The impact of climate change on U.S. surface ozone levels is investigated. We simulated two 10 year periods using the global chemical transport model MOZART-2 (Model of Ozone and Related chemical Tracers version 2): 1990–2000 and 2090–2100. In each case, MOZART-2 is driven by meteorology from the National Center for Atmospheric Research (NCAR) coupled Climate Systems Model (CSM) 1.0 forced with the Intergovernmental Panel on Climate Change (IPCC) Special Report on Emissions Scenarios (SRES) A1 scenario. During both periods the chemical emissions are fixed at 1990s levels, so that only changes in climate are allowed to impact ozone. The simulated surface ozone concentration during the 1990s is compared with observations from the Environmental Protection Agency's Aerometric Information Retrieval System (EPA AIRS) monitoring sites. Model-measurement correlations are high, but MOZART-2 overpredicts ozone especially over the eastern United States. The impact of climate change is calculated separately for background ozone and for the ozone generated through U.S. NOx emissions. Our results show that the response of ozone to climate change in polluted regions is not the same as in remote regions. MOZART-2 predicts a 0–2 ppbv decrease in background ozone in the future simulation over the United States but an increase in ozone produced internally within the United States of up to 6 ppbv. The decrease in background ozone is attributed to a future decrease in the lifetime of ozone in regions of low NOx. Over the western United States the decrease in background ozone approximately cancels the increase in locally produced ozone. As a result, the main impact of future climate change on ozone is centered over the eastern United States, where future ozone increases up to 5 ppbv. We predict that in the future over the northeast United States, up to 12 additional days each year will exceed the maximum daily 8-hour averaged ozone limit of 80 ppbv. Various climatic factors are identified which impact the net future increase in ozone over the United States including changes in temperature, water vapor, clouds, transport, and lightning NOx. Significant future changes are generally not found in planetary boundary layer height and precipitation.
 The surface concentration of ozone affects human health and the environment. Because of its importance the U.S. Environmental Protection Agency (EPA) maintains an extensive network of sites to monitor ozone levels. Surface ozone is mainly produced from volatile organic compounds (VOCs), CO and NOx (NO and NO2) which are emitted from traffic, industries and biogenic sources [Environmental Protection Agency (EPA), 2002]. According to EPA  both 8-hour and 1-hour averaged ozone levels have improved considerably within the United States with a 12% decrease in the maximum daily 8-hour averaged (MDA8) ozone concentration between 1981 and 2000. The preliminary conclusion [EPA, 2002] is that this improvement is due to a reduction in anthropogenic VOC and NOx emissions. This paper addresses the expected ozone changes over the next 100 years in a warmer climate.
 The ozone concentration over the United States can be considered to be composed of two components: ozone produced from ozone precursors emitted in the United States and ozone produced from other sources, including that transported from the stratosphere. The ozone not produced from U.S. precursors is often referred to as background ozone. While the ozone produced inside the United States plays the most significant role in U.S. air quality, the contribution of background ozone to surface air quality cannot be ignored [Jacob et al., 1999; Lin et al., 2000; Lefohn et al., 2001; Fiore et al., 2002, 2003]. Using a 3-D global model of tropospheric chemistry driven by assimilated 1995 meteorology, Fiore et al.  investigated the origin of background ozone over the United States on summer afternoons. They showed that anthropogenic emissions from outside United States enhanced afternoon surface ozone over the United States by 4–7 ppbv in summer. Berntsen et al.  and Jaffe et al.  show that Asian emissions influence air quality over the northwestern United States, especially during spring. They estimated that the present Asian ozone contribution during springtime is 4 ppbv over the west coast of Washington State. Using 18 years of observations, Jaffe et al.  show that background ozone has been increasing on the rural west coast of North America. They suggest that this increase is due to increased emissions of ozone precursors from Asia. Jacob et al.  further predicts that increased Asian emissions over the next decade will have significant impact on the western United States air quality.
 A number of studies have used global chemical transport models and future emission scenarios to predict future ozone concentrations. Prather et al.  reported on 14 independent model simulations using Intergovermental Panel on Climate Change (IPCC) Special Report Emission Scenarios (SRES). They found the change in zonal mean surface ozone in the 21st century between 25° and 40°N to be particularly large as surface ozone increased in summer by more than 25 ppbv. The increases were especially large under the SRES A2 and the A1FI scenarios which have overall higher emissions. However, these model simulations were not run using 21st century meteorology. Among other factors, changes in atmospheric dynamics, cloudiness, boundary layer height, water vapor, temperature and the production of NOx by lightning may alter these conclusions.
 A number of studies [Brasseur et al., 1998; Grewe et al., 1999, 2001; Johnson et al., 1999, 2001; Stevenson et al., 2000] suggest that climate change itself, particularly the increase in relative humidity will act to decrease ozone concentrations in the future. Zeng and Pyle  predict climate change increases tropospheric ozone due to enhanced stratosphere-troposphere exchange. The effect of changes in both climate and emissions has also been considered in a number of model simulations [e.g., Brasseur et al., 1998; Grewe et al., 1999, 2001; Johnson et al., 1999, 2001; Stevenson et al., 2000; Zeng and Pyle, 2003; Grenfell et al., 2003]. For example, Johnson et al.  applied the 3 dimensional Lagrangian chemistry model STOCHEM using an emission estimate from the IPCC IS92a scenario to a future climate. They found that global mean tropospheric ozone increased by 6.4 ppbv between 1990 and 2075 because of changes in both emissions and climate. Hogrefe et al.  used a regional model centered over the eastern United States to examine U.S. air quality in the future. They showed that large changes in the average MDA8 ozone concentration may be expected by 2080 because of changes in the import of ozone to their regional model domain, regional climate change and changed emissions. However, because of the nature of their study they could not account for the impact of changes in the global climate on their chemical boundary conditions.
 The purpose of this study is to examine how climate change affects ozone levels near the surface, with a focus on changes in U.S. air quality. We simulate two 10 year periods using a global chemical transport model: 1990 through 2000 and 2090 through 2100. During both periods we keep the chemical emissions constant, so that only changes in climate are allowed to impact ozone. In particular we examine the effect of climate change on the two components of U.S. ozone: the background ozone and on the ozone produced from precursors emitted in the United States.
 To simulate the impact of climate change on U.S. air quality we drive the chemical transport model, MOZART-2 (Model of Ozone and Related chemical Tracers version 2) with meteorological fields generated using the National Center for Atmospheric Research (NCAR) coupled Climate Systems Model (CSM) 1.0 [Boville and Gent, 1988] forced with the IPCC SRES A1 scenario. We simulate a control period (1990–2000) and a future period (2090–2100). Further details on these CSM 1.0 simulations can be found in Mahowald and Luo . During the two 10 year periods the wind, temperature, surface pressure, specific humidity and surface fluxes of heat and surface stress are saved from the CSM 1.0 simulation at a frequency of three hours for input into MOZART-2.
 Except for the model emissions (see below), the version of MOZART-2 used here includes only minor modifications to that described by Horowitz . Evaluations of the model against measurements are given by Horowitz , Emmons et al.  and Tie et al. . The model transport is based on the MATCH model [Rasch et al.,1997; Mahowald et al., 1997]. It uses the flux-form semi-Lagrangian advection scheme of Lin and Rood , the shallow convection scheme by Hack , the deep convection of Zhang and McFarlane , and the boundary layer parameterization of Holtslag and Boville . The model chemistry consists of 63 chemical species and 140 chemical reactions. The source of NOx from lightning is parameterized following Price et al.  and depends on the diagnosed convective cloud height and the cold cloud thickness. The vertical distribution of the NOx source follows a C-shape profile based on Pickering et al. . Depending on the frequency and height of the diagnosed model convection the source of lightning NOx varies both seasonally and interannually. In particular the average global lightning NOx source between 1990 and 2000 is 3.9 ± 0.1 Tg N yr−1, but 4.5 Tg ± 0.09 N yr−1 between 2090 and 2100.
 Surface emissions are kept the same during both simulation periods so as to isolate the effect of climate change alone on U.S. air quality. Except for isoprene and terpene, the emissions used in these simulations are valid for 1997 and are from the Precursors of Ozone and their Effects in the Troposphere (POET) Project [Olivier et al., 2003]. The emissions of the nonmethane hydrocarbons (NMHCs) have been adapted to the MOZART chemical mechanism. The POET anthropogenic, biomass burning, biofuel and agricultural waste emissions are based on the Emission Database for Global Atmospheric Research 3.2 (EDGAR 3.2) emissions (see Olivier et al.  and van Aardenne et al.  for a discussion of EDGAR emissions). The emission factors from Andreae and Merlet  are used for biomass burning. Additional biogenic and oceanic emissions from Müller and Brasseur  are also included in the POET database. The isoprene and terpene emissions are based on the Guenther et al.  emission algorithm. We derive the emissions using a 10-year climatology based on the National Centers for Environmental Prediction (NCEP) meteorological reanalysis. In this study the biogenic emissions remain fixed on an interannual basis. In particular the future emissions are the same as used in the control simulation and do not respond to future climate change. The resulting isoprene emissions over the United States during July are 6.4 TgC. This value is between the U.S. isoprene emissions from the July 1996 Global Emissions Inventory Activity (GEIA) (7.1 TgC) and those estimated by Palmer et al.  using Global Ozone Monitoring Experiment (GOME) formaldehyde measurements (5.7 TgC). The U.S. emissions of NOx used here (1.64 107 kg N/day) underestimate the EPA estimate for 1999 (2.08 107 kg N/day) by approximately 20%. The mixing ratio of CH4 is specified at current values in the lowest model layer as a function of latitude, longitude and season to remove any long-term model drift.
 The model resolution used in this study is approximately 2.8° latitude x 2.8° longitude with 18 hybrid vertical levels. The lowest level is located at approximately 993 hPa. To distinguish between ozone generated inside the United States and imported from outside the United States, we tag the U.S. emissions of NOx. This tag is consistently maintained throughout the full suite of oxidized nitrogen species. The ozone produced through the photolysis of tagged NO2 is also tagged and defined as O3US (except for a couple of minor reactions, tropospheric O3 is only produced within MOZART through the photolysis of NO2). O3US is destroyed through the same set of reactions which destroy O3. Thus O3US serves as a tag of ozone produced anywhere in the troposphere through U.S. NOx emissions. This procedure has been tested by tagging all modeled NOx emissions. In this test it is found that the tagged ozone very closely reproduces the original O3 distribution. Since the troposphere is generally NOx limited, tagging NOx is a reasonable method for attributing ozone to particular emission sources.
where (R2) represents the primary loss of ozone due to increased water vapor. In these simulations we also find that the density weighted troposphere ozone (defined simply as the ozone on all model levels below 300 hPa) decreases in both hemispheres during the future simulation (not shown). In both hemispheres the future ozone decrease is much larger during the summer season.
 We have also included a tracer of stratospheric ozone in these simulations (O3S). This tracer is relaxed to the stratospheric ozone concentration above the tropopause, but is destroyed in the troposphere at the computed loss rate of odd oxygen. With no change in future stratosphere-troposphere exchange, increased water vapor in the future climate will reduce the future tropospheric burden of O3S [see Collins et al., 2003]. In our simulations the Northern Hemisphere (N.H.) density weighted tropospheric burden of O3S is larger during winter and spring in the future, but it is smaller during the summer and fall. In the Southern Hemisphere (S.H.) the tropospheric burden of O3S is generally larger during all seasons in the future, although most notably during winter. Thus, as in the works by Collins et al.  and Zeng and Pyle , we find evidence for increased future stratospheric-tropospheric exchange of ozone, particularly during the winter months. During summer large future increases in the ozone loss rate act to mask any increases in stratosphere-troposphere exchange. Note, however, that even in the future simulations of the N.H. summer, when the column burden of O3S decreases, O3S is somewhat elevated in the vicinity of the subtropical jet implying increased future exchange. As the tropospheric burden of ozone decreases during all seasons in the future, the future change in tropospheric ozone loss must override any increases in stratosphere-troposphere exchange.
Figure 1 shows the average summertime June, July and August (JJA) change in ozone between the future and control simulations in the lowest MOZART layer. The significance of the results are evaluated with Student's t-test using 9 years of data for the future and control simulations (allowing 1 year each for spin-up) assuming a 95% confidence level (subsequent comparisons are made in the same manner). Figure 1 demonstrates that ozone generally decreases in response to future climate change at the surface. The zonally averaged response in Figure 1 is negative at all latitudes (not shown).
 The largest decreases in future ozone are largely confined to the northern oceans. Regions with increased future ozone are generally found over land: the northeast United States, Africa and South America. These regions coincide with locations of large NOx emissions, including large biomass burning emissions (i.e., in Africa and South America). It is clear from Figure 1 that the effect of climate change on ozone is different between those regions with low NOx emissions and those regions with higher emissions. The mechanisms for this are discussed below in the context of ozone change over the United States. As future NOx emissions evolve the locations where global change acts to further degrade air quality will also evolve.
 The response of ozone to future climate change over the United States is relatively large and positive. It is statistically significant over much of the country east of the Rockies and north of the Gulf coast (Figure 1). First we evaluate surface ozone over the United States in our control simulation against measurements (section 4). Section 5 identifies various processes which affect ozone in a future climate over the United States. Some of these processes are applicable only to the U.S. mainland because of its particular geography and climate. Other identified processes will be equally valid in other regions. Section 6 discusses future changes in the background and nonbackground components of ozone over the United States. Changes in the import and export of ozone are discussed in section 7. The paper concludes with a general discussion of the results.
4. Evaluation of MOZART
 In this study, we mainly focus on surface ozone over the United States in summer, as it is during the summer season during which most of the exceedances of the Environmental Protection Agency's (EPA) ozone standard occurs. Here we evaluate the control run of MOZART (for JJA from 1991 to 1999) against measurements from the EPA's Aerometric Information Retrieval System (AIRS) monitoring sites from 1995 to 2000. In each case we compare the measured and simulated average MDA8 ozone concentrations, the standard used by the EPA for assessing ozone violations. For this comparison the measured station data are binned onto the MOZART grid. Figures 2a and 2b show the average JJA modeled and measured concentrations. Figure 2c shows the difference between the measured and modeled MDA8 ozone. Averaging stations onto the MOZART grid tends to average out local regions of high ozone concentrations.
Figure 3 shows the summer simulated versus measured MDA8 over the western and eastern United States and Table 1 summarizes the comparison between MOZART and the EPA data. Over the western United States the simulated surface ozone shows good agreement with the average, maximum and minimum measured concentrations (Table 1). MOZART clearly overestimates the surface ozone concentration over the eastern United States, particularly at high ozone concentrations (Table 1 and Figures 2c and 3). In both the western and eastern United States the spatial correlations between the modeled and measured MDA8 ozone are high.
Table 1. Statistics for Simulated MDA8 Ozone (ppbv) Averaged From 1991 to 1999 During JJA Over the Eastern (Points East of 100°W) and Western United States (Points West of 100°W), Where the Model is Only Sampled at EPA Measurement Sitesa
Western United States
Eastern United States
Minimum (maximum) values give the 9 year averaged minimum (maximum) concentration within each region.
 The causes of the high ozone bias in MOZART, and its sensitivity to model processes is under continuing investigation. There are a number of processes which MOZART does not include which might alleviate the bias. Like most global models MOZART does not include elevated point sources of emissions, or a module to parameterize the chemical effect of these sources. This version of MOZART also does not include aerosol induced changes in photolysis rates, nor the loss of NO2, HO2, NO3 or O3 on aerosols. However, Fiore et al.  estimates photolytic changes due to aerosols have little affect on ozone over the U.S. Model simulations by Tie et al.  and Martin et al.  also suggest that the net effect of aerosols over the United States results in only a small decrease in ozone.
 Surface ozone is difficult to simulate in global models because of the nonlinearity of the chemistry and the heterogeneity of surface emissions of photochemical precursors. At high NOx concentrations ozone production is less efficient per NOx molecule [Liu et al., 1987]. In particular, simulated ozone production in a coarse grid is expected to be overestimated as the emissions of ozone precursors are artificially diluted [Sillman et al., 1990; Liang and Jacobsen, 2000]. Kumar et al.  shows that increasing model grid size leads to increased ozone production. Thus we would expect that MOZART, and other global models with coarse representations of emissions to overestimate measured ozone. Ozone also increases dramatically in MOZART when the emissions are distributed through the depth of the boundary layer. This implies an important sensitivity to the vertical model resolution, such that decreasing the vertical resolution increases the ozone production. We note that the vertical resolution of most climate models, including the CSM, is rather coarse.
 In the analysis below we concentrate on ozone differences between the future and present simulations. It is unlikely that these differences will be significantly affected by MOZART's positive bias in surface ozone. Lamarque et al.  shows that the ozone change in MOZART between 1890 and 1990 is comparable to that found in other studies. Furthermore, in a series of sensitivity experiments to different sets of emissions, little difference was found between the relative ozone perturbations of MOZART and GEOS-CHEM (A. Fiore, personal communication, 2004).
5. Changes in Factors Affecting U.S. Air Quality: 1990–2090
 The meteorology in the future experiment is obtained using the CSM forced with the IPCC SRES A1 scenario. Details of the climate in the CSM during the 2090s can be found in Dai et al. , albeit assuming a different emission scenario than used here. They used A Consortium for the Application of Climate Impact Assessments Business as Usual (ACACIA-BAU) climate forcing scenario. The resulting climate forcing is similar to the IPCC SRES A1 scenario used here. Here we discuss details of the climate change particularly pertinent to ozone chemistry.
 The surface temperature over the United States increases by 2–4°C in the future simulation (Figure 4a). Modeling studies by Sillman and Samson  and Aw and Kleeman  show that summertime ozone concentrations increase as temperature increases. Sillman and Samson  show that this sensitivity is largely driven by a decrease in the net formation of PAN. The production of PAN both ties up NOx and reduces the source of peroxy radicals. This is consistent with our simulations (see Figure 10) which show future decreases in the concentration of PAN and increases in NOx. The impact of temperature increases should predominantly affect ozone production in regions of high emissions. This provides a partial explanation for the asymmetrical response in future ozone between polluted locales and the ocean basins seen in Figure 1.
 There is a significant 10–20% increase in water vapor (Figure 4b) in the future simulation over much of the eastern and central United States. The largest increase in water vapor occurs over the central United States. An examination of circulation differences between the future and control simulations suggests increased onshore flow from the Gulf of Mexico in the future (not shown). This increased flow is particularly pronounced over the central United States, coincident with the maximum change in water vapor. The location of the increased onshore flow is consistent with an increase in the strength of the Great Plains low-level jet. Various factors have been proposed which modulate the strength of this jet including flow over the Rocky Mountains, thermodynamic processes related to heating over the Great Plains, sea surface temperature anomalies and remote teleconnections including those connected with El Niño/Southern Oscillation (ENSO) [see Byerle and Paegle, 2003]. It is difficult to determine if the future change in onshore flow can be explained by changes in climate. Hu  shows large interannual variations in the moisture flux to the central United States between 1950 and 2000.
 The effect of increased water vapor on ozone, and its relation to NOx emissions, can be inferred from earlier work on changes in stratospheric ozone. Liu and Trainer  and Fuglestvedt et al.  have shown that decreases in stratospheric ozone can be expected to increase tropospheric ozone production when NOx concentrations are sufficiently high. Hydroperoxy radical production is enhanced through increased ozone photolysis (reactions (R1)–(R3)). At low NOx this leads to ozone destruction (R4), but at sufficiently high NOx increased hydroperoxy radical production acts to convert NO to NO2 and thus increase ozone production:
It is clear that an increase in water vapor in (R2) acts in the same way as an increase in the photolysis rate of ozone in (R1). In particular, increases in water vapor are expected to increase ozone in regions of high NOx concentrations, but to decrease ozone in regions of low NOx. This is consistent with the differences seen in Figure 1.
 In the future simulation large and significant increases in photolysis rates occur across much of the United States at the surface (see Figure 5): the photolysis rate of O3 → O1D increases up to 10% and the photolysis rate of NO2 increases up to 4% (not shown). The photolysis increase at the surface results from a decrease in the low-level cloud water (Figure 6). The decreases in cloud water are statistically significant over most of the United States, except in a swath in the center of the country coincident with the large water vapor increases. The decrease in cloud water is consistent with the findings of Del Genio and Wolf  who show observationally that the low-cloud liquid water path over the Great Plains decreases with increasing temperature because of a decrease in cloud thickness. As discussed above, an increase in the photolysis rate of ozone can either increase or decrease ozone depending on the NOx concentration. Sillman and Sampson  show that in both urban and rural environments typical of the eastern United States a general increase in photolysis rates acts to increase ozone.
 No significant future large-scale changes in precipitation were found over the United States during the summer months. However, the column production of lightning NOx significantly increases (Figure 7) over much of the central and eastern United States in the future run, coincident with the region of large water vapor increases. Price and Rind  also show an increase in lightning over the United States in a climate with doubled CO2. As the parameterized production of lightning NOx is nearly proportional to cloud height to the fifth power [Price et al., 1997], lightning NOx is sensitive to even small changes in cloud height. It is certainly possible that cloud height will increase in the 2090s because of the warmer and wetter boundary layer. Because of the assumed C-shaped vertical distribution of lightning NOx [Pickering et al., 1998] changes in lightning will have an immediate effect on the chemistry of the continental boundary layer.
 Regional high-pollution episodes generally occur under anticyclonic conditions dominated by clear conditions, high temperatures, low wind speeds and restricted boundary layer ventilation [e.g., Rao et al., 2003; Vukovich and Sherwell, 2003]. Mickley et al.  argue that these conditions are likely to be more prevalent in the future because of a decrease in the frequency of synoptic systems ventilating the boundary layer leading to an increase in the severity of summertime pollutant episodes. The cumulative frequency distribution of surface pressure is shown in Figure 8 in both the future and control simulations within a study region. This region is located in the eastern and central United States (between 35° and 45°N and 280° and 264°W) and is represented by 24 grid points. We chose the study area to encompass the region of large ozone changes (see Figure 12), but not to comprise any ocean grid points. Pressure differences between the future and control simulations are important at the lower pressures, suggesting that frontal passages during the summer months will not be as intense in a future climate. At the high-pressure end of the cumulative frequency distribution there is very little difference between the future and control simulations.
 Not only are the frontal passages weaker in the future simulation, but the synoptic frequency of frontal passages decreases. The standard deviation of the 2–6 day band-passed surface pressure decreases by 11% in our study region. Dai et al.  found a similar decrease in storm track activity in their simulation using the CSM. Mickley et al.  found that surface cyclone activity decreased by approximately 10–20% in a future simulation using the Goddard Institute of Space Studies (GISS) model. The decrease in synoptic activity is in general agreement with a number of observational studies over the northern midlatitudes and North America [Zishka and Smith, 1980; Agee, 1991; McCabe et al., 2001].
 These changes in circulation are also reflected in CO and O3. The 2–6 day band passed standard deviation of CO (ozone) decreases by 10% (7.5%) in the future simulation. In addition, when we correlate the lagged time series of CO (or ozone) with itself, the autocorrelation times of CO and ozone increase. For example, at 2 days the average ozone autocorrelation over the study region is 0.25 in a future climate and 0.13 in the control simulation (not shown). Mickley et al.  also find the autocorrelation times increase in the future. Finally, we have calculated the duration time of pollution and clean events in the region defined above. These events are defined respectively as the time for which the concentration of CO is greater than (less than) one standard deviation above (below) its mean at a grid point (Figure 9). For both types of events and in both climates the number of events which last longer than a set period decreases exponentially with an approximate 2 day timescale. In both climates it is rare for either type of event to last longer than 10 days at any one grid point. There is a clear bias for longer-lasting events in the future. Hogrefe et al.  reports a similar result. This tendency is somewhat more pronounced for pollutant events, but clean events also tend to last longer for times greater than 2 days. This suggests the effects of a more sluggish circulation in the future. These effects are manifest both for pollutant events and clean air events. We further discuss these effects on the ozone distribution in the following section.
 The height of the planetary boundary layer has an immediate effect on the dilution of emitted species. We find no significant future differences in boundary layer height over the eastern third of the country and a decrease of 30–60 m in the west (not shown). Mickley et al.  find significant increases in the boundary layer height in their simulation of the future climate over the United States. Hogrefe et al.  also report significant increases in mixing layer depths over the eastern half of the United States in a future climate. Both Hogrefe et al.  and Mickley et al.  use the meteorology from the GISS climate model for input, a model which simulate greater vertical mixing away from the surface in a future climate [Rind et al., 2001]. In the recent Community Climate Systems Model version 3 (CCSM3) future simulations (see http://www.ccsm.ucar.edu/experiments/ccsm3.0/#plots) changes in the boundary layer height over the United States depend on the resolution of the simulation. Because of the fact that the change in boundary layer height is not consistent between models and resolutions, we conclude the future change in boundary layer height over the United States remains uncertain.
6. Changes in U.S. Pollutants: 1990–2090
 Changes in the future climate act to change future concentrations of PAN, NOx, OH and H2O2 (Figure 10). The pattern of increase is a complicated function of the changes in water vapor, temperature, insolation, emissions and chemistry. For all species the changes are significant over most of the continental United States. Over most of the continental United States PAN decreases and NOx increases, acting to increase future ozone production. The largest decrease in PAN occurs over the northeast United States, where the emissions of NOx are particularly high. The largest increase in NOx occurs to the west of this region, somewhat south of Chicago. As discussed above, the future increase in temperature has an important affect on the PAN to NOx ratio. Following the redistribution of lightning NOx recommended by Pickering et al.  the future increase in lightning NOx also acts to increase NOx at the surface. Overall, however, the vertical distribution of lightning NOx remains uncertain. In addition, the overall increase in surface lightning NOx is small in comparison to anthropogenic emissions. In our formulation ozone produced from lightning NOx is counted as background ozone, but as we show below (see Figure 12) the background ozone concentration decreases over the continental United States in the future simulation.
 In the future simulation global tropospheric OH increases on a mass weighted basis during JJA by nearly 3.5% with an almost 4.5% increase in the N.H. Our simulations also predict OH to increase dramatically over the central United States in the future (see Figure 10), where the water vapor increase is particularly large (Figure 4). The increase in OH decreases the lifetime of a number of emitted species. It can increase or decrease ozone production depending on the concentration of NOx (reactions (R2)–(R5)). H2O2 also increases by approximately the same percentage as the increase in OH, despite the fact that the production of H2O2 increases as the square of HO2. H2O2 is important in the aqueous phase oxidation of SO2, and thus has an important impact on aerosol formation. Table 2 summarizes the percentage change of a number of variables between the 2090s and 1990s when averaged over the eastern and western United States.
Table 2. Percentage Change of a Number of Variables Between the Future and Control Simulations Averaged Over All Model Grid Points Between 30 and 50°N Over the Eastern United States and Western United Statesa
Western United States
Eastern United States
The east-west division is assumed to be at 100°W.
Cloud water is integrated from the surface to 700 hPa.
Lightning NOx is integrated throughout the depth of the atmosphere.
 By tagging the U.S. NOx emissions, the associated U.S. ozone production is calculated. The difference in net ozone production between the 2090s and 1990s is shown in Figure 11. The differences are positive throughout most of the continental United States, except along the southern border. The maximum difference in ozone production coincides with the maximum difference in NOx concentration.
Figure 12 shows the simulated MDA8 ozone concentration over the United States during the 1990s (O3), the portion of this due to background ozone (O3B) and the portion due to ozone production from U.S. NOx emissions (O3US). For example, the concentration of O3B shown in Figure 12 represents the portion of the MDA8 ozone concentration which is composed of background ozone. The contribution of O3B ranges between 5 and 40% of ozone, and maximizes on the west coast. These concentrations represent the “background” levels of ozone over the United States during the summer months and include sources of ozone production from outside the United States, ozone produced from lightning NOx emissions, aircraft emissions, and ozone transported from the stratosphere. These levels are substantially lower than those reported by Fiore et al.  who predict a maximum background ozone concentration of 30 ppbv over the southwestern United States and 20 to 25 ppbv over the eastern United States. While we use a different technique to calculate background ozone than used by Fiore et al. , the main difference in the results is most likely in the definition of what composes background ozone. Fiore et al.  considered soil and biomass burning emissions to be part of the background (A. Fiore, personal communication, 2005); we considered these emissions to contribute to O3US. MOZART predicts a significant 2–6 ppbv increase in O3US in the 2090s over much of the United States and a 0–2 ppbv decrease in O3B, also significant over much of the country except for parts of Texas and the northeast. The negative change in O3B in the western United States tends to cancel out the positive change in O3US. As a result these simulations suggest the change in ozone due to climate change will be primarily centered over the eastern half of the country, where we predict O3 increases up to 5 ppbv. Table 3 summarizes the MDA8 changes in O3, O3US and O3B over the western and eastern United States. Hogrefe et al.  predict an ozone change of 5 ppbv in a future climate (their temperature changes by 5.8°C over the eastern half of the United States compared with up to 3°C in our simulations) when weighted over the location of the EPA ozone monitoring stations. Table 3 and Figure 12 show the importance of correctly taking into account the influence of climate on the chemical boundary conditions in a future run. This is particularly important over the western United States where the change in imported ozone is significant.
Table 3. U.S. MDA8 of O3, O3US and O3B for the Control Simulation and the Percentage Change Between the Future and Control Simulations (Future Minus Control Divided by Control) Averaged Over All Model Grid Points Between 30 and 50°N Over the Eastern United States and Western United Statesa
Western United States
Eastern United States
The east-west division is assumed to be at 100°W.
Figure 13 gives the number of days the MDA8 ozone concentration is greater than 80 ppbv in the 2090s versus the 1990s. In part because of the cancellation between O3US and O3B, the change is only large and significant over the northern and eastern part of the country. The maximum increase is approximately 12 days. To put this number in perspective New England exceeded this standard by an average of 29 days per year from 1993 to 2003 (U.S. Environmental Protection Agency, Region 1: New England, http://www.epa.gov/region01/eco/ozone/). Hogrefe et al.  find a comparable change in their future simulations. In the future the number of high-ozone days decreases over the southeastern and south central United States. In this region there is an increased inflow from the Gulf of Mexico in the future climate (not shown).
 In the future simulations of Mickley et al.  the upper 10% of cases have increased concentrations of a CO-like tracer, where the loss of this tracer is given by reaction with the present-day monthly averaged OH field. The cumulative frequency distributions of carbon monoxide (Figure 14a) in our simulations show very little difference between the future and control simulations. The cumulative frequency distribution of ozone shows increases at all probabilities (Figure 14b). This is consistent with the results given in the regional study of Hogrefe et al. . The minimum ozone concentrations are definitely higher in the future climate; the maximum concentrations also tend to be slightly higher, except in the upper 4% of the cases.
 The daily mean ozone concentration increases by 1.7 ppbv between the control and future simulations in the study region examined in Figure 14. In the lowest 16% of the cumulative distribution ozone increases by approximately 0.4 ppbv in the future simulation relative to the control simulation, or on average 0.025 ppbv per percent; in the highest 16% of the points the ozone increase is approximately 0.2 ppbv, or on average 0.0125 ppbv per percent; in the middle 68% of the distribution the increase is 1.1 ppbv, or approximately 0.016 ppbv per percent. Most of the future ozone increase is the result of the accumulation of the relatively small increases in the middle range (i.e., in the 16 to 84% range) of the distribution. However, the steepest increases occur on the low end of the distribution. It is possible that in the future the weaker synoptic activity is less effective than at present in flushing out the boundary layer with clean air, but we have not verified this. At any rate our findings do not support the conclusion that increased future ozone is due to an increase in the intensity of high-pollution events.
7. Import and Export of Ozone to the United States
 In this section we examine changes in the export and import of ozone to the United States in a future climate. Figure 15 shows significant differences between the future and control simulations in the latitudinal profile of ozone at 130°W, just west of the continental United States. Because of the predominantly westerly winds north of 30°N, differences in Figure 15 will be reflected in the import of ozone to the United States. At all latitudes large and significant ozone decreases occur below approximately 600 hPa. Ozone increases are evident above 600 hPa near 15°N and 55°N.
 Some of the differences in Figure 15 can be traced to differences in the distribution of “stratospheric” ozone. At 130°W the change in future O3S mirrors the ozone change given in Figure 15: O3S increases above 600 hPa near 15°N and 55°N, and decreases elsewhere. The regions where O3S decreases can be explained by future decreases in the tropospheric lifetime of ozone. The future increase in O3S in the vicinity of the subtropical jet has been noted above (section 3). The increase above 600 hPa and 55°N is a local feature: in the zonal average both O3S and ozone decrease in this region. This local response can be attributed to slight changes in the long-wave pattern in the upper troposphere in the future climate.
 In summary while some of the changes in the import of ozone to the west coast of the United States can be attributed to increases in the input of stratospheric ozone, others must be attributed to changes in tropospheric chemistry. Consistent with Figure 1, and argued elsewhere, the lifetime of ozone is expected to decrease in the future in remote regions. This is supported by additional simulations (not shown) where we tag ozone produced from Asian sources. We find that the net impact of the Asian sources of ozone on the United States decreases in the future because of the greater loss of ozone over the ocean basins. In addition, some of the changes in the import of ozone along the west coast of the United States can also be attributed to changes in transport. Figure 16 shows the differences at 130°W of a tracer of Asian CO emissions. This tracer has a constant loss rate of 1/40 days−1, so that differences in this trace species between the 2090s and 1990s can only be attributable to transport. Figure 16 shows a decrease in the transport of this species to 130°W in the middle and lower troposphere, and an increase in the upper troposphere. These changes are largely statistically significant. However, we cannot unequivocally attribute these changes to differences in climate. They may simply be due to transport differences between the simulation periods.
Figure 17 gives the latitude height cross section of O3US at 70°W, off the eastern coast of North America. The two outflow maxima in Figure 17a, one near 300 hPa and 35°N, and one near the surface and 40°N correspond very closely to those obtained for North American CO in GEOS-CHEM simulations [Li et al., 2006]. Li et al.  argue that the upper level maximum is associated with deep convection of ozone precursors. Figure 17b shows future changes in O3US at 70°W. The increases in the upper tropospheric export of O3US eventually impacts the upper troposphere over southern Europe. Future O3US decreases at the surface at 70°W. We believe this reflects the average response to different flow regimes. When O3US is transported to 70°W from the United States mainland (e.g., to the west of cold fronts) its future concentration should increase; when O3US is transported to 70°W from the remote marine boundary layer (e.g., to the east of cold fronts) its future concentration should decrease. It is likely that the latter event predominates and accounts for the overall decrease in O3US at 70°W (Figure 17b). At the surface, future increases in O3US are mostly confined over land (see Figure 12).
8. Discussion and Conclusions
 In this study we have run the MOZART-2 chemical transport model for the period 1990–2000 and from 2090 to 2100, where MOZART was driven using meteorology from the CSM forced with the IPCC SRES A1 scenario. The emissions of NOx from United States were tagged in these runs so as to distinguish between background ozone and ozone produced from United States emissions. Emissions of all species were kept constant during both periods to examine the impact of climate change on ozone.
 Significant differences were found in the chemistry between the two simulations. We note that the CSM is known to have a low sensitivity to changes in CO2 [Cubasch et al., 2001], so that the ozone response to climate change might also be expected to be on the low end of the range of model predictions. The chemical response to future climate change occurs on local, regional and global scales. We have identified a number of factors which affect the future chemical response on both the regional and global scales. It is outside the scope of this study to complete a sensitivity analysis with respect to the relative importance of the various factors.
 We argue that in general the impact of climate change alone (i.e., assuming emissions remain constant) on future ozone levels will be to decrease surface ozone in remote regions and increase it in polluted regions. When examined on the large scale (e.g., in a zonal average) surface ozone concentrations decrease in our future simulation. This has also been found to be the case in a number of other studies [Brasseur et al., 1998; Grewe et al., 1999, 2001; Johnson et al., 1999, 2001; Collins et al., 2003] and is attributable to increased water vapor leading to enhanced ozone destruction. As a result the ozone lifetime generally decreases in the troposphere, ozone plumes generally decay faster and the background levels of ozone decrease. This is a robust consequence of climate change as the probability of increased tropospheric water vapor is virtually certain [Cubasch et al., 2001].
 However, distinct from the general decrease in ozone concentrations, surface concentrations of ozone increase in a future climate in regions with high NOx emissions (northeast United States, Africa and South America) including those with biomass burning. Thus future climate change is likely to exacerbate the air quality in regions with high NOx emissions. We have identified two processes which are expected to occur on a global scale in a future climate and which act to increase ozone production in locations with large NOx emissions: an increase in temperature and an increase in water vapor. An increase in temperature and water vapor are robust attributes of climate change, occurring across a broad range of models and model resolutions [Cubasch et al., 2001]. The results of Liu and Trainer  imply an increase in water vapor will decrease ozone over regions with low NOx and increase ozone in regions with high NOx. Temperature increases should also most profoundly affect regions of high emissions, where the increases can significantly alter the ratio of NOx to PAN.
 On a regional scale, the effects of climate change on ozone may be augmented or mitigated by more local processes. In particular we have examined the effect of climate on a number of chemically relevant processes over the continental United States including changes in: boundary layer height, precipitation, circulation, transport, clouds, and synoptic activity. We discuss these processes in more detail below. Significant differences occur between the future and control simulations in some of these processes. However, we cannot be certain that these changes are a robust feature of climate change. First, the changes may be simply a feature of normal interdecadal variability irrespective of climate. Secondly, there is generally limited confidence in our ability to simulate regional impacts of climate change, in particular for surface quantities [McAvaney et al., 2001]. More study will be needed to evaluate the robustness of our predictions.
 Despite these qualifications the decrease in synoptic activity over the United States appears to be a fairly robust feature of climate change. We find a decrease in synoptic activity of approximately 10% during the summer months over a study region. This decrease has been reported in the CSM [Dai et al., 2001] and in other models [see Mickley et al., 2004]. It is consistent with observational trends of reduced baroclinicity during the summer months [Zishka and Smith, 1980; Agee, 1991; McCabe et al., 2001]. However, the effect of these changes on ozone levels is not well understood and may be offset by changes in convection or boundary layer mixing. We do find that the autocorrelation timescales of O3 and CO increase in a future climate and that high-pollution episodes tend to last longer in agreement with Hogrefe et al. . However, to offset this we find clean episodes also tend to last longer. The role of these circulation changes in determining future ozone is not well understood. Mickley et al.  concludes that the reduced cyclone frequency in a future climate will lead to an increase in the severity of summertime pollution episodes due to an increase in the intensity of pollution events at the high end of probability distribution. We find little evidence of this in the present study. Here we find that the future ozone increase occurs throughout its cumulative probability distribution, although it is most marked at the high and low ends of the distribution.
 We find a significant decrease in low-level cloud water over much of the continental United States consistent with the observational study of Del Genio and Wolf . As a result surface photolysis rates increase leading to an increase in surface ozone. However, clouds and their change under a future climate must be regarded as highly uncertain [Stocker et al., 2001]. We find no significant change in precipitation between the future and control simulations. Giorgi  shows that future precipitation change over the United States in summer is of inconsistent sign between models. Also, in future climate simulations different resolutions of the CCM3 do not give a consistent increase in short-wave radiation at the surface of the continental United States (see http://www.ccsm.ucar.edu/experiments/ccsm3.0/#plots).
 We also find a significant increase in lightning NOx production in the future simulation in agreement with Price and Rind . However, not only is lightning NOx subject to the uncertainty in the cloud parameterization, but the basic parameters determining the distribution and source of lightning NOx remain highly uncertain [e.g., see Prather et al., 2001].
 We find no significant future change in the height of the planetary boundary layer over the eastern United States, although other studies report a significant increase in planetary boundary layer height [e.g., Hogrefe et al., 2004]. We conclude additional work is necessary before we understand how climate change impacts the planetary boundary layer, transport within the boundary layer and transport out of the boundary layer. The simulation of boundary layers remains a persistent difficulty in global simulations with sensitivity to numerics and vertical resolution [Stocker et al., 2001].
 We do find a significant difference in the transport of Asian emissions to the west coast of the United States with the future transport decreasing at low levels and increasing at higher levels. Likewise in the future simulation we find differences in the circulation over the United States with increased southerly flow through the midsection of the country. This increased flow affects the surface distribution of water vapor and temperature (Figure 4), and is likely responsible for the pattern of change of future lightning NOx production (Figure 7) and ozone (Figure 12). Again more research is needed to understand how robust these future circulation changes are.
 The background concentration of ozone was explicitly computed within the model by tagging the U.S. emissions of NOx and calculating the resultant ozone production. In the control simulation we find the maximum background level of ozone is 20 ppbv throughout the continental United States, with the highest concentrations on the west coast. We find the background ozone decreases on the order of 10% between the future and control simulations. Overall, the change in background ozone can be viewed as a competition between increased ozone production over high-emission regions combined with a shorter ozone lifetime in travel across remote regions. The net effect is a decrease in the background level of ozone over the United States. This study clearly shows that when taking into account changes in U.S. pollutant levels in a future climate changes in the background levels of ozone and other pollutants must be accounted for.
 Over the western United States decreases in the background ozone largely cancel increases in U.S. ozone production and little change in net ozone is apparent. Over the eastern and central United States we predict ozone levels will increase by up to 5 ppbv. This results in an increase of up to 12 days in which the 8-hour ozone concentration exceeds 80 ppbv.
 This study only considered changes in U.S. air quality directly attributable to changes in climate. We predict that despite any changes to emissions, air quality is likely to deteriorate in a future climate. We have argued that this is likely to be a robust result, regardless of the uncertainties in the regional impact of climate change over the United States. A number of factors may exacerbate this increase. Warmer temperatures are likely to increase demand for energy during the summer months, increasing emission levels. Regardless of U.S. anthropogenic emissions, biogenic emissions are likely to increase in a future climate resulting in increased ozone levels [Sanderson et al., 2003]. In addition, with the industrialization of eastern Asia it is likely that emissions over Asia will increase. The impact of future climate change is likely to augment this increase because if the NOx emissions increase sufficiently, the impact of climate change is likely to further increase ozone levels. While increased water vapor in a future climate should mitigate the effect of these emissions over the United States, a substantial increase in Asian emissions may still be expected to impact the United States.
 We would like to thank Denise Mauzerall for providing us with the EPA AIRS ozone data; Louisa Emmons, Jean-Francois Lamarque, Sasha Madronich, Natalie Mahowald, Brian Ridley, and two anonymous reviewers for their helpful suggestions and comments; Jean-Francois Lamarque for calculating the average isoprene emissions; and Natalie Mahowald for providing the meteorological fields. Most of this work was performed during a visit by one of us (K.M.) to NCAR, which was made possible largely through a grant from Ministry of Education, Culture, Sports, Science and Technology of Japan and the NCAR visiting scientist program. NCAR is operated by the University Corporation for Atmospheric Research under sponsorship of the National Science Foundation.