastern United States. The simulated O 3 :[NO y À NO x ] relationships, CH 2 O concentrations, and response of O 3 to global fossil fuel emission changes from 1980 to 1995 further imply a good simulation of the photochemical environment.

Background O 3 Over the United States
[44] We apply the model to quantify the contribution of transport from outside North America to afternoon O 3 concentrations in surface air over the United States. We define North America to be the region north of 13°N, and between 69°W and 126°W. We label, or tag, O 3 in the model by its region of production, as done by Wang et al. [1998c]. Tagging O 3 involves archiving 3-D fields of daily mean production and loss frequencies of the extended odd oxygen family (O x = O 3 + O + NO 2 + 2NO 3 + PANs + HNO 4 + HNO 3 + 3N 2 O 5 ) from our standard simulation. O x will be hereafter referred to as O 3 since O 3 usually accounts for over 95% of O x . The archived production and loss frequencies are used to drive an off-line simulation, in which total O 3 is divided into individual tagged tracers produced in different regions of the atmosphere. Each tracer is subjected to the same chemical loss and dry deposition frequencies as the total O 3 , but is produced only within its specific source region. This approach allows surface O 3 over the United States to be deconstructed into two separate components: (1) O 3 produced inside the North American boundary layer (defined as extending up to 3 km altitude or $700 hPa) and (2) the O 3 background (produced outside). The O 3 background is further decomposed into contributions from tropospheric production and stratospheric injection.
[45] Five months of initialization (January -May 1995) were found appropriate for the tagged tracer model. After this time the initial conditions contribute only a few percent to the total O 3 . The sum of O 3 from all source regions is within a few percent of the total O 3 in the standard fullchemistry simulation.
[46] Production in the North American boundary layer is found to account for 50% of surface afternoon (1300 -1700 LT) O 3 concentrations over the United States in summer, and the remainder is mainly from tropospheric production outside the region. We find that stratospheric injection of O 3 never contributes more than 2 ppbv to these concentrations. This result is consistent with that of Follows and Austin [1992], who showed that O 3 originating in the stratosphere composes less than 5% of zonally averaged O 3 near the surface.
[47] Figure 10 shows the mean summer afternoon (1300 -1700 LT) background O 3 concentration in the model surface layer. Average background concentrations are 15-35 ppbv, with highest values in the west. The larger background in the west reflects the arid climate with its deep boundary layer mixing, slow deposition velocities, and long O 3 lifetime.
[48] Indeed, the lifetime of O 3 in the U.S. boundary layer is critical for understanding the magnitude of the background. The mean lifetime of O 3 in the GEOS-CHEM mixed layer in summer is displayed in Figure 11. This lifetime is calculated from 24-hour average O 3 concentrations, deposition velocities, and O x chemical loss rates summed over the local afternoon (1300 -1700 LT) mixed layer depth (Figure 1b). The afternoon mixed layer is assumed to represent the depth of the planetary boundary layer that is in daily contact with the surface. In the western states, the chemical lifetime for O 3 in the mixed layer is as long as 10 days, while the lifetime against dry deposition is about 6 days (mixing depths are about 2 km on average, with average deposition velocities of 0.4 cm s À1 ), yielding an O 3 lifetime of 3 -5 days ( Figure 11). The lifetime in the eastern states is typically less than 2 days, reflecting shallow mixing depths (1 km on average), faster deposition velocities (0.5 cm s À1 on average), and more active chemical loss. The short chemical lifetime of O 3 in the east is due in part to reactions with biogenic hydrocarbons. Table 4 quantifies the relative importance of various loss pathways for O 3 in the mixed layer for different regions. Reaction of O 3 with biogenic hydrocarbons contributes 10-20% to O 3 loss in the eastern United States while deposition of HNO 3 contributes 5 -10%. Aqueous reaction with SO 2 in clouds is a negligible sink for O 3 [Jacob, 2000].
[49] The short lifetime of O 3 in the mixed layer over the eastern United States effectively reduces the contribution of the background during episodic pollution events associated with stagnation and strong subsidence inversions. To illustrate this point, we show in Figure 6 (triangles) the simulated afternoon background at Harvard Forest, as a function of (NO y À NO x ), which serves here as an index of aged pollution. The background decreases with increasing (NO y À NO x ), and drops below 12 ppbv at the highest O 3 concentrations. Figure 6 indicates that inferring a background contribution to episodes of elevated O 3 from either average conditions or from the intercept of an O 3 :(NO y À NO x ) correlation plot leads to a substantial overestimate of [50] Our general finding that background O 3 is low during high-O 3 episodes leads us to examine whether the background is ever substantial on days when O 3 is elevated above 80 ppbv. Figure 12 presents the probability distribution of afternoon background O 3 concentrations in the model surface layer for two populations: (1) the ensemble of summer afternoon 1995 data in the United States and (2) the data subset in which total afternoon surface O 3 was above 80 ppbv (dashed line). The mode for the full summer 1995 data set is 23 ppbv, whereas it is 11 ppbv for the polluted data subset. There is, however, a small population from that subset (21 occurrences spanning 12 summer days) that have background concentrations above 30 ppbv and total surface O 3 above 80 ppbv. This population increases to 59 occurrences spanning 26 summer days if we consider background concentrations above 25 ppbv. Further analysis reveals that these values occur on days with convective precipitation, which is associated with subsidence from above that transports background O 3 down to the surface. Alongside the subgrid-scale convective activity in the model, there is enough photochemical production of O 3 in the mixed layer to sustain elevated levels of surface O 3 . This finding reveals that conditions do exist under which the background O 3 can contribute substantially to total surface concentrations above 80 ppbv. However, some of the background transported convectively to the surface may have been produced from North American anthropogenic precursors previously lofted to the free troposphere. Further  analysis of the origin of the background is presented in the next section.

Origin of the Background
[51] Because of chemical nonlinearity, the background O 3 defined by the tagged tracer simulation cannot be quantitatively related to precursor emissions; sensitivity simulations with modified emissions are necessary. In this section, we use a series of sensitivity simulations to quantify (1) the natural background and (2) the enhancement to the background from anthropogenic emissions outside of North America.
[52] We conducted three sensitivity studies in the coarser resolution (4°Â 5°) model in which we removed all anthropogenic emissions of NO x , CO, and nonmethane hydrocarbons (including NO x emitted from aircraft and fertilizer, but not biomass burning): (1) globally, (2) within North America, and (3) only outside of North America. Natural sources of NO x from soil and lightning remain. We consider biomass burning as a natural source because of the lack of information on trends in biomass burning over the past century. Anthropogenic biomass burning is minimal during the summer months in the Northern Hemisphere, with episodic boreal fires (over two thirds of which are ignited naturally by lightning [Stocks, 1991]) predominantly responsible for biomass burning emissions (Duncan et al., submitted manuscript, 2002).
[53] We define the North American domain to be north of 12°N latitude and between 62.5°W and 127.5°W longitude. All sensitivity simulations are initialized for a full year from 1 June 1994 through 31 May 1995 and results from 1 June to 31 August 1995 are examined. One year provides sufficient initialization to remove the effects of initial O 3 concentrations on the results.
[54] Sensitivity simulation 1 serves to isolate a ''natural'' background. Figure 13a displays the resulting distribution of O 3 in surface air over the United States in summer. The corresponding spatial concentration statistics are shown in Table 3. The average afternoon (1300-1700 LT) concentration from June-August is 25 ppbv (compared to 51 ppbv in the standard simulation), with a spatial variance of 10 ppbv 2 (72 ppbv 2 in the standard simulation). Similar to the results from our tagged tracer study, we find the natural background to be highest in the arid southwestern United States. In comparison to the results presented here, preindustrial model simulations show surface O 3 concentrations over the United States of 10-20 ppbv; these lower values reflect the effect of lower concentrations of preindustrial methane, and the use of 24-hour average values as compared to the afternoon values we present here Mickley et al., 1999;Lelieveld and Dentener, 2000].
[55] Figure 13b shows results from sensitivity simulation 2 which includes anthropogenic emissions only outside of North America. The mean summer afternoon O 3 concentration in the model surface layer over the United States is 30 ppbv, with a spatial variance of 12 ppbv 2 (Table 3). Figure 13c shows the difference between simulations 2 and 1, which represents the enhancement in the O 3 background in the United States due to anthropogenic emissions outside North America (i.e., from Asia and Europe). We find that this enhancement to summer afternoon O 3 concentrations is 4 -7 ppbv on average and is largest in the west. [56] Figure 14 displays the results of the sensitivity simulations in terms of the frequency distribution of surface afternoon O 3 in the standard simulation at Harvard Forest (open triangles). The natural background as defined by sensitivity simulation 1 (open squares) does not exhibit the decrease with higher O 3 that was found in the tagged tracer simulation discussed in section 5 (solid triangles in Figure 6). The tagged tracer simulation did not account for O 3 production from natural precursors in the North American boundary layer, which compensates for decreased transport from the free troposphere during stagnation events. The impact of anthropogenic emissions outside North America (open diamonds in Figure 14) is 2 -9 ppbv over the range of conditions encountered at Harvard Forest in the model, and tends to be highest for O 3 concentrations in the range 55-75 ppbv.
[57] In Figure 15, we examine the O 3 enhancement from Asian and European emissions for the entire population of summer afternoon O 3 concentrations over the United States in the model surface layer. The decrease in this enhancement for the highest O 3 concentrations (>80 ppbv) is consistent with our finding in section 5; the lifetime of O 3 is short enough that the background is depleted under the stagnant conditions associated with high-O 3 episodes. Even so, Asian and European emissions enhance summer afternoon O 3 levels in surface air by 2 -7 ppbv under these episodic conditions. The low tail of the O 3 frequency distribution (below 40 ppbv) is associated with only small Figure 14. Cumulative probability distribution of simulated afternoon O 3 concentrations in the model surface air at Harvard Forest, Massachusetts (triangles). The corresponding contributions from the natural background are shown as squares (sensitivity simulation 1) and the anthropogenic enhancement due to emissions in Asia and Europe is shown as diamonds (difference between sensitivity simulations 2 and 1). enhancements from emissions outside of North America. We attribute these clean values to boundary layer transport of air masses originating from the tropics or the Arctic, with minimal impact from Asian and European emissions. Generalizing from the results at Harvard Forest (Figure 14), we observe in Figure 15 a maximum influence of Asian and European emissions (up to 14 ppbv) for concentrations of 50 -70 ppbv. These conditions reflect a combination of subsidence of O 3 from the free troposphere (often associated with convective precipitation events) and fast photochemical production of O 3 in the U.S. boundary layer, as discussed in section 5; the present analysis allows us to quantify the exogenous pollution component of this subsiding background. Our result implies that if a more stringent national O 3 standard were to be adopted in the United States (such as the 60 ppbv standard in Japan and many European nations), it would be essential to consider the major role played by Asian and European anthropogenic emissions in exceedances of the standard.
[58] Sensitivity simulation 3, with anthropogenic emissions only within North America, was conducted to explore how chemical non-linearities may affect our estimate of the impact of foreign emissions on domestic air quality (Figure 13c). This simulation yields a mean summer afternoon surface O 3 of 46 ppbv, which is 5 ppbv less than the value from the standard simulation (Table 3). The difference between the standard simulation and simulation (3) enables us to make a second estimate of the impact of anthropogenic emissions from outside North America on surface background concentrations in the United States. We find the range of influence derived from this approach to be consistent with our previous approach: 4 -7 ppbv, with a similar spatial distribution. The difference between sensitivity simulations 1 and 2 on the one hand, and between the standard simulation and sensitivity simulation 3 on the other hand, allows us to bracket the effect of chemical nonlinearity in terms of quantifying the O 3 enhancement over North America due to anthropogenic emissions from outside. We find that the effect of nonlinearity is less than 1 ppbv.

Conclusions
[59] We have applied a three -dimensional global model of tropospheric chemistry driven by assimilated meteorological observations for 1995 to identify the origin of background O 3 in surface air over the United States on summer (June through August) afternoons (1300 -1700 LT). A tagged tracer simulation enabled us to quantify the contribution to O 3 from chemical production outside the North American boundary layer, and to examine the importance of the background on both average and highly polluted days. Sensitivity simulations were employed to quantify the natural component of this background and the enhancement from emissions outside of North America (from Asia and Europe).
[60] Extensive evaluation of model results with 1995 observations from several networks was conducted to assess the ability of the model to simulate regional chemistry over the United States under the range of conditions from relatively clean to highly polluted. Comparison with surface and aircraft measurements from the summer of 1995 indi-cates that the GEOS-CHEM model reproduces much of the observed variability in the summertime afternoon O 3 distribution, including the observed high-O 3 events. Consistent with previous analyses of O 3 observations in surface air over the United States [Lefohn et al., 1998;Lin et al., 2000], we find that a comparison of simulations with 1995 versus 1980 global fossil fuel emissions reveals a decrease in the high end of the O 3 probability distribution (due to domestic hydrocarbon emission controls) and an increase in the lower half of the distribution (from rising Asian emissions). The most severe limitation for our study is the inability of the model to capture background levels of O 3 on clean days in the southeastern United States. This problem arises from excessive convection over the Gulf of Mexico and the Caribbean, which mixes down too much O 3 from aloft.
[61] With our tagged O 3 tracer simulation, we find that chemical production outside of the North American boundary layer (surface -700 hPa) contributes an average 15-35 ppbv background to summertime afternoon O 3 in surface air over the United States. This background is mainly produced in the free troposphere. Contribution from the stratosphere is negligible (less than 2 ppbv). The background is higher in the western than in the eastern United States, reflecting the longer O 3 lifetime in the west. We find that the lifetime of O 3 (actually the extended odd oxygen family) in the mixed layer over the eastern United States is only about 2 days because of shallow mixing depths, relatively high deposition velocities, and significant contributions to O 3 loss from reactions with biogenic hydrocarbons and HNO 3 deposition. Because of this short lifetime, background O 3 concentrations in the east generally drop to below 12 ppbv during high-O 3 episodes associated with regional stagnation. Consequently, an estimate of the background O 3 contribution based on observations under clean conditions severely overestimates the actual background under the heavily polluted conditions associated with exceedances of the national air quality standard.
[62] We find that in 3% of events where O 3 exceeds 80 ppbv in the model, background O 3 transported from outside the North American boundary layer contributes 30 -40 ppbv. In 9% of all such events, background O 3 contributes above 25 ppbv. These events are associated with deep convection upstream that entrain background O 3 from the free troposphere, followed by rapid production of additional O 3 within the U.S. boundary layer.
[63] Our sensitivity simulations show that anthropogenic emissions from Asia and Europe enhance afternoon O 3 in surface air over the United States in summer afternoons by 4-7 ppbv on average. This enhancement is maximum for O 3 concentrations in the 50-70 ppbv range when subsidence from the free troposphere associated with convection events combines with subsequent O 3 production in the U.S. boundary layer. The enhancement from these exogenous anthropogenic emissions is less (2 -7 ppbv) under the heavily polluted conditions leading to exceedances of the 80 ppbv threshold. Nevertheless, it is important to consider this relatively small influence of Asian and European emissions in the context of the substantially less-than-linear response of O 3 to domestic emission controls Jacob et al. 1999;Sillman, 1999;Ryerson et al., 2001]. If emissions from developing countries follow their projected increasing trends [Prather and Ehhalt, 2001], we can expect this contribution to rise in future years. Furthermore, if the O 3 standard in the United States is lowered to below 70 ppbv in the future (as it already is in many nations, to protect vegetation as well as human health), then consideration of the large enhancement from anthropogenic emissions outside North America will become an essential element in the development of an emission control strategy.