References

Uncertainties in calculated impacts of climate forecasts on future regional air quality are investigated us- ing downscaled MM5 meteorological fields from the NASA GISS and MIT IGSM global models and the CMAQ model in 2050 in the continental US. Differences between three fu- ture scenarios: high-extreme, low-extreme and base case, are used for quantifying effects of climate uncertainty on re- gional air quality. GISS, with the IPCC A1B scenario, is used for the base case simulations. IGSM results, in the form of probabilistic distributions, are used to perturb the base case climate to provide the high- and low-extreme scenarios. Impacts of the extreme climate scenarios on concentrations of summertime fourth-highest daily maximum 8-h average ozone are predicted to be up to 10 ppbV (about one-seventh of the current US ozone standard of 75 ppbV) in urban areas of the Northeast, Midwest and Texas due to impacts of mete- orological changes, especially temperature and humidity, on the photochemistry of tropospheric ozone formation and in- creases in biogenic VOC emissions, though the differences in average peak ozone concentrations are about 1-2 ppbV on a regional basis. Differences between the extreme and base scenarios in annualized PM2.5 levels are very location depen- dent and predicted to range between 1.0 and +1.5µg m 3 . Future annualized PM2.5 is less sensitive to the extreme cli- mate scenarios than summertime peak ozone since precipita- tion scavenging is only slightly affected by the extreme cli- mate scenarios examined. Relative abundances of biogenic VOC and anthropogenic NOx lead to the areas that are most responsive to climate change. Overall, planned controls for decreasing regional ozone and PM2.5 levels will continue to be effective in the future under the extreme climate scenar- ios. However, the impact of climate uncertainties may be substantial in some urban areas and should be included in assessing future regional air quality and emission control re- quirements.


Introduction
Impacts of future climate change on regional air quality have been investigated for different regions, years and emission scenarios. Due to uncertainties inherent in climate forecasts, the Intergovernmental Panel on Climate Change (IPCC) investigated multiple scenarios associated with different projections of future anthropogenic emissions of greenhouse gases (GHGs) as part of IPCC's Special Report on Emissions Scenarios (SRES) (IPCC, 2007;Nakićenović, 2000). Hogrefe et al. (2004) predict an increase in spatially averaged summertime daily maximum 8-h O 3 concentrations of 4.2 ppbV in the 2050s over the eastern US based on the IPCC A2 scenario and assuming anthropogenic precursor emissions and boundary conditions to remain constant. Murazaki and Hess (2006) suggest an increase of up to 12 additional days in the northeast of the continental US each year exceeding daily maximum 8-h average ozone concentrations of 80 ppbV in the decade 2090s compared with 1990s, assuming that future precursor emissions remain at 1990 levels and GHG emissions follow A1 scenario. Racherla and Adams (2006) predict an increase up to 5 ppbV in ozone concentrations over the eastern US while the global burdens and the lifetimes of all the fine particulate matter species decreased by 2 to 18% between 1900s and 2050s assuming climate will follows the IPCC A2 scenario and anthropogenic Published by Copernicus Publications on behalf of the European Geosciences Union.

866
K.-J. Liao et al.: Climate uncertainty and air quality emissions remain constant. Sanderson et al. (2003) predict a 10-20 ppbV increase in ozone concentrations during July due to a combined effect of changes in vegetation and prescribed IPCC IS92a CO 2 emissions in 2090s compared with 1990s over the majority of the US.
The objective of this study is to investigate how uncertainties inherent in climate change forecasts impact regional air quality predictions over the continental US using multiple climate futures. Given that model inputs (e.g., regional meteorology and precursor emissions) and parameterizations/assumptions lead to uncertainties in regional downscaling of future climate and air quality modeling (which have been presented elsewhere, e.g., Bergin et al., 1998;Russell and Dennis, 2000;Hanna et al., 2001;Hanna et al., 2005;Gustafson and Leung, 2007), the purpose of this study is not to specifically forecast future air quality but to quantify the impact of climate uncertainties on regional air quality forecasts, particularly focusing on ground-level ozone and PM 2.5 (particulate matter with an aerodynamic diameter less than 2.5 µm) due to their adverse health-related effects (Bernard et al., 2001;Galizia and Kinney, 1999;Johnson and Graham, 2005). Of particular interest are the uncertainties associated with the "climate penalty" (increases in levels of air pollutants caused by climate change, Mickley et al., 2004) and investigating if uncertainties in climate predictions suggest alternative emission control strategies.

Downscaling of global climate models to a meso-scale meteorological model
The meso-scale meteorological model, MM5 (The Fifth-Generation NCAR/Penn State Mesoscale Model) (Grell et al., 1994;Seaman, 2000), is used to downscale outputs from the NASA Goddard Institute of Space Studies (GISS) global climate model (GCM) (Rind et al., 1999) to regional scale for studying effects of climate on regional air quality in year 2050. 2050 is chosen for this study as a compromise between non-trivial climate modification and a reasonable horizon for regional air quality planning. The GISS-MM5 climate fields, following the IPCC A1B scenario, are used as base case meteorological fields in this study. Details in GISS global climate simulation and downscaling of GISS global climate to meso-scale climate are described by Mickley et al. (2004) and Leung and Gustafson (2005), respectively. The IPCC A1B emission scenario assumes a future world of very rapid economic growth with a balanced case between fossil and non-fossil energy sources and projects mid-level increases in greenhouse gas emissions and temperatures (Nakićenović, 2000). As such, the IPCC A1B is used for the base case GISS-MM5 simulations within multiple IPCC scenarios. For assessing uncertainties in climate projections and their associated effects on regional air quality, it is useful to investi-gate uncertainties in individual, but covering, climate variables (e.g., temperature, absolute humidity, etc.) in terms of their probabilistic distributions instead of qualitative assessments. In this study, climate fields from MIT's Integrated Global System Model (IGSM) Reilly et al., 1999) simulations, in the form of probabilistic distributions, are used to quantify uncertainties inherent in forecasts of future changes, and their associated effects on regional air quality. The IGSM is composed of: (a) the Emissions Prediction and Policy Analysis (EPPA) model, designed to project emissions of climate-relevant gases and the economic consequences of policies to limit them (Babiker et al., 2000), (b) the climate model, a 2-D zonally-averaged land-ocean resolving atmospheric model, coupled to an atmospheric chemistry model, (c) a 2-D ocean model consisting of a surface mixed layer with specified meridional heat transport, diffusion of temperature anomalies into the deep ocean, an ocean carbon component, and a thermodynamic sea-ice model (Sokolov and Stone, 1998;Holian et al., 2001;Wang et al., 1998), (d) the Terrestrial Ecosystem Model (TEM 4.1) (Melillo et al., 1993;Tian et al., 1999), designed to simulate carbon and nitrogen dynamics of terrestrial ecosystems, and (e) the Natural Emissions Model (NEM) that calculates natural terrestrial fluxes of CH 4 and N 2 O from soils and wetlands . The probabilistic distributions of changes in climate fields in 2050 were derived from a set of 1000 ensemble simulations (Webster et al., 2003). In configuring this ensemble of simulations, the model uncertainty is included by using a joint PDF of three climate model parameters, i.e., climate sensitivity, ocean heat uptake, and aerosol radiative forcing along with PDF of predicted anthropogenic emissions of major greenhouse gases which is calculated using Monte Carlo analysis of the EPPA model. The IGSM provides 2-D longitudinally and monthly averaged meteorological fields. For uncertainty analyses, extreme cases from probabilistic distributions of climate fields are of interest for policy-making. The use of the 2-D model allowed development of wider probabilistic distributions from which a wide variety of proposed policies and extreme future cases can be chosen . Temperature and absolute humidity fields from the GISS-MM5 climate are chosen for perturbations as they are strongly correlated with regional ozone and secondary PM 2.5 levels (Sillman and Samson, 1995;Wunderli and Gehrig, 1991;Wise and Comrie, 2005;Strader et al., 1999;Seinfeld and Pandis, 2006;Nenes et al., 1998). Climate fields used are associated with the 0.5th, 50th and 99.5th percentiles of temperature and absolute humidity from IGSM. The 50th percentile of both the meteorological parameters are adjusted to the GISS-MM5 by minimizing the discrepancies in temporal and spatial resolutions between the 50th IGSM and GISS-MM5 outputs and used to develop perturbation fields along with the GISS-MM5 based on the following processes: First, a three-dimensional time-dependent variable of the GISS-MM5 climate is decomposed into a spatially re-running MM5 with new initial and boundary conditions (i.e., M (y, x, z, t) INT 0.5% , M (y, x, z, t) INT 50% and M (y, x, z, t) INT 99.5% from step 2) in order to get conservative meteorological fields.
We keep the GISS-MM5 field (M (y, x, z, t) GISS−MM5 ) as the base case scenario in order to compare with current pollutant levels (note that new fields of the IGSM 50th percentile climate are not identical to the GISS-MM5 fields). The highand low-extreme fields are calculated as follows (Eq. 3a and b): It is recognized that using MM5 for downscaling the temperature and absolute humidity distributions from the IGSM outputs may not capture the full range of uncertainty in climate change, however, the new fields do capture the extreme cases from the probabilistic distributions of both the meteorological fields, and the other meteorological fields (e.g., precipitation, wind fields, etc.) are dynamically consistent and responsive to the perturbations in temperature and absolute humidity.

Emission and air quality modeling
The simulation domain in this study covers the continental US as well as parts of Canada and Mexico. For more detailed analysis, the continental U.S is divided into five regions -West, Plains, Midwest, Northeast and Southeast (Fig. 1 (Byun and Schere, 2006) for simulating impacts of climate uncertainties on regional air quality in 2050. Details of the projections of future precursor emissions and regional air quality modeling approach are given elsewhere (Tagaris et al., 2007;, and summarized here. The IPCC A1B emission scenario projects decreases in SO 2 , NO x and non-methane volatile organic compound (NMVOC) emissions from Organization for Economic Co-operation and Development (OECD) countries, including the US and Canada, in 2050s as compared with 1990s based on long-term energy-use and economic trends, which don't include currently planned precursor emission control regulations (Nakićenović, 2000). For future regional air quality modeling, projections in regional anthropogenic precursor emission changes are required in addition to national and global emission trends. In this study, projections of regional anthropogenic precursor emissions were developed for North America integrating currently planned emission controls (i.e., US Clean Air Interstate Rule (CAIR); Houyoux, 2004) and long-term economic and population growth, based on the A1B scenario. Although the same projected emission inventories are applied in the uncertainty simulations in 2050, simulated emissions of precursors of pollutants for the three climate scenarios are not identical since emissions (especially biogenic volatile organic compounds (VOCs)) respond to changes in meteorological fields (e.g., temperature, precipitation, etc.) (see Table S2 http://www.atmos-chem-phys.net/9/865/2009/ acp-9-865-2009-supplement.pdf).
The highest daily maximum 8-h average ozone (MDA8h O 3 ) levels, which are often associated with adverse health effects in epidemiologic studies and used for assessing attainment of the US National Ambient Air Quality Standards (NAAQS) for ozone (Bernard et al., 2001;Levy et al., 2001), consistently occur in summer. Three summer months (June, July and August) in the year-2050 are chosen as the target period for studying the impact of climate uncertainties on the average and 4th highest MDA8h O 3 (4th MDA8h O 3 ) concentrations. The 4th highest value is chosen as being more stably predicted by chemical transport models than the maximum in any location. For PM 2.5 , one month from each of the four seasons (i.e., January, April, July and October) in 2050 is chosen for studying the impact of climate uncertainties on annualized PM 2.5 levels because PM 2.5 has distinct seasonal variation and an annual healthbased standard (http://www.epa.gov/air/criteria.html, last access: 15 December 2008). Interannual variability of meteorology is a critical issue since only the year 2050 is chosen as the future episode examined in this study. The analysis for the interannual variability of climate fields has been presented by Tagaris et al. (2007): the results show that cumula-tive distribution function (CDF) and spatial distribution plots for temperature and absolute humidity are similar for the three consecutive future years (2049)(2050)(2051). The former paper provides information on interannual variability, and this paper looks at perturbations to the modeled base meteorology.

Meteorology
The annualized temperatures (average temperatures of January, April, July and October) of the 2050 base GISS-MM5 climate, based on the IPCC A1B scenario, are predicted to be 0.4-2.4 K warmer than 2001, depending on the region, whereas absolute humidity values are simulated to be approximately 0.5 g/Kg (9%)-0.7 g/Kg (14%) higher than 2001 (Table 1). Spatial distributions of annualized temperature and absolute humidity between the base case GISS-MM5 and IGSM 50th percentile climate are found to be similar (Fig. S1 http://www.atmos-chem-phys.net/9/865/2009/ acp-9-865-2009-supplement.pdf), although regional average values differ slightly since the IGSM 50th percentile data has been used for re-running MM5 after being adjusted to the base case GISS-MM5 (Table 1). On the other hand, annualized temperatures and absolute humidity of the two 2050 extreme scenarios are predicted to change approximately from −0.8 K (low-extreme) to +2.1 K (high-extreme) and −0.4 g/Kg (−7%) (low-extreme) to +1.1 g/Kg (+19%) (high-extreme), respectively, as compared with the 2050 base scenario on a regional basis (Table 1). Summer (JJA) temperatures and absolute humidity values are predicted to be higher for the 2050 base case than 2001 climate (Table 1). Differences between the high-extreme and base case scenarios are found to be larger than differences between the lowextreme and base scenarios for both temperature and absolute humidity (Figs. 2 and 3). This reflects that the probability density functions of predicted temperatures and absolute humidity are not normally distributed but have a long right-hand tail in the IGSM outputs (Table S1) (Webster et al., 2003). Annualized precipitation is found to be somewhat different for the three scenarios in 2050, with a 0.1 mm/day decrease in summer precipitation in the Plains for the highextreme scenario as compared with the base case (Fig. 4). The differences in precipitation between the 2050 base case and extreme scenarios are driven by the perturbations in temperatures and absolute humidity and based on the extreme percentiles of probabilistic distributions of temperature and absolute humidity changes derived from global modeling in 2050.

Emissions
Both sulfur dioxide (SO 2 ) and nitrogen oxides (NO x ) emissions are forecast to be 51% lower for the 2050 base case   Absolute Humidity (g/Kg)  (Tagaris et al., 2007;Woo et al., 2008). For the two extreme 2050 scenarios, annualized and summer-average SO 2 , NO x , NH 3 , PM 2.5 and anthropogenic VOC emissions are predicted to change slightly compared with the 2050 base  (Table S2 http://www.atmos-chem-phys.net/9/865/ 2009/acp-9-865-2009-supplement.pdf). However, predicted total VOC emissions vary significantly as biogenic VOC emissions are much more sensitive to temperature changes than other precursor emissions (Tables 1 and S2). Responses of VOC emissions to the extreme climate scenarios are also found to change spatially. The low-extreme scenario results in an approximately 0-17% decrease in total VOC (=anthro-pogenic+biogenic VOC) emissions compared with the 2050 base scenario. For the high-extreme scenario, higher biogenic VOC emissions cause an increase of up to about 22% in annualized and 29% in summer-average total VOC emissions compared with the base case in 2050 on a regional basis (Table 1). 3.3 Summer-average ozone and summertime fourthhighest daily maximum 8-h average ozone Summer-average ozone and daily maximum 8-h ozone concentrations are found to be slightly affected by the extreme climate scenarios in 2050 with typical changes of about 1-2 ppbV between the extreme and base case climate scenarios on a regional basis (Table 2). For the peak ozone levels, summertime (JJA) 4th MDA8h O 3 (4th MDA8h O 3 in the summer of 2050) concentrations for the high-extreme scenario are predicted to increase up to 10 ppbV as compared with the 2050 base case in urban areas of the Northeast, Midwest and Texas in the continental US (Fig. 5). Such differences are attributed to the changes in meteorological fields (e.g., temperature, humidity, etc.) and meteorologyinduce changes in precursor emissions (e.g., increases in biogenic VOC emissions) on the photochemistry of tropospheric ozone. Sensitivity analyses show that peak ground-level ozone levels and ambient temperatures are positively correlated with each other (Baertsch-Ritter et al., 2004;Menut, 2003;Dawson et al., 2007;Aw and Kleeman, 2003). Sillman and Samson (1995) found higher temperatures increase decomposition of peroxyacyl nitrates (PANs) and generate nitrogen dioxides (NO 2 ) during the daytime and hence cause higher peak ozone levels. Higher absolute humidity (water vapor concentration) increases hydroxyl radicals (OH), resulting in faster oxidation of VOCs, forming peroxy radicals (e.g., HO 2 , RO 2 ) which react with nitrogen oxides (NO) to form NO 2 (Seinfeld and Pandis, 2006). Even when changes in precursor emissions are not considered, concentrations of summertime (JJA) 4th MDA8h O 3 in urban areas are more sensitive to changes in temperatures and absolute humidity than the other areas due to their higher concentrations of PANs, VOC, CH 4 and CO, and also expected to find a greater simulated impact from the high-extreme scenario than the base case in 2050. When temperature-induced increases in  VOC emissions (especially biogenic VOC emissions) (Table 1 and Fig. 6) are considered, higher ozone levels are found in NO x -saturated (or VOC-sensitive) urban areas, e.g., Chicago and New York, and the effects of the high-extreme climate scenario are predicted to be more significant. Moreover, for five urban areas in the continental US (i.e., Atlanta, Chicago, Houston, New York and Los Angeles), our previous results show that concentrations of daily maximum 8-h average ozone are predicted to positively respond to VOC emissions on some days for the base case GISS-MM5 simulations in 2050, especially in Chicago and New York, al-though ozone formation is predicted to be more NO x -limited in 2050 than 2001 due to currently planned emission controls . On the other hand, increases in VOC emissions are also found to decrease 4th MDA8h O 3 formation in some areas of the West region where ozone formation is VOC-saturated. Furthermore, lower levels of predicted summer precipitation for the high-extreme scenario also lead to more ozone formation and an increase in the differences between the high-extreme and base case scenarios in the polluted urban areas (Fig. 4). The spatial distribution of changes in 4th MDA8h ozone between the 2050 extreme and base case climate scenarios does not follow that of temperature, absolute humidity and precipitation since groundlevel ozone formation is affected by the combined effects of different climate fields and local precursor emissions as well.
Differences in concentrations of summertime 4th MDA8h O 3 are predicted to be approximately +/−3 ppbV between the base case and low-extreme scenarios (Fig. 5). Concentrations of summertime 4th MDA8h O 3 are found to be less sensitive to the low-extreme climate scenario than the highextreme scenario due to smaller differences in meteorological fields between the base case and low-extreme scenario as well as non-linear responses of ozone concentrations to emission changes (Cohan et al., 2005). Tagaris et al. (2007) present an about 20% decrease in concentrations of summeraverage daily maximum 8-h ozone and less number of exceedance days of ozone concentration of 85 ppbV in five US cities between 2000-2002 and 2049-2051, mainly due to currently planned emission controls in the future. Here, a maximum change of 10 ppbV in 4th MDA8h O 3 (about oneseventh of the current NAAQS of ozone of 75 ppbV) is found in 2050 for the high-extreme climate scenario, which may significantly offset the effectiveness of currently planned emission reductions in urban areas with high concentrations of PANs, VOC, CH 4 and CO as well as VOC-sensitive ozone formation regimes.
3.4 Annualized PM 2.5 PM 2.5 levels are influenced by the changes between the climate scenarios in several ways. Higher temperatures favor semi-volatile compounds (e.g., secondary organic aerosols (SOAs) and ammonium nitrate (NH 4 NO 3 )) to remain in the gas phase. On the other hand, increases in temperatures and humidity result in higher emissions of SOA precursors and faster oxidation of SO 2 , NO x and VOCs, increasing formation of condensable compounds, such as sulfate, nitrate and semi-volatile organic species (SVOCs). Further, changes in precipitation can have a dramatic effect on frequency of washout and fine particle concentrations (Racherla and Adams, 2006). Overall, the net effects of different mechanisms of PM 2.5 production and loss result in a −1.0 to +1.5 µg m −3 difference in annualized PM 2.5 levels (average of daily PM 2.5 levels of January, April, July and October) between the extreme and base scenarios in 2050 (Fig. 5). Larger differences in PM 2.5 levels between the extreme and base scenarios are found in the Southeast and Midwest of the continental US due to higher PM 2.5 precursor emissions (e.g., anthropogenic SO 2 , NO x , VOC, etc.) in those regions (Table S2). The 1.0 µg m −3 decrease in annualized PM 2.5 levels in the Midwest for the high-extreme scenario is mainly due to lower sulfate and nitrate in the winter (January) compared with the base case (Table 3), while the 0.5 µg m −3 Table 3. Annualized and seasonal (January, April, July and October) total PM 2.5 levels and composition (sulfate, nitrate, secondary organic aerosol (SOA) and primary organic aerosol (POA)) for the three climate scenarios in 2050 for the five regions and US Unit: µg m −3 . increase in annualized PM 2.5 levels in the Southeast for the low-extreme scenario is due to higher nitrate in the winter (January) than the base case (Table 3). SOA is predicted to be influenced by changes in biogenic VOC emissions as well as modifications in the formation rates under the effects of the extreme climate scenarios. In CMAQ version 4.3, the SOA gas-particle partitioning model is based on the Secondary Organic Aerosol Model (SORGAM) (Schell et al., 2001), which doesn't account for SOA formation from isoprene. Some studies show that isoprene significantly contributes to SOA formation (Claeys et al., 2004), and SOA levels are predicted to be underestimated without including isoprene as a SOA precursor (Zhang et al., 2007;Morris et al., 2006;Pun and Seigneur, 2007). In this study, the changes in PM 2.5 levels attributed to the extreme climate scenarios are dominated by sulfate and nitrate PM 2.5 in most of the US regions since SOA formation is not fully captured in current regional air quality models (Table 3).
Impacts of climate uncertainties on PM 2.5 concentrations also show a seasonal trend. Monthly-average PM 2.5 concentrations are predicted to be lower in January but slightly higher in July for the high-extreme scenario compared with the 2050 base case (Table 3); this is mainly because tem-peratures change the partitioning of semi-volatile compounds between the gas-phase and particle-phase. Higher temperature and humidity increase sulfate aerosol formation due to faster gas-and aqueous-phase oxidation rates of SO 2 . Rae et al. (2007) have shown that increases in temperature and changes in oxidant concentrations are simulated to decrease 1% of Aitken-mode sulfate aerosols but increase of 9.2% of accumulation-mode sulfate in 2100 assuming climate and emission-induced oxidant levels will follow the IPCC SRES A2 scenario. Total sulfate concentrations are expected to increase by 6.8% in 2100 compared with 1990 (Rae et al., 2007). Effects of climate on nitrate are more complicated than sulfate due to high vapor pressure for particle-phase ammonium nitrate (Seinfeld and Pandis, 2006). Aw and Kleeman (2003) present that nitrate aerosol may slightly increase with cool temperature (<290 K) but decrease with hot temperature (>290 K) as temperature increases. The combined effects of changes in sulfate and nitrates show that the high-extreme climate scenario with associated increases in temperatures in January induces more nitrates and sulfates to be in the gas-phase lowering PM 2.5 concentrations. The seasonal trend is reversed in the low-extreme scenario (Table 3). Wise and Comrie (2005) show that, from a long-term statistical analysis, PM is not as weather-dependent as ozone in the southwestern US since low precipitation is found in the studying region. The results in this study also show that annualized PM 2.5 levels are not as sensitive as concentrations of summertime peak ozone with respect to the extreme climate scenarios examined since one of the main removal mechanisms of PM 2.5 , precipitation scavenging, is found to slightly affect annualized PM 2.5 levels between the extreme climate scenarios (Fig. 4). Our previous study shows that annual average PM 2.5 levels are predicted to decrease by about 23% (regionally varies from −9% to −32%) in 2050 as a combined effect of future climate change and CAIR emission controls (Tagaris et al., 2007). The results here imply that, on a regional basis, future emission controls will still be effective in decreasing annualized PM 2.5 levels with respect to the extreme climate scenarios if precipitation is only slightly affected.

Response of air quality to emission controls under extreme climate scenarios
In addition to simulating how the alternative extreme scenarios impact pollutant levels, we also investigate the responses of ozone and PM 2.5 levels to emission controls under the extreme climate scenarios. CMAQ, with the Decoupled Direct Method-3D (DDM-3D) (Dunker et al., 2002;Yang et al., 1997), is used to quantify sensitivities of ozone and PM 2.5 to precursor emissions. First-order sensitivities (S i,j ) of pollutant concentration i (C i ) (i.e. ozone and PM 2.5 ) to source emissions j (E j ) (i.e., anthropogenic VOC, anthropogenic NO x and total SO 2 emissions) are defined as (Yang et al., 1997): First-order sensitivities represent the locally linear responses of pollutant concentrations to emission changes and have the same units as the concentrations. Sensitivities of summertime 4th MDA8h O 3 to anthropogenic NO x emissions (S 4th MDA8h O 3 ,A NO x ) are predicted to slightly decrease for the low-extreme scenario but increase for the high-extreme scenario as compared with the base case in 2050. The differences are mainly attributed to the climate effects on biogenic VOC emissions and photochemistry. Biogenic VOC emissions are predicted to increase for the high-extreme scenario, especially in the Southeast region and west coast of the continental US, compared with the base case in the summer of 2050 (Fig. 6). Higher biogenic VOC emissions for the high-extreme attribute to a more NO x -limit environment for ozone formation and increase S 4th MDA8h O 3 ,A NO x . The effects of the extreme climate scenarios on sensitivities of summertime 4th MDA8h O 3 to anthropogenic VOC emissions (S 4th MDA8h O 3 ,A VOC ) are predicted to be small (Fig. 7). For the responses of PM 2.5 to emission changes under the extreme climate scenarios, sensitivities of annualized PM 2.5 to SO 2 emissions (S PM 2.5 ,SO 2 ) are predicted to slightly increase for the high-extreme scenario on a regional basis, because of higher temperature, humidity, decreased rainfall in some regions, and faster oxidation of precursors as compared with the base scenario. Higher temperatures for the high-extreme scenario favor particulate NH 4 NO 3 to dissociate to its gas phase precursors and cause slight decreases in sensitivities of annualized PM 2.5 concentrations to anthropogenic NO x emissions (S PM 2.5 ,A NO x ) (Fig. 7). Overall, on a regional basis, the effectiveness of NO x and SO 2 emission controls for reducing peak ozone and PM 2.5 levels changes little, though climate-driven increases in extreme ozone levels may require additional controls to reach applicable air quality standards.
Annualized monthly maximum daily PM 2.5 levels (average of monthly maximum daily PM 2.5 for January, April, July and October) and their sensitivities are also examined   (Table 4), since ammonium nitrate (NH 4 NO 3 ) formation is very responsive to changes in meteorology (particularly temperature, Stelson and Seinfeld, 1982). Responses of annualized monthly maximum daily PM 2.5 to SO 2 emissions are also predicted to be different under the extreme climate scenarios in Chicago due to high SO 2 emissions in the Midwest region of the US (Table 4). For four of the five cities examined, Los Angeles being the exception, reductions in SO 2 and anthropogenic NO x emissions are predicted to continue to be still effective for decreasing annualized monthly peak daily PM 2.5 levels under the extreme climate scenarios in 2050. While reductions in anthropogenic NO x and VOCs emissions are predicted to be effective for decreasing annualized monthly peak PM 2.5 levels in Los Angeles for the three climate scenarios in 2050. Overall, the results for the five cities show that, although the effectiveness of emission controls for decreasing peak PM 2.5 levels responds to the extreme climate scenarios, the directions of currently planned emission controls will not be significantly affected.

Conclusions
Uncertainties associated with simulations of the extreme climate scenarios are found to have a rather moderate effect on predicted emissions of VOC and concentrations of fourthhighest daily maximum 8-h average ozone in year 2050. Differences in concentrations of fourth-highest daily maximum 8-h average ozone between the extreme climate scenarios and base case are found up to 10 ppbV (about oneseventh of the current ozone standards of 75 ppbV) in some polluted urban areas due to the combined effects of higher temperature, absolute humidity and VOC emissions, though the change in summer-average ozone is minimal on a regional basis (∼1 ppbV). Differences between the extreme and base scenarios in annualized PM 2.5 levels are predicted to range between −1.0 and +1.5 µg m −3 . For the five US cities examined, sensitivities of annualized monthly maximum daily PM 2.5 to anthropogenic NO x emissions are predicted to influenced by the extreme climate scenarios since temperature changes significantly affect ammonium nitrate (NH 4 NO 3 ) formation. Responses of annualized monthly maximum daily PM 2.5 to SO 2 emissions change most in the Midwest due to locally high emissions. Overall, future annualized PM 2.5 is predicted to be less sensitive to the extreme climate scenarios than summertime fourth-highest daily maximum 8-h average ozone since precipitation scavenging is not significantly changed with the extreme climate scenarios where perturbations in temperatures and absolute humidity from the global model are used to drive changes in the regional meteorological modeling. Planned controls for decreasing regional ozone and PM 2.5 will continue to be effective in the future under the extreme climate scenarios. However, the impact of climate uncertainties may be substantial in some urban areas and should be included in assessing future regional air quality and emission control requirements.