This analysis provides an up-to-date assessment of long-term (1990–2010) rural ozone trends using all available data in the western (12 sites) and eastern (41 sites) USA. Rather than focus solely on average ozone values or air quality standard violations, we consider the full range of ozone values, reporting trends for the 5th, 50th and 95th percentiles. Domestic ozone precursor emissions decreased strongly during 1990–2010. Accordingly 83%, 66% and 20% of summertime eastern U.S. sites experienced statistically significant ozone decreases in the 95th, 50th and 5th percentiles, respectively. During spring 43% of the eastern sites have statistically significant ozone decreases for the 95th percentile with no sites showing a significant increase. At the 50th percentile there is little overall change in the eastern U.S. In contrast, only 17% (2 sites) and 8% (1 site) of summertime western U.S. sites have statistically significant ozone decreases in the 95th and 50th percentiles, respectively. During spring no western site has a significant decrease, while 50% have a significant median increase. This dichotomy in U.S. ozone trends is discussed in terms of changing anthropogenic and biomass burning emissions. Consideration is given to the concept that increasing baseline ozone flowing into the western U.S. is counteracting ozone reductions due to domestic emission reductions. An update to the springtime free tropospheric ozone trend above western North America shows that ozone has increased significantly from 1995 to 2011 at the rate of 0.41 ± 0.27 ppbv yr−1. Finally, the ozone changes are examined in relation to regional temperature trends.
 Within the United States, ozone has been recognized since the 1940s and 1950s as an air pollutant detrimental to human health and vegetation [National Research Council, 1991], and today ozone pollution is a widespread problem in many regions around the world resulting from both local and distant sources [Dentener et al., 2011]. The U.S. Clean Air Act of 1970 created a National Ambient Air Quality Standard (NAAQS) for ozone requiring regions that exceed the standard to implement ozone precursor emission reductions such that the NAAQS can be attained. Two decades later, a National Research Council review of ozone trends across the country concluded that “Despite the major regulatory and pollution-control programs of the past 20 years, efforts to attain the National Ambient Air Quality Standard for ozone largely have failed.” In those days the U. S. Environmental Protection Agency's (EPA) principal statistical measure of ozone concentrations was the composite nationwide average of the second highest 1-h daily maximum concentrations in a given year, which decreased by 14% between 1980 and 1989 but with large interannual variability. The National Research Council noted that it was difficult to ascertain progress in reducing ozone pollution when this primary metric was highly susceptible to meteorological variation and when only 10 years of data were available.
 As time went by, additional years of data allowed for more robust trend analyses, and different metrics were devised to give a better idea of ozone changes relevant for human and vegetative exposure [Lefohn and Shadwick, 1991; Fiore et al., 1998, and references therein]. Using summertime afternoon ozone data from 549 ozone monitoring sites across the U.S., Fiore et al.  found that no large region of the U.S. experienced a significant increase in ozone during 1980–1995. Significant ozone decrease were mainly confined to the New York City, Los Angeles and Chicago metropolitan regions, with decreases of the 90th percentile being more pronounced than for the medians. These urban ozone decreases appeared to be related to decreases in volatile organic compound (VOC) emissions as national NOx emissions were constant over the 15 year period.
 Up-to-date national ozone trend information is provided by the U.S. EPA on its website (http://www.epa.gov/airtrends/ozone.html). Using the annual 4th highest daily maximum 8-h average ozone mixing ratio (the metric used to determine exceedance of the NAAQS for ozone, currently set at 75 ppbv) from 247 monitoring sites across the country the EPA shows that ozone decreased from 97 ppbv during 1980–1984 to 75 ppbv during 2006–2010. The overall national trend is a 28% decrease from 1980 to 2010 and a 17% decrease from 1990 to 2010. These trends are based on data that were primarily collected in urban areas with most sites concentrated in the eastern USA or along the U.S. west coast. Using May–September average ozone values at 180 rural and urban sites across the country the EPA shows that 2001–2010 ozone decreases were similar for rural and urban sites. However, ozone decreases were greater in the east than in the west. While the average of all sites shows a nationwide decrease, not all sites have decreasing ozone. From 1980 to 2008 71% of U.S. sites showed statistically significant decreases in the 4th highest daily maximum 8-h average ozone value, and 2% had statistically significant increases; for 1994–2008 these values were 51% and 1%, respectively [Lefohn et al., 2010]. As described below, these results show that U.S. emission controls are reducing the frequency and magnitude of extreme ozone episodes.
 Because ozone is a secondary pollutant its production is closely related to the primary emissions of precursor trace gases. The EPA National Emissions Inventory estimates that for the period 1990–2010, total U.S. anthropogenic emissions declined by 49%, 58% and 44% for the ozone precursors NOx (=NO + NO2), CO and VOC, respectively (U.S. Environmental Protection Agency, National Emissions Inventory (NEI) air pollutant emissions trends data, http://www.epa.gov/ttnchie1/trends/, 2012, hereinafter referred to as EPA, online report, 2012). According to several recent photochemical modeling studies, these reductions in ozone precursors are responsible for the downward trend in ozone observed in the eastern U.S. and in southern California, at least for the median and extreme ozone values during the ozone season that lasts from May through September [Frost et al., 2006; Kim et al., 2006; Gilliland et al., 2008; U.S. Environmental Protection Agency, 2009; Butler et al., 2011; Hogrefe et al., 2011; Pozzoli et al., 2011].
 While emission reductions appear to be reducing the frequency of high ozone events, several studies have shown that mixing ratios of the lower ozone percentiles, such as the 5th, 10th or 20th percentiles are increasing across the country [Lin et al., 2000; Lefohn et al., 2010; Hogrefe et al., 2011]. Lin et al.  suggested that the increase in the lower ozone percentiles was due to an increase in the baseline ozone flowing into the U.S. (baseline ozone is defined as the observed ozone at a site when it is not influenced by recent, locally emitted or produced anthropogenic pollution [Dentener et al., 2011]). A modeling study by Jacob et al.  lent support to this hypothesis, which calculated that rising Asian anthropogenic emissions can increase the quantity of baseline ozone impacting the U.S., even offsetting ozone reductions caused by decreased domestic emissions. Many studies concurrent with, and subsequent to the findings of Jacob et al.  have provided ample in situ evidence that Asian pollution plumes reach the lower free troposphere and the surface of western North America [Jaffe et al., 1999; McKendry et al., 2001; Parrish et al., 2004a; Dentener et al., 2011 ].
 The possibility of significant changes in baseline ozone due to rising Asian emissions has the strongest implications for the western United States, the region of the country most likely to be impacted by Asian outflow [Reidmiller et al., 2009; Brown-Steiner and Hess, 2011; Lin et al., 2012a]. Several recent studies have shown that mean and median ozone values are increasing at some rural sites in the western USA, especially during spring. Analysis of springtime ozone measurements from rural sites on the U.S. west coast and Lassen Volcanic National Monument in elevated northeast California, as well as hydrocarbon measurements from field campaigns [Parrish et al., 2004b], showed that ozone increased from the 1980s through 2002, and that the photochemical environment of the eastern North Pacific Ocean marine boundary layer had also changed [Jaffe et al., 2003; Parrish et al., 2004b]. These studies suggested that the changes in the baseline air entering North America were related to increased anthropogenic emissions in Asia. Focusing on elevated National Park ozone monitoring sites across western North America, Jaffe and Ray  found that deseasonalized mean daytime ozone values increased at 7 of the 9 sites analyzed. Similarly, Oltmans et al.  reported that the positive trend at elevated Lassen Volcanic National Monument continued through 2006, the most recently available data at the time of their analysis. Returning to the west coast, Parrish et al.  combined surface ozone measurements from several U.S. coastal sites and filtered the data to characterize baseline air masses flowing onshore from the North Pacific Ocean marine boundary layer. The analysis showed statistically significant increases in mean ozone for winter, spring and summer from the 1980s through 2007. An increasing ozone trend has also been measured in the free troposphere above western North America during springtime (1984–2008), with the strongest rate of increase associated with the air masses with greater transport from the atmospheric boundary layer of south and east Asia [Cooper et al., 2010]. In response to these findings, several recent studies have focused on the transport mechanisms that bring enhanced ozone plumes from the free troposphere to the surface of the western USA with an emphasis on identifying plumes heavily influenced by Asian emissions [Weiss-Penzias et al., 2006; Parrish et al., 2010; Huang et al., 2010; Cooper et al., 2011; Pfister et al., 2011; Lin et al., 2012a].
 The studies described above have made great progress toward understanding and quantifying the impact of local and upwind emissions on U.S. surface ozone trends. In addition many recent studies report rural and urban ozone trends across the United States and provide insight into the reasons for the trends, or lack thereof [Chan, 2009; Chan and Vet, 2010; Lefohn et al., 2010; National Park Service, 2010; Hogrefe et al., 2011; Pozzoli et al., 2011; Sather and Cavender, 2012]. Individually, these studies have various foci, such as exceedances of the NAAQS for ozone, and cover a range of time periods through 2005 or 2008. However, the literature is lacking a current overview of U.S. ozone trends that is sensitive to the seasonal variability in ozone chemistry and transport processes, and that is aimed at understanding long-term (20-year) trends for low, mid- and high mixing ratios. The goals of this study are to provide an up-to-date nationwide survey of surface ozone trends, discuss the observed trends in light of regional and global changes in ozone precursor emissions and temperature, and to serve as a benchmark for modeling studies of U.S. ozone trends. Specifically, this study is aimed at understanding regional trends in ozone and therefore only analyzes rural measurements, which are less susceptible to ozone titration by fresh NO emissions, and only examines daytime measurements (11:00–16:00 local time) when the atmospheric boundary layer is well mixed and when nighttime surface deposition is not an issue. Rather than focusing on the broad “ozone season” of May–September as is generally done by studies interested in extreme ozone events, this study examines ozone trends separately for spring (March, April and May), and summer (June, July, August) which have contrasting transport patterns [Moody et al., 1998; Cooper et al., 2002] and photochemistry [Emmons et al., 2003]. While most of the emphasis is placed on spring and summer when exceedances of the NAAQS for ozone are most likely, winter ozone trends are also briefly reported to allow the contrast of ozone trends between seasons with strong and weak domestic ozone production. As noted by the studies above, ozone trends can differ for high, mid- and low ozone events, therefore trends are calculated for the 95th, 50th and 5th ozone percentiles.
 Due to the interannual variability of ozone [National Research Council, 1991; Parrish et al., 2012], determining robust ozone trends requires many years of data and this study focuses on the 21-year period of 1990–2010. The rural ozone measurements in this study were obtained from the U.S National Park Service (NPS) and The Clean Air Status and Trends Network (CASTNET). Data from one additional rural site, Whiteface Mountain Summit, New York were obtained from the University of Albany. Information on the locations of the 53 sites retained for this study can be found inTable 1 and Figure 1.
Table 1. Names and Locations of the 53 CASTNET (CN) and National Park Service Gaseous Pollutant Monitoring Program (NPS) Rural Ozone Monitoring Sites Used in This Studya
Network Site Abbreviation
NP, NM, and NS denote National Park, National Monument and National Seashore, respectively. Site abbreviations used in Figure 1 are listed in parentheses in the Network Site Abbreviation column.
Western United States
Grand Canyon NP
Grand Canyon NP
Sequoia/Kings Canyon NPs
Lassen Volcanic NP
Manzanita Lake Fire Station
SW of East entrance
Joshua Tree NP
Joshua Tree NP
Lost Horse Ranger Station
Rocky Mountain NP
West Glacier Horse Stables
Eastern United States
Cape Cod NS
Fox Bottom Area
Mount Greylock Summit
Blue Ridge Parkway
Route 191 (S Brevard Rd)
Whiteface Mountain Summit
U. of Albany/ NY DEC
Deer Creek State Park
Laurel Hill State Park
Penn State University
M.K. Goddard State Park
Kane Experimental Forest
Edgar Evins State Park
Great Smoky Mountains NP
Great Smoky Mountains NP
Cedar Creek Park
 The NPS Air Resources Division administers an extensive air monitoring program throughout the park system, designed to establish current air quality conditions and to assess long-term trends of air pollutants that affect park resources [National Park Service, 2010]. The data are also used to determine compliance with the NAAQS and to assess national and regional air pollution control policies. The NPS data used in this study were collected by the National Park Service Gaseous Pollutant Monitoring Program and downloaded from the NPS Gaseous Pollutant and Meteorological Data Access Page: http://ard-request.air-resource.com. Additional NPS data from the Brevard Rd., Saratoga, and Mt. Greylock sites were downloaded from the U.S. EPA Air Quality System Datamart website: http://www.epa.gov/ttn/airs/aqsdatamart/access/interface.htm.
 CASTNET is a national air quality monitoring network (administered and operated by EPA's Clean Air Markets Division) that collects data to assess trends in air quality, atmospheric deposition, and ecological effects due to changes in air pollutant emissions. The network began in 1991 with the incorporation of 50 sites from the National Dry Deposition Network, which had been in operation since 1987. CASTNET operates more than 80 regional sites throughout the contiguous United States, Alaska, and Canada, located in rural areas where urban influences are minimal. The CASTNET data used in this study were downloaded from: http://epa.gov/castnet/javaweb/index.html.
 A total of 53 sites were available for this analysis, all spanning the 1990–2010 period. While all of the sites are in rural areas, none can be considered to be truly remote from all anthropogenic emission sources. To minimize any local effects and to ensure that the measurements are as representative as possible of regional ozone, only midday measurements were analyzed, when the daytime atmospheric boundary layer is well-mixed. Twelve sites are located in the western U.S. and 41 sites in the eastern U.S. No sites were available in the Great Plains.
 Two of the sites experienced monitor relocations. From 1990 to 1996 measurements from Yellowstone National Park were made at the Lake Yellowstone site on the northwest shore of Lake Yellowstone. In 1996 the monitor was moved 1.5 km to the northwest to the Water Tank site, which is located 40 m above Lake Yellowstone. Because the monitor was moved simultaneous measurements at both sites were not available for comparison. While the local transport patterns are slightly different for the two sites due to the impact of the lake breeze at the lower elevation site, using data from the well-mixed midday period minimizes the differences as discussed byJaffe and Ray . Measurements from Joshua Tree National Park were made at the Lost Horse Ranger Station site from 1990 until September 1993, when the monitor was moved to the Black Rock site, 20 km to the northwest. The terrain is similar for the two sites but the new location is closer to urban areas. While the relocation of the monitor covers a substantial distance, air quality in Joshua Tree National Park is overwhelmingly dominated by air pollution transport from the Los Angeles Basin which regularly places the rural park in exceedance of the NAAQS for ozone [Sullivan et al., 2001; Rosenthal et al., 2003; Langford et al., 2010]. Therefore the ozone measured at Joshua Tree National Park is largely a reflection of emissions in the Los Angeles Basin and the variation in local emissions is minor in comparison.
Jaffe and Ray report ozone trends through 2004 at several National Park Sites. They noted that the inlet height for many National Park monitoring sites was increased from 3.5 m to 10 m above ground level in the mid-1990s. These sites included several used in the present study: Lassen Volcanic National Park, Rocky Mountain National Park, Yellowstone National Park and Glacier National Park.Jaffe and Ray  determined that the increase of the inlet height resulted in a statistically significant increase in ozone at night when the sites are typically influenced by temperature inversions, but they found no significant change in ozone during daytime conditions when surface warming has eroded the nighttime inversion.
 Ozone trends were calculated as follows. In order to minimize local effects and to ensure that the ozone measurements are representative of the well-mixed daytime atmospheric boundary layer, only hourly average data reported at 11, 12, 13, 14, 15 and 16 local time were considered. Trends were calculated separately for spring (March, April, May), summer (June, July, August), and winter (December, January, February). If a site had less than 50% data availability in any month in any season then that particular season was discarded. For each site and each season, data must be present for at least 18 out of 21 years during 1990–2010. Accordingly, any site with a reported trend in this analysis has greater than 85% data completeness over the 21 year study period, based on seasonal availability. All available daytime hourly measurements were used to compute the seasonal 5th, 50th and 95th ozone percentiles for each year. From this point forward, the 5th, 50th and 95th ozone percentiles for spring will be referred to as SpO305, SpO350 and SpO395, respectively. Likewise the summer percentiles will be referred to as SuO305, SuO350 and SuO395, and winter as WiO305, WiO350 and WiO395. The trend, or ozone rate of change (ppbv per year), over 1990–2010 was calculated separately for the 5th, 50th and 95th ozone percentiles with a straight line fit through the data using the least squares method of simple linear regression. The trends were calculated with a reference year of 2000. The p value indicates the statistical significance of the linear relationship, determined by first calculating R, the correlation coefficient between ozone and time. The null hypothesis that R2 = 0 (no linear relationship) was tested using the standard F-statistic (ratio of the mean square regression to the mean square residual). If the probability p associated with the F statistic was small (p ≤ 0.05), the null hypothesis was rejected with a confidence level ≥ 95%.
 To examine the impact of the Denver metropolitan region on ozone in the nearby Rocky Mountains, surface hourly ozone, CO and NO2 measurements from the Denver region were obtained from the U.S. EPA Air Quality System Data Mart website (http://www.epa.gov/ttn/airs/aqsdatamart/). Surface CO measurements from Niwot Ridge (40.05°N, 105.63°W, 3526 m a.s.l.) in the Rocky Mountains west of the Denver metropolitan region were measured by the NOAA Earth System Research Laboratory Carbon Cycle Group using weekly flask samples [Novelli et al., 2003].
 In addition to the surface ozone analysis, this study also provides an update to the free tropospheric springtime ozone trend above western North America that was originally calculated for 1984–2008 and reported by Cooper et al. . This update follows the same general methodology as Cooper et al.  and adds all available ozone measurement from 2009 to 2011, made between 3.0 and 8.0 km above sea level (a.s.l.) in the region 25°–55°N, 130°–90°W. The only difference in methodology between the present study and Cooper et al.  is that this analysis uses all available data between 3.0 and 8.0 km a.s.l. from all years during 1984–2011, whereas Cooper et al.  used a particle dispersion model to identify and remove measurements made in the lowermost stratosphere that occasionally extended below 8.0 km a.s.l. Using the original 1984–2008 data set of Cooper et al.  we compared the ozone rate of change with and without measurements made in the lowermost stratosphere. Due to the small percentage of data points in the lowermost stratosphere, the stratospheric measurements made little impact on the 50th ozone percentile values (seasonal medians were no more than 1 ppbv greater in the data set with stratospheric measurements) and made little difference in the rate of increase of the ozone 50th percentile.
 Ozone data for 2009–2011 were collected by (1) electrochemical concentration cell ozonesondes, accuracy: ±10% [Smit et al., 2007]; (2) an ozone lidar, accuracy: ±5–25% [McDermid et al., 2002]; (3) MOZAIC commercial aircraft, accuracy: ±(2 ppbv + 2%) [Thouret et al., 1998]; and (4) a variety of research aircraft flights with instrument accuracies that are generally better than ±5% or ±5 ppbv. The ozone measurements were made at the following locations: 1) NOAA ozonesondes from the monitoring sites of Boulder, CO and Trinidad Head, CA, as well as the IONS-2010 California sites of Shasta, Pt. Reyes, San Nicolas Island and Joshua Tree National Park [Cooper et al., 2011]; 2) Environment Canada ozonesondes from the monitoring sites of Kelowna, Bratt's Lake and Edmonton; 3) MOZAIC commercial aircraft profiles at Portland, OR and Calgary, Alberta; 4) lidar profiles from the NASA JPL Table Mountain facility, CA; 5) the NASA WB-57 aircraft profiles from Houston, TX during the spring 2011 MACPEX experiment; and 6) NOAA Global Monitoring Division aircraft profiles from Briggsdale, CO, Estevan Point, BC, Trinidad head, CA, Beaver Crossing, NE, West Brook, IA, and Southern Great Plains, OK. The total number of profiles from all platforms for 2009, 2010 and 2011 was 99, 186 and 100, respectively.
 Several gridded data sets were used in the analysis:
 1) Column NO2trends were calculated for several regions within midlatitude North America using monthly gridded data from the polar-orbiting GOME and SCIAMACHY sensors, produced and made freely available by the Tropospheric Emission Monitoring Internet Service in The Netherlands (www.temis.nl). These products are based on the methodology of Boersma et al.  and Richter et al. . Data were available for 1996–2011 at 0.25° × 0.25° horizontal resolution.
 2) The EDGARv4.1 global anthropogenic NOx emission inventory for 2005 at 0.1° × 0.1° resolution was provided by European Commission, Joint Research Centre (JRC)/Netherlands Environmental Assessment Agency (PBL): Emission Database for Global Atmospheric Research (EDGAR), release version 4.1 http://edgar.jrc.ec.europa.eu, 2010.
 3) Gridded Population of the World, Version 3 (GPWv3) data at 2.5 min horizontal resolution for 1990 and 2010 were produced and made available by the Center for International Earth Science Information Network (CIESIN), Columbia University; and Centro Internacional de Agricultura Tropical (CIAT) (Gridded Population of the World, Version 3 (GPWv3). Palisades, NY: Socioeconomic Data and Applications Center (SEDAC), Columbia University, http://sedac.ciesin.columbia.edu/gpw, 2005, hereinafter referred to as CIESIN, online report, 2005).
 4) Interannual variability of NOx emissions from wildfires was calculated for several regions within midlatitude North America using monthly gridded data from the Global Fire Emissions Database (GFED3) [van der Werf et al., 2010]. Data were available for 1997–2010 at 0.5° × 0.5° horizontal resolution.
3.1. Trends in Population, Emissions, and Free Tropospheric Baseline Ozone
 Before presenting the ozone trend analysis, and to facilitate the discussion of the results, this section describes the changes in population, ozone precursor emissions, and free tropospheric baseline ozone across the United States, all of which affect surface ozone levels. During 1990–2010 the U.S. experienced a broad increase and geographical shift in its human population. Overall, the population of the contiguous U.S., southern Canada and northern Mexico increased from 287 to 352 million, a 22% increase (Figure 2) that also shifted southwards and westward, with the greatest rate of increase in the west at 37%. Even faster rates of increase occurred in several western sub-regions, with the population of central Colorado and southeast Wyoming increasing by 44%, and the population of northwest Arizona and Las Vegas increasing by 95% (Figure 2).
 Despite the increase in population, ozone precursor emissions in the eastern and western U.S. have decreased. As mentioned in the Introduction, the EPA estimates that total U.S. anthropogenic emissions declined by 49%, 58% and 44% for NOx, CO and VOC, respectively, during 1990–2010. However, the changes have not been smooth with most of the reductions occurring after 1996, and there can be strong seasonal variations, for example power plant NOx emissions in the eastern U.S. are much less during May–September in an effort to reduce ozone during the typical ozone season [Butler et al., 2011]. An independent and up-to-date assessment of regional ozone precursor changes can be gathered from the satellite tropospheric column NO2retrievals made freely available by the Tropospheric Emission Monitoring Internet Service in The Netherlands. This gridded product is ideal for the purposes of this paper because the retrievals are available (1996-present) for much of the study's time period, and because ozone production in rural areas of the U.S. is typically NOx limited in spring and summer [McKeen et al., 1991; Fiore et al., 2009]. Even in the western USA where biogenic VOC emissions are weakest, ozone is much more sensitive to changes in anthropogenic emissions than biogenic [Fiore et al., 2011]. Therefore, for the remainder of this study, NO2 (or NOx) will be used as the primary indicator of ozone precursor emission distributions and trends.
Figure 3a shows the change in tropospheric column NO2 above the contiguous U.S., southern Canada and northern Mexico during spring and summer (column amounts in spring are greater than summer due to the longer NO2 lifetime in spring). Changes in column NO2 do not exactly match changes in emissions due to the change in NO2 lifetime that occurs with changes in ambient NO2 concentrations [Lamsal et al., 2011]. From 1996 to 2011 the total mass of NO2 above the U.S. decreased by 41% in spring and 33% in summer, in general agreement with the 49% emissions decrease estimated by the EPA for the longer period of 1990–2010. Column NO2 declines are also observed within the large regions of the northeastern, southeastern and western U.S. in spring and summer, all statistically significant based on linear regression analysis at the 95% confidence level (Figure 3a). Small western regions that cover the Colorado/Wyoming Front Range, northwest Arizona/Las Vegas, and Yellowstone National Park and surroundings (Figure 3b) also show declining column NO2. However, trends are only statistically significant (based on linear regression analysis at the 95% confidence level) for the northwest Arizona/Las Vegas region.
 The column NO2 retrievals detect NO2 from all sources including wildfires. Due to the sporadic occurrence of wildfires, which may be missed by the polar orbiting satellites that measure column NO2, trends in wildfires across the U.S. may not be well represented by the monthly gridded column NO2 product. Because wildfires have the potential to produce large amounts of ozone in the U.S. [Jaffe and Wigder, 2012], monthly NO2 emissions from fires are also examined across the U.S., based on monthly estimates in the Global Fire Emissions Database (GFED). Figures 3c and 3d shows GFED biomass burning NO2 emissions for the same regions as Figure 3a for spring and summer, 1997–2010. Emissions are greater in summer than spring, with both seasons showing large interannual variability. Linear regression analysis indicates that none of the examined regions has a statistically significant trend. The strongest fire season in the western USA from 1997 until 2010 was summer (JJA) 2007 which emitted 94 Gg NO2. According to the EDGAR v4.1 inventory for 2005, the anthropogenic NO2 emitted from the western USA in this season was 784 Gg. Therefore, the strongest fire season in the western U.S. from 1997 to 2010 was equal to 12% of the anthropogenic NO2emissions. Given the lack of a trend in wildfire emissions across the U.S. and their relatively low rates compared to anthropogenic sources, U.S. fires are not expected to influence long-term ozone trends during 1997–2010.
 An important consideration for rural surface ozone trends across the U.S. is the change in baseline ozone that has been observed during springtime. Cooper et al.  used all available free tropospheric ozone measurements above midlatitude North America from April–May, 1995–2008 to show that median ozone values in the 3.0–8.0 km range were increasing at the rate of 0.63 ± 0.34 ppbv yr−1. The rate of increase was stronger when the analysis was limited to air masses with a greater likelihood of transport from south and east Asia. Adequate data coverage was not available for a summertime analysis. Because free tropospheric ozone can be transported to the surface of the U.S. [Cooper et al., 2011; Lin et al., 2012a], it is likely that a trend in free tropospheric ozone could influence ozone trends at the surface.
 Analysis of satellite column NO2 retrievals (not shown) indicates that NO2 emissions in east Asia have doubled from 1996 to 2011, increasing at a relative rate similar to CO2 emissions (as first reported by Richter et al. ), while emissions in midlatitude North America and Europe have decreased. In response, surface ozone has decreased in the eastern U.S. (this study) and Europe [Logan et al., 2012], and increased in east Asia [Beig and Singh, 2007; Ding et al., 2008; Wang et al., 2009; Lin et al., 2010; Li et al., 2010; Parrish et al., 2012]. Apparently, the sphere of influence of the increase of emissions and ozone within Asia extends downwind as far as the free troposphere above western North America [Cooper et al., 2010]. Globally there has been little change in total anthropogenic NOx emissions from 1990 to 2010, while emissions in China have doubled [Granier et al., 2011]. If global NOx emissions remain constant but continue to shift to east Asia it can by hypothesized that the ozone production efficiency within east Asia will decrease with increasing emissions [Wu et al., 2009], while the exported Asian pollution plumes will become diluted with air masses from regions with decreasing ozone production. As a result the high ozone rate of increase observed downwind of Asia (0.63 ± 0.34 ppbv yr−1 for 1995–2008 [Cooper et al., 2010]) may not be sustained and could decrease.
Figure 4 is an update to the results of Cooper et al.  showing the April–May free tropospheric ozone trend above western North America for 1995–2001 and 1984–2011. For 1995–2011 the median ozone rate of increase is 0.41 ± 0.27 ppbv yr−1, 30% lower than, but still within the statistical uncertainty of the rate of increase determined for 1995–2008 by Cooper et al. . Several more years of data are required to determine if the ozone rate of increase has in fact declined. Regardless, from 1995 to 2011, free tropospheric ozone above western North America has increased significantly by 6.5 ppbv, and from 1984 to 2011 ozone increased by 14 ppbv.
3.2. Present-Day Rural Ozone Distribution
Figure 1 shows the locations of the 53 U.S. rural ozone monitoring sites used in this analysis. Twelve sites had data available for 1990–2010 in the western U.S., all located in mountainous or elevated regions, with 10 sites over 1 km a.s.l. and 6 sites over 2 km a.s.l. While all sites are in rural areas, ozone monitors at Pinnacles National Monument (PN), Sequoia/Kings Canyon National Parks (SK), Joshua Tree National Park (JT) and Rocky Mountain National Park (RM) are less than 100 km from large urban areas (Figure 1b). Ozone monitor density is greater in the eastern U.S. with 41 available sites from northern Maine to northern Florida. Due to the lower terrain of the eastern U.S. the monitor elevations are much lower than in the west with only 5 sites having elevations greater than 1 km a.s.l. Eastern U.S. rural ozone monitors also have a closer proximity to large urban areas due to the greater population, which exceeded the western population by a factor of 3 in 2010 (CIESIN, online report, 2005). In addition, 2005 anthropogenic NOxemissions in the eastern USA exceeded western emissions by a factor of 3.8 (EDGAR v4.1). These distributions are based on an east-west dividing meridian of 102°W and include the regions of southern Canada and northern Mexico, as shown inFigure 2.
 Present-day (2006–2010) rural ozone mixing ratios are shown across the USA for spring and summer inFigures 5 and 6, respectively. Figure 5 and Table 2 show that the range of SpO395 values in the west are similar to those in the east with Joshua Tree National Park (JT) being the only site with notably high ozone due to its location downwind of the Los Angeles Basin. SpO350 and SpO305 values are typically greater in the west than in the east. Given the greater emissions of ozone precursors in the eastern U.S. it might be expected that greater ozone values would be found in the east than in the west. However, previous studies have shown that ozone in the free troposphere (>2 km a.s.l) is similar on the east and west coasts of the U.S. in springtime due to the rapid eastward transport and lack of stagnation events during this season [Cooper et al., 2005], and that free tropospheric ozone is much greater than at the surface [Cooper et al., 2011]. The elevated terrain of the western USA is much more likely to be influenced by descending free tropospheric air masses than the eastern USA, air masses that can contain enhanced ozone due to transport from upwind emission regions such as Asia or from the stratosphere [Brown-Steiner and Hess, 2011; Cooper et al., 2011; Zhang et al., 2008; Lin et al., 2012a, 2012b]. While U.S. springtime surface ozone mixing ratios are sensitive to local emissions [Reidmiller et al., 2009; Lin et al., 2012a], terrain and the impact of baseline air masses appear to have a strong influence on the nationwide rural ozone distribution. These impacts are most apparent for the greater SpO305 and SpO350 values in the west, while for SpO395, ozone production in the eastern USA appears to match the effects of enhanced baseline ozone on the western USA.
Table 2. Average of the Ozone Percentiles at Each Site in the Western and Eastern U.S. for Spring and Summer During 1990–1994 and 2006–2010
Western U.S. 95th %
Western U.S. 50th %
Western U.S. 5th %
Eastern U.S. 95th %
Eastern U.S. 50th %
Eastern U.S. 5th %
 During summer, eastward transport is weaker than in spring and emissions from upwind regions have less of an impact on U.S. surface ozone relative to domestic emissions [Reidmiller et al., 2009]; ozone transport from the stratosphere is also less in summer than in spring [Stohl et al., 2003; James et al., 2003a, 2003b]. The diminished role of long range transport in summer might be expected to result in a smaller contrast between western and eastern ozone. However, average western surface ozone is still greater than eastern ozone for the 5th, 50th and 95th percentiles, and the difference between east and west in summer is actually greater than in spring (based on the 50th percentile; Table 2) indicating that terrain is still an important factor in summer for the nationwide ozone distribution. This comparison includes all 12 sites in the western USA, however, two sites, Joshua Tree and Sequoia/Kings Canyon are heavily influenced by anthropogenic emissions in southern California. When these two sites are omitted, the western summertime 95th, 50th and 5th ozone percentiles decrease to 65, 51 and 37 ppbv, respectively. This omission makes the western SuO395 1 ppbv less than the eastern USA value, but SuO350 and SuO305 are still greater than the eastern USA values by 5 and 9 ppbv, respectively.
3.3. 1990–2010 Rural Ozone Trends Across the U.S.
 The 1990–2010 rural ozone trends are shown in Figures 7 and 8 and statistically significant trends are summarized for the eastern and western U.S. in Table 3. In spring 41% of the eastern U.S. sites have statistically significant ozone decreases for the 95th percentile with no sites showing a significant increase. In contrast, no site in the western U.S. has a significant decrease, while 25% have a significant increase. At the 50th percentile there is little overall change in the eastern U.S. while 50% of western sites have a significant increase, and no western site has a significant decrease. At the 5th percentile increases outweigh decreases in both the east and west. Two sites in the west, Lassen National Volcanic Monument (LV) and Yellowstone National Park (YS) have significant increases for all three percentiles. The most likely explanation for the contrast in eastern and western ozone trends appears to be emission reductions causing decreases of extreme ozone events in the eastern U.S. [Hogrefe et al., 2011], while increased baseline ozone is likely causing ozone to rise at many sites in the western U.S. [Parrish et al., 2004b, 2009, 2012; Cooper et al., 2010]. We hypothesize that the increase in baseline ozone may also contribute to the rise in the eastern USA 5th percentile. However, decreasing NOx emissions may also contribute to the rise in the eastern USA 5th percentile as urban areas will experience less ozone titration, and under weak photochemical conditions the result is the export of plumes to rural areas with greater ozone concentrations.
Table 3. Regional Ozone Trends (ppbv yr−1) by Season, Calculated by Averaging the Trends of the 50th Ozone Percentiles of All Sites in That Regiona
Numbers in parentheses are the minimum and maximum trends in each region. Also shown are the percentages of sites in the western or eastern USA that experienced statistically significant changes in ozone, either positive or negative, at the 5th, 50th, or 95th ozone percentiles. For example, in the western USA in spring, 25% of sites had statistically significant positive trends of the 95th ozone percentile.
Number of sites
n = 12
n = 40
0.19 ppbv yr−1 (0.00, 0.43)
−0.03 ppbv yr−1 (−0.41, 0.51)
95th % positive
50th % positive
05th % positive
95th % negative
50th % negative
05th % negative
Number of sites
n = 12
n = 41
0.10 ppbv yr−1 (−0.39, 0.41)
−0.45 ppbv yr−1 (−0.87, 0.07)
95th % positive
50th % positive
05th % positive
95th % negative
50th % negative
05th % negative
Number of sites
n = 11
n = 36
0.12 ppbv yr−1 (−0.11, 0.46)
0.12 ppbv yr−1 (−0.13, 0.39)
95th % positive
50th % positive
05th % positive
95th % negative
50th % negative
05th % negative
 In summer the impact of domestic emission controls appears to have strongly affected the eastern U.S. across the entire ozone range, with 83%, 66% and 20% of these sites experiencing statistically significant ozone decreases in the 95th, 50th and 5th percentiles, respectively. No eastern site has a significant increase, in any percentile. Despite emission reductions in the west only 17% and 8% of sites have statistically significant ozone decrease in the 95th and 50th percentiles, respectively, all occurring in the polluted region of central California (Pinnacles National Monument (PN) and Sequoia/Kings Canyon National Parks (SK)). Surprisingly, Joshua Tree National Park (JT) has no significant downward trend in summer with the 5th percentile actually increasing significantly, despite improvements in ozone air quality in the Los Angeles Basin (http://www.arb.ca.gov/adam/index.html). Overall, western U.S. sites show more of an increase than a decrease in summer (Table 3).
Table 2summarizes very interesting seasonal changes in the timing of peak ozone values. In the western U.S., median ozone in summer was 3 ppbv greater than in spring during 1990–1994. But for the present-day (2006–2010) summertime ozone is only 1 ppbv greater than spring. This decreased seasonal contrast is primarily caused by greater ozone increases in spring than summer. Spring/summer differences for the 5th and 95th percentiles have also diminished over the past 20 years. The contrast is even stronger for the eastern U.S. In 1990–1994 eastern U.S. median ozone in summer exceeded spring by 5 ppbv, but today summer ozone is actually 1 ppbv less than spring, entirely due to decreasing ozone in summer. Spring/summer differences for the 5th and 95th percentiles have also disappeared. While ozone across the USA used to peak in summer (1990–1994), today there is a very broad spring/summer maximum. Seasonal ozone shifts have been observed at other long-term monitoring sites in the U.S. since the 1980s [Bloomer et al., 2010; Lefohn et al., 2010].
 While this paper focuses on spring and summer, for completeness, winter ozone trends are reported in Figure 9 and summarized in Table 3. This comparison shows how ozone trends are different in winter when domestic ozone production is at a seasonal minimum. Significant ozone decreases only occur at 9% of sites in the east or west and only for the 95th percentile. In contrast, ozone increases are much more common. In the west roughly a quarter to a third of all sites has significant increases across the range of ozone values. In the east just one site has a significant increase in WiO395 while 44% and 58% of sites have significant increases in WiO350 and WiO305, respectively.
3.4. Impact of Temperature Trends
 As discussed in Section 3.1 changes in emissions and baseline ozone can impact surface ozone trends across the USA, but another potential impact is regional climate change. Of all the meteorological variables affecting ozone concentrations in polluted regions, temperature is the most important [Jacob and Winner, 2009]. Surface ozone and temperature are highly correlated in the eastern United States with observed ozone increases of 2–3 ppbv per °C [Bloomer et al., 2009]. In California maximum daily ozone increases with maximum daily temperature with a slope of 2–4 ppbv per °C averaged across the 1990s and 2000s, although the slope becomes very small or even negative at temperatures above 39°C [Steiner et al., 2010]. The increase in ozone with temperature is linked to 1) the temperature-dependent lifetime of peroxyacetylnitrate (PAN), 2) the temperature dependent biogenic emission of isoprene, and 3) air mass stagnation and sunny skies that accompany high surface temperatures [Jacob and Winner, 2009]. Many modeling studies have calculated the response of ozone to a warmer future climate due to rising concentrations of anthropogenically produced greenhouse gases. The robust findings indicate increased surface ozone in polluted regions, but decreased surface ozone in remote regions due to higher water vapor concentrations [Jacob and Winner, 2009]. Murazaki and Hess  examined the potential impact of climate change on U.S. surface ozone from the period 1990–2000 to 2090–2100. They calculated that a 2–4°C increase in surface temperature would result in surface ozone enhancements of up to 6 ppbv in polluted regions. Baseline ozone in the free troposphere above the U.S. west coast would decrease by 2–6%, due to the shorter lifetime of ozone in a warmer climate, resulting in a decreased impact at the surface of the western U.S., up to 2 ppbv. Wu et al. found similar results for the shorter time period of 2000–2050 (U.S. summertime temperature increases as much as 2–3°C), except they found little change in ozone in the mid-troposphere of northern midlatitudes.Wu et al.  also found that in April, shifting synoptic scale transport patterns could result in increased baseline ozone impacting some regions of the western U.S.
 Global average surface temperature increased by 0.074°C ± 0.18°C per decade when estimated by a linear trend for the 100 year period,1906–2005 [Intergovernmental Panel on Climate Change (IPCC), 2007], while the rate of increase since the late 1970s is 0.15°C–0.20°C per decade [Hansen et al., 2010]. As global temperatures have increased since the 19th century so too has the global tropospheric ozone burden, primarily due to rising anthropogenic emissions of ozone precursors [Lamarque et al., 2005]. Based on the model studies of ozone response to future climate change, one might assume that past ozone changes have also been influenced by climate change since the 19th century. A recent intercomparison of 10 atmospheric chemistry models run with year 2000 emissions but with 2000s and 1850s climate shows a range of responses of tropospheric ozone to the observed temperature increase [Stevenson et al., 2012]. Six out of 10 models indicate ozone decreases at the surface of the northern hemisphere midlatitudes due to observed climate change, but the decreases are small (<2 ppbv). Four out of 10 models indicate regions of both positive and negative surface ozone changes, but these changes are also small (±2 ppbv). In the free troposphere of northern midlatitudes the models indicate a range of ozone changes that are also small (±2 ppbv).
 Given the small and variable response of modeled ozone to observed climate change from the 1850s to 2000s we do not expect a strong impact from climate change over the much shorter time periods of 1990–2010 (surface ozone analysis) and 1984–2011 (western North America free tropospheric ozone analysis). Table 4shows the changes in surface temperature in the contiguous U.S., as well as in 10 sub-regions (corresponding to the states and regions with ozone measurements), for spring, summer and winter for 1895–2011 and 1990–2010, based on the slope of the linear regression line fit through the yearly average temperature values for each season (temperature data from NOAA National Climatic Data Center:http://www.ncdc.noaa.gov/temp-and-precip/time-series/). Using the full data record (1894–2011) the average surface temperature across the contiguous USA increased significantly by 0.84°C, 0.70°C and 1.05°C during spring, summer and winter, respectively, rates similar to the global average. However, regionally, the trends are much more variable with weaker (and mostly insignificant) trends in the east and stronger trends in the west. On the much shorter time scale of 1990–2010, the only region with a statistically significant trend is Colorado in summer, where temperatures increased by 1.3°C over 20 years. Using the ozone/temperature relationship of Bloomer et al.  we can estimate that the increase in temperature over Colorado potentially increased summertime ozone by 3–4 ppbv from 1990 to 2010.
Table 4. Seasonal Changes in Temperature (°C) in the Contiguous USA and in 5 Eastern Regions and 5 Western States From 1990 to 2010 and 1895–2011a
The changes in temperature are based on the slope of the linear regression line fit through the yearly average temperature values for each season. Entries in bold indicate temperature changes that are statistically significant (p < 0.05). States in the five eastern regions are as follows: South: Kansa, Oklahoma, Arkansas, Texas, Louisiana, Mississippi; Southeast: Virginia, North Carolina, South Carolina, Georgia, Alabama, Florida; Central: Missouri, Illinois, Indiana, Ohio, Tennessee, Kentucky, West Virginia Northeast: Maine, Vermont, New Hampshire, Massachusetts, Rhode Island, Connecticut, New York, New Jersey, Pennsylvania, Delaware, Maryland; East North Central: Minnesota, Iowa, Wisconsin, Michigan.
E. North Central
 In the free troposphere, northern hemisphere springtime (MAM) mid-tropospheric (850–300 hPa) temperatures increased significantly by 0.7°C from 1984 to 2011, as determined from radiosonde measurements compiled by NOAA's Radiosonde Atmospheric Temperature Products for Assessing Climate (RATPAC) (data downloaded fromhttp://www1.ncdc.noaa.gov/pub/data/ratpac/). As described above, relatively large temperature increases of 2–4°C should result in increased tropospheric water vapor and a decrease in ozone lifetime in remote regions of the lower troposphere. However, model results are less conclusive for the mid-troposphere [Stevenson et al., 2012] and at present the model evidence does not allow us to conclude that the observed increase in mid-tropospheric temperatures would have caused a significant change in ozone.
 From the analysis of surface and mid-tropospheric temperature trends over the relatively short time spans of 20 and 27 years, respectively, we conclude that there is no convincing evidence that the observed temperature changes have produced a measurable change in ozone in the free troposphere above western North America during springtime. At the surface there are no significant 1990–2010 temperature trends in 10 sub-regions of the U.S. in spring, summer or winter, except for Colorado in summer. The increase in summertime surface temperatures in Colorado may have increased surface ozone by 3–4 ppbv.
3.5. Ozone Trends at Selected Sites
 Further indications of the impact of baseline versus domestically produced ozone on trends across the western and eastern U.S. can be inferred by examining the ozone trends at several regionally representative sites. Figure 10shows the spring and summer ozone trends at five sites in the eastern U.S.: Ashland (AL) a fairly remote site in northern Maine; Cape Cod (CC) located in a rural area of the Cape Cod peninsula, downwind of Boston; Whiteface Mountain Summit (WF) at 1.5 km a.s.l. in northern New York state; Big Meadows (BM) at 1.1 km a.s.l. on a ridge in Shenandoah National Park; and Cove Mountain (CM) at 1.2 km a.s.l. in Great Smoky Mountains National Park. These sites were chosen to indicate ozone trends at elevated sites along the spine of the eastern U.S. (WF, BM, and CM), as well as ozone trends in remote northern Maine (AL), and to compare them to trends immediately downwind of the Boston/New York City metropolitan region (CC). Despite the wide range of environments represented by these five sites, ozone values for the 5th and 50th percentiles have little springtime variation from north to south, consistent with the hypothesis that the eastern U.S. in spring has a strong influence from upwind air masses that reduce regional variability. However, the 95th percentile does exhibit a large north-south gradient with less ozone at Ashland and Whiteface Mountain Summit. High ozone events in the rural eastern U.S. are often associated with warm temperatures and stagnation events which increase the influence of local emissions and are less likely to impact the northern sites of Ashland and Whiteface Mountain Summit [Logan, 1989; Moody et al., 1998; Cooper and Moody, 2000; Fiore et al., 2002; Fischer et al., 2004]. In summer, transport speeds are lower which increases the influence of regional emissions on rural ozone and leads to greater ozone variability between sites, in comparison to spring. Figure 10 also shows the seasonal shift in the timing of peak ozone discussed in Section 3.3. In the early 1990s Cove Mountain and Big Meadows had higher ozone in summer than in spring, but in recent years, spring and summer ozone values are similar due to the decrease in summertime values. In the early 1990s the 95th percentile at Whiteface Mountain Summit was greater in summer than spring and the 50th and 5th percentiles were similar for both seasons, but today spring values are greater for all three percentiles. Similarly, Ashland's present-day ozone is less in summer for all three percentiles. Only Cape Cod shows greater ozone in summer, but this applies just to the 95th percentile and this difference narrowed during 1990–2010. We have no explanation for the significant increase in SpO350 at Cape Cod.
Figure 11 shows the spring and summer ozone trends for the 6 monitoring sites in the western U.S. with the greatest elevations: Grand Canyon (GC), Gothic (GO), Rocky Mountain (RM), Centennial (CN), Pinedale (PD) and Yellowstone (YS). These sites were chosen because they are located within the region covered by Wyoming, Colorado, Utah, Arizona and New Mexico, the high elevation portion of the western U.S. with the strongest influence from baseline ozone [Zhang et al., 2011; Lin et al., 2012a]. In springtime there is little variability among the 6 sites, with ozone values being more similar in 2006–2010 than 1990–1994. Overall, ozone precursor emissions in the western U.S. declined from 1990 to 2010, while springtime baseline ozone mixing ratios have increased in the marine boundary layer [Parrish et al., 2009] and in the free troposphere (this study). The reduced ozone variability in recent years may be due to greater influence from baseline ozone and less influence from regional emissions. Present-day ozone is similar at each site between spring and summer, with the exception of the 95th percentile at Rocky Mountain National Park which is greater in summer than spring.
 Despite the similarity in ozone among the sites in spring and summer, and despite the strong influence of baseline ozone across this region, the 1990–2010 trends don't agree among sites. Three sites (Yellowstone, Rocky Mountain and Grand Canyon) show strong significant increases in SpO350 while the other three sites are essentially flat. The same is generally true for SuO350, although the positive trend at Rocky Mountain in not significant. The three sites with the strongest ozone increases are all in National Parks and are generally closer to urban areas than Gothic, Pinedale and Centennial. This raises the possibility that regional changes in ozone precursor emissions caused the ozone increases in the National Parks. Figure 2 shows that the population in the regions surrounding the National Park sites increased by 31%–95% during 1990–2010, but column NO2 in the same regions did not increase, and actually declined significantly in the northwest Arizona/Las Vegas region (Figure 3b). Therefore, regional changes in ozone precursor emissions do not seem to be the cause of the increase in ozone.
 To examine this issue in more detail, ozone and ozone precursor trends are calculated for the Denver metropolitan region, a major urban area that can export ozone and ozone precursors to nearby Rocky Mountain National Park and surroundings in summer and to a much lesser extent in spring [Fehsenfeld et al., 1983; Oltmans and Levy, 1994; Brodin et al., 2010]. Figure 12a shows urban NO2trends at the two Denver sites with long-term measurements. In both spring and summer NO2decreases at both sites although only one site in each season has a statistically significant decrease. Long-term and reliable CO measurements are available from one Denver site and show very strong decreases over the past 20 years (Figure 12b) similar to those seen in other large urban areas of the U.S. [Bishop and Stedman, 2008; Parrish et al., 2011; Warneke et al., 2012]. In contrast, CO at the long-term monitoring site of Niwot Ridge (3526 m a.s.l.), which is 25 km south of Rocky Mountain National Park and west of Denver, is much lower than in Denver and comparable to the baseline CO mixing ratios observed by aircraft at 3500 m above Trinidad Head-on the U.S. west coast (http://www.esrl.noaa.gov/gmd/dv/iadv/). Sampling at Niwot Ridge is avoided if there are upslope winds that could bring Denver pollution to the site; therefore Niwot Ridge is broadly representative of the western U.S. lower troposphere. Despite steep decreases in anthropogenic CO emissions across the western U.S., Niwot Ridge shows only a weak decrease in CO in spring and no trend in summer, suggesting that this high elevation region of the Rocky Mountains is strongly influenced by the free troposphere.
Figure 13shows ozone trends from 1990 or the mid-1990s through 2010 at 4 sites in the Denver metropolitan region. These sites tend to exceed the NAAQS for ozone every summer and are somewhat isolated from fresh urban emissions such that they are regionally representative of the conurbation. During spring these sites show a slight increase in 50th percentile ozone but none of the trends are statistically significant. Trends in summer are not statistically significant either. The lack of change in ozone mixing ratios in the Denver region despite the strong decrease in ozone precursors is currently unexplained and requires detailed chemical transport modeling to identify the cause. Possibilities are increasing temperatures in summer (as described above), increase in regional baseline ozone, or increase in local ozone production from unidentified emissions sources. In any event, decreasing ozone precursors in Denver and the lack of a trend in Denver cannot explain the springtime increase in ozone at Rocky Mountain National Park especially since springtime meteorology does not favor transport from Denver to the Park. The ozone rate of increase at Rocky Mountain National Park is similar in spring and summer although during summer the statistical significance is just outside the 95% confidence level. A portion of the summertime ozone increase might be due to significantly increasing temperatures in Colorado during summer.
4. Discussion and Conclusions
 An area of current research is focusing on the uncertainty of ozone precursor emissions from petroleum and natural gas development across the U.S. According to the U.S. Energy Information Administration, oil and gas production in the lower-48 states has increased in recent years with oil production projected to increase through the year 2020, and gas production projected to increase through 2035 [U.S. Energy Information Administration, 2011]. Extraction of oil and gas can result in the emission of ozone precursor gases [Schnell et al., 2009], and several modeling studies indicate that these emissions can produce regional scale ozone enhancements [Rodriguez et al., 2009; Kemball-Cook et al., 2010; Carter and Seinfeld, 2012]. U.S. EPA estimates indicate that NOx emissions from U.S. petroleum and related industries increased by a factor of 4 from 2000 until 2010, but only amounted to 4% of total U.S. anthropogenic NOx emissions in 2010 (EPA, online report, 2012). Likewise VOC emissions increased by a factor of four from 2000 until 2008 but have since decreased, and accounted for 9% of total U.S. anthropogenic VOC emissions in 2010. On the other hand, trace gas measurements within the gas development region of northeastern Colorado indicate that current inventories may underestimate emissions from gas development by as much as a factor of two [Pétron et al., 2012]. Improved oil and gas production emission inventories need to be developed and utilized by chemical transport models to better assess any impact of oil and gas development on rural ozone levels and trends.
 Another area of uncertainty is the impact of changing baseline ozone mixing ratios on surface ozone across the rural United States. Reidmiller et al.  summarize the results of a multimodel experiment that explored the impact of 20% ozone precursor emission reductions on U.S. surface ozone. Sixteen chemical transport models were run for the year 2001 for two scenarios: one with global 2001 emissions at 100%, and one with 2001 emissions reduced by 20% in East Asia, South Asia, Europe and North America. Focusing first on the impact of foreign emissions on U.S. surface ozone, they found that the U.S. was more sensitive to emission changes in East Asia than Europe or South Asia. Regionally, the western U.S. was more sensitive to Asian emissions than was the eastern U.S. Focusing on a 20% reduction in ozone precursors within just North America they found that the U.S. was far more sensitive to domestic changes than to foreign changes and that these domestic precursor emission reductions produce the greatest ozone reductions in the northeastern U.S. in summer, especially for high ozone events. They also concluded that domestic precursor emission reductions would reduce ozone in all regions of the U.S. in spring and summer.
 These model results are in general agreement with many of the findings of this study. During 1990–2010, ozone precursor emissions dropped greatly across the U.S. in western, northeastern and southeastern regions and the largest decreases in ozone have been observed in the northeast U.S. in summer and at the high end of the ozone distribution. However, the ozone increases at several sites in the western U.S. in summer and especially in spring, along with some eastern U.S. sites experiencing ozone increases in the SpO305, are not congruent with the decreases in domestic emissions. According to the findings of Reidmiller et al.  if ozone trends in the U.S. were solely controlled by domestic emissions we should expect to see ozone decreases in all regions of the U.S. in both spring and summer. The limited ozone reductions in the western U.S., as well as ozone increases at many sites, lend weight to the argument that ozone increases from upwind sources are offsetting ozone reductions from decreased domestic emissions. The increase in ozone flowing into the western U.S. is supported by observations ([Parrish et al., 2009; Cooper et al., 2010] and results from this study) and the sensitivity of western U.S. ozone to changes in ozone precursor emissions in East Asia has been proposed by chemical transport model studies [Jacob et al., 1999; Reidmiller et al., 2009].
 As discussed in Section 3.4, temperature trends can impact ozone trends, however statistically significant 1990–2010 temperature trends have only been observed in Colorado during summer where increasing temperatures may have contributed to increasing ozone. An additional uncertainty partially influenced by climate change is the impact of land use and land cover changes on atmospheric chemistry [Fall et al., 2010; Diffenbaugh, 2009; Wu et al., 2012]. Climate change can affect plant productivity and shift biomes while human activity changes the land surface area covered by forests and cropland. These changes in land use and land cover affect the emission of biogenic VOCs and NO that act as ozone precursors and also affect ozone dry deposition rates. A recent model study calculated the change in surface ozone between 2000 and 2050 from the combined impact of changes in climate, CO2 abundance and agricultural land use. During summer most of the western U.S. showed no change in ozone while the northern Great Plains and the eastern U.S. showed ozone decreases of 1–5 ppbv. Therefore, land use change can decrease ozone in the eastern U.S., the opposite impact of a warming climate [Murazaki and Hess, 2006; Wu et al., 2008]. The impacts of climate change and land use change are difficult to assess over the relatively short time period of 20 years. In contrast the strong changes in anthropogenic emissions in the USA and east Asia over the past 20 years appear to be the most likely reason for the observed ozone changes.
 Due to the paucity of long-term observations of ozone precursors in the rural USA and in the free troposphere above the USA, our ability to attribute the causes of the ozone trends based on observations alone is limited. Further insight into the impact of changing global ozone precursor emissions on U.S. surface ozone can be gained through long-term chemical transport model simulations. These simulations should cover the 1990–2010 time period using reanalysis wind fields that include observed regional temperature trends. The simulations should also account for land use changes and include realistic emission inventories that reflect the changes in the global distribution of ozone precursor emissions. If the base-case runs can reproduce the ozone trends in the eastern and western U.S., a test case should be run with Asian emissions held constant at 1990 levels. If Asia is the cause of the positive or level ozone trends in the western U.S., the test case should result in downward ozone trends across the western U.S.
 In conclusion, this analysis has provided an up-to-date assessment of long-term (1990–2010) rural ozone trends in the western and eastern United States, focusing on the 5th, 50th and 95th ozone percentiles in spring, summer and winter. We found very different springtime trends in the eastern and western U.S. Forty-one percent of the eastern sites have statistically significant negative ozone trends for the 95th percentile with no site showing a significant increase. In contrast, no western site has a significant decrease, while 25% have a significant increase. At the 50th percentile there is little overall change in the eastern U.S. while 50% of western sites have a significant increase. At the 5th percentile, increases outweigh decreases in both the east and west.
 During summer the impact of domestic emission controls appears to have strongly affected the eastern U.S. across the entire ozone range, with 83%, 66% and 20% of these sites experiencing statistically significant ozone decreases in the 95th, 50th and 5th percentiles, respectively. Despite emission reductions in the west only 17% and 8% of sites have statistically significant ozone decreases in the 95th and 50th percentiles, respectively.
 Wintertime trends in the eastern U.S. are very different. Significant ozone decreases only occur at 9% of sites in the east or west and only for the 95th percentile. In contrast, ozone increases are much more common. In the west roughly a quarter to a third of all sites have significant increases across the range of ozone values. In the east just one site has a significant increase in the 95th percentile while 44% and 58% of sites have significant increases in the 50th and 5th percentiles, respectively.
 Many previous modeling studies indicate that the ozone precursor emission reductions that have occurred across the U.S. during 1990–2010 should have resulted in broad ozone reductions across the country. The limited ozone reductions in the western U.S. suggest that increasing baseline ozone is counteracting domestic emission reductions. Finally, an update to the springtime free tropospheric ozone trend above western North America shows that ozone has increased significantly from 1995 to 2011 at the rate of 0.41 ± 0.27 ppbv yr−1.
 Ozone data from the National Park Service Gaseous Pollutant Monitoring Program were collected by the National Park Service and downloaded from the NPS Gaseous Pollutant and Meteorological Data Access Page: http://ard-request.air-resource.com. Assistance in retrieving the data was provided by Jessica Ward, Air Resource Specialists, Inc. and John Ray, National Park Service Air Resources Division. Whiteface Mountain Summit ozone data were collected by the University of Albany with instrumentation provided by the New York State Department of Environmental Conservation, and provided by James J. Schwab, Atmospheric Sciences Research Center, University at Albany. Surface CO measurements from Niwot Ridge measured and made available by the NOAA Earth System Research Laboratory Carbon Cycle Group. We gratefully acknowledge the strong support of the MOZAIC program by the European Communities, EADS, Airbus and the airlines (Lufthansa, Austrian, Air France) who have carried the MOZAIC equipment free of charge since 1994. Jean-Pierre Cammas and Philippe Nedelec at CNRS - Laboratoire d'Aérologie, France, provided access to the MOZAIC data. The EDGARv4.1 global NOx emissions inventory was provided by European Commission, Joint Research Centre (JRC)/Netherlands Environmental Assessment Agency (PBL): Emission Database for Global Atmospheric Research (EDGAR), release version 4.1 http://edgar.jrc.ec.europa.eu, 2010. We acknowledge the free use of tropospheric NO2 column data from the GOME and SCIAMACHY sensors from www.temis.nl. The global topography data at 4 min resolution were downloaded from: NOAA National Geophysical Data Center, Boulder, www.ngdc.noaa.gov/mgg/global/global.html. Finally, we thank three anonymous referees whose comments and suggestions improved the analysis. Funding for the Trinidad Head ozonesondes was provided by the NOAA ESRL Health of the Atmosphere Program.