Corresponding author: D. D. Parrish, NOAA ESRL Chemical Sciences Division, 325 Broadway, Boulder, CO, USA. (David.D.Parrish@noaa.gov)
 At northern midlatitudes the abundance of tropospheric O3 has increased by a factor of approximately 2 since the 1950s. The cause of this increase is generally attributed to increasing anthropogenic precursor emissions, but present chemical and transport models cannot quantitatively reproduce its magnitude. Here we show another manifestation of changes in O3 abundance—a shift of the seasonal cycle at northern midlatitudes so that the observed peak concentrations now appear earlier in the year than in previous decades. The rate of this shift has been 3 to 6 days per decade since the 1970s. We examine possible reasons to explain this shift and suggest it is due to changes in atmospheric transport patterns combined with spatial and temporal changes in emissions. Detailed modeling is necessary to test these hypotheses; this investigation will provide useful guidance for improving global chemistry-climate models and stringent tests of the model results.
 Ozone (O3) is one of the most important trace constituents of the troposphere, where it acts as a pollutant in rural and urban areas [Royal Society, 2008], as a greenhouse gas, thereby contributing to global climate change [Gauss et al., 2006], and as the precursor of the hydroxyl radical, which is responsible for oxidation and removal of many atmospheric pollutants [Levy, 1971]. In the troposphere, O3 originates from downward transport from the stratosphere and from photochemical production from precursors emitted by natural and anthropogenic sources. It is destroyed photochemically and by dry deposition at the surface [Crutzen, 1973]. Ozone and its precursors are also transported between continents. It can also be formed over downwind regions from decomposition of nitrogen oxide (NOx) reservoirs, in particular peroxyacetyl nitrate. Thus, O3 concentrations over a particular region depend not only on local and regional emissions, but also on import from upwind emission regions [HTAP, 2010].
 Analysis of surface data from northern midlatitudes shows that the abundance of tropospheric O3 has changed, increasing by about a factor of 2 since the 1950s, with generally increasing surface O3 in all seasons at relatively remote sites at northern midlatitudes [e.g., Staehelin et al., 1994; Parrish et al., 2012, and references therein]. This increase slowed, at least over Europe, by the late 1990s, and at some European sites O3 has begun to decrease, particularly in summer [e.g., Logan et al., 2012]. Chemical transport models generally attribute this general O3 increase to increasing anthropogenic precursor emissions and changes in meteorology or natural emissions [e.g., Stevenson et al., 2012]. However, attribution varies significantly between studies and models have difficulties quantitatively reproducing the magnitude of observed trends [e.g., Pozzoli et al., 2011]. Here we show another manifestation of changes in O3 abundance, a shift of the seasonal cycle at northern midlatitudes so that observed seasonal maxima now appear earlier in the year than they did 20 years ago.
 Over populated continental areas at northern midlatitudes, surface O3 exhibits a well-known summertime maximum that is generally attributed to the peak in local and regional photochemical production at this time of year. Over more remote continental areas surface O3 generally has a late spring maximum attributed to both enhanced stratospheric input and hemisphere-wide photochemical production during that season [Holton et al., 1995; Monks, 2000]. Sites within the marine boundary layer exhibit a spring maximum with a summer minimum due to enhanced photochemical destruction in that season. The general long-term O3 increase has important seasonal differences that necessarily imply a concurrent change in the seasonal cycle of O3. For example, at Hohenpeissenberg in Germany, where O3 has been continuously measured since 1971, summer was the season of maximum O3 throughout at least the first three decades of the data record. However, in recent years O3 concentrations in spring have been comparable to those in summer [see Parrish et al., 2012, Figure 2]. Here we quantitatively examine how the seasonal cycle in tropospheric O3 at northern midlatitudes has evolved over the past four decades, possibly in response to changing anthropogenic emissions, natural variability, and changing global climate.
2 Data Sets and Analysis
 To quantify long-term changes in lower tropospheric O3 concentrations at northern midlatitudes, Parrish et al.  examined 11 data sets that included measurements in Europe, North America, and Asia. Here we examine five of those data sets that are most appropriate (those with longest continuous data records from single sites with year-round coverage) for quantifying the evolution of the O3 seasonal cycle over decadal time scales. These data sets include four European records: two from alpine sites (Zugspitze, Germany and Jungfraujoch, Switzerland), a lower elevation, midcontinental site (Hohenpeissenberg, Germany) and a marine boundary layer site (Mace Head, Ireland). The European data were provided by coauthors of this paper; the German and Swiss data are as described by Gilge et al. . The Jungfraujoch and Hohenpeissenberg data are also available from the World Data Centre for Greenhouse Gases at http://ds.data.jma.go.jp/gmd/wdcgg/. The Mace Head data are filtered for baseline conditions based on the NAME Lagrangian dispersion model as described by Derwent et al. . A single North American site is also examined. Monthly averages at Lassen NP were calculated from hourly average data provided by the U.S. National Park Service, Air Resources Division (available at http://www.nature.nps.gov/air/monitoring/network.cfm). Parrish et al.  give more details of these data sets and sites, including a map of site locations. For all sites, the full data records are included in the analyses.
 We use two approaches to quantify the O3 seasonal cycle change and its uncertainty at a particular site. First is to fit a sine function to monthly average data from 5 year time periods at the beginning and at the end of the data record, and then to determine if the parameters of the functional fits are statistically significantly different between the two time periods. These fits are to a three-parameter equation
where yo is the annual average O3, A is the magnitude of the seasonal cycle, χ the month (where 12 months corresponds to 2π radians), and ϕ is the phase shift of the seasonal cycle. The derived value of ϕ allows calculation of the date of the year of the annual O3 maximum: (π/2−ϕ)*365/2π. Equation ((1)) also can be fit to sequential, overlapping 5 year periods over the entire data record at each site, which then provides a determination of the continuous evolution of the parameters in equation ((1)). In this latter application, the yo parameter in equation ((1)) is obtained from the average of the seasonal averages over the 5 year period. The seasonal O3 averages for each year are directly available from the quadratic fits to the seasonal O3 trends presented by Parrish et al. . This reduces the least-squares regression fit from three to two parameters. Three-parameter regressions to equation ((1)) give results in close (statistically not significantly different) agreement with the two-parameter fits, but the two-parameter fits give somewhat more precise determinations of the A and ϕ parameters. Confidence limits for all derived parameters are calculated from standard propagation of error techniques, properly accounting for the number of independent data points included in the regressions.
 The second approach is to calculate the date of the year when the cumulative ozone concentration reaches 50% of the year's total. This date is determined through three steps designed to be least affected by any gaps of missing data. First, the cumulative hourly O3 concentration for the entire year is calculated by summing the accumulated hourly average O3 concentrations in each of the 12 months. The accumulated hourly average O3 concentration for each month is calculated as the product of the monthly average O3 concentration multiplied by the hours in the month. Second, the monthly accumulated hourly averages are examined to determine the month during which the cumulative ozone concentration reaches 50% of the year's total. In all cases to be considered that month is either June or July. Third, sequential hourly O3 averages before or after July 1 are summed to find the date and time corresponding to 50% of the year's total; periods of a few hours or less of missing data are linearly interpolated. This process allows for the unbiased treatment of data sets with some missing data, as long as the data are complete enough to (1) allow calculation of accurate monthly average O3 concentrations for all months of the years considered, and (2) provide nearly continuous records during the June–July period that includes the 50% cumulative O3 date. The one data set treated in this manner (Lassen NP) meets these criteria for all years considered.
 In the final section, we will discuss column NO2 maps. They are derived from 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. . Data are available for 1996–2011 at 0.25°×0.25° horizontal resolution.
 There is substantial evidence that the seasonal cycle has shifted to earlier in the year at all five of the sites investigated here. For the four sites included in Figure 1, the seasonal maximum occurred 10 to 48 days earlier during the later 5 year period compared to the earlier period; however, only at the two sites with the longest data records, Zugspitze and Hohenpeissenberg, are the differences significant at the 95% confidence level. As expected, significant increases in annual average O3 are also evident at all sites, reflecting the general O3 increase that has occurred at northern midlatitudes [e.g., Parrish et al., 2012].
 The temporal evolution of the date of the seasonal maximum obtained from fitting a sine function to running 5 year periods (Figure 2) indicates a shift of the seasonal cycle to earlier in the year for all five stations. Table 1 gives the parameters for the linear, least-squares fits to these seasonal maxima as a function of the centers of the 5 year periods. The three European continental sites all have late spring to early summer maxima (dates interpolated to the year 2000 are given in Table 1) that move to earlier in the year at statistically significant and similar rates of 5 to 7 days per decade. Mace Head, a European marine boundary layer site, has a late winter to early spring maximum that also moves to earlier in the year, although the rate of evolution (3.0 ± 3.7 days/decade) is smaller than found at the continental sites, and is not statistically significant at the 95% confidence limit. Oltmans et al.  also noted an indication of a shift (from April–May to April) of monthly average maximum O3 concentrations at Mace Head. These months are later than the March seasonal maxima shown in Figure 2. However, Oltmans et al.  considered the full data set rather than the baseline-selected data considered here. Their results are likely affected by episodes of NOx-driven depletion and/or photochemical production of O3 in polluted European air masses transported to Mace Head. Nevertheless, the seasonal shift they note parallels that in Figure 2. Lassen NP (the one North American site) has a seasonal cycle that is only poorly represented by a sine function, but also shows indications of a seasonal cycle similar to the European sites together with a similar shift although the shift is not statistically significant. A statistically significant shift in the date of the year when the cumulative ozone concentration reaches 50% of the year's total at Lassen NP (3.6 ± 3.0 days/decade) is apparent (Figure 2). The magnitude of this shift cannot be directly compared to the shift in the seasonal cycle maxima given in Table 1, because these are different measures of the phase of the seasonal cycle.
Table 1. Temporal extent of data records and results of linear regressions (Figure 2) to the parameters of the sine function least-squares fits (based on Eq. ((1))) to five data sets. Indicated uncertainties are 95% confidence limits of the respective quantities
Date of Seasonal Maximum in 2000
Shift in Seasonal Maximum (Days/Decade)
14 June ± 2.0 days
5.6 ± 4.1
10 June ± 3.6 days
5.1 ± 3.5
9 June ± 3.3 days
6.4 ± 2.4
Mace Head, Ireland
2 March ± 1.9 days
3.0 ± 3.7
Lassen NP, California, USA
14 June ± 11 days
14 ± 19
 In summary, these five sites, which provide the longest continuous records of surface O3 concentrations at relatively remote sites in the northern midlatitudes, all show substantial evidence for the seasonal cycle shifting so that the yearly maximum O3 appears earlier in the year. Long-term increases in northern midlatitude, lower troposphere O3 have occurred from the 1950s to the present [e.g., Staehelin et al., 1994; Parrish et al., 2012]. The O3 seasonal cycle has changed concurrently with this increase in concentrations (Figures 1 and 2). Importantly, this seasonal cycle shift is not due simply to local or regional influences. The Mace Head data analyzed here have been filtered to remove recent European influence [see Parrish et al., 2012]. A separate analysis of a Jungfraujoch data set that was similarly filtered [Cui et al., 2011] yields a seasonal shift statistically consistent with those shown in Figure 2a. Although these sites provide only sparse coverage of the northern midlatitudes, the sites do represent baseline (here understood as representative of continental to hemispheric scales) O3 [Parrish et al., 2012], so we hypothesize it is a hemisphere-wide phenomenon. To support this hypothesis, it would be useful to extend this analysis to other O3 data sets (e.g., those described in Logan et al. ), although we note that few existing long-term records are appropriate for this type of analysis. We now discuss possible causes for this seasonal cycle shift.
 The seasonal cycle in tropospheric O3 at northern midlatitudes has evolved over the past four decades in response to changing anthropogenic emissions, natural variability on annual or decadal time scales, and changing global climate. Climate change may have caused changes in stratosphere to troposphere transport, tropospheric transport patterns, and increased temperatures, water vapor, and changes in natural emissions (e.g., lightning, forest fires, and vegetation). Here we suggest possible hypotheses to explain the observed changes in the O3 seasonal cycle. The following discussion considers several possibilities related to (1) changes in transport and pollution transport pathways, (2) changes in emissions and ozone photochemistry, and (3) factors related to climate change.
 First, changes in atmospheric transport may have influenced pollution transport pathways or transport of O3 from the stratosphere, possibly in response to changing climate. Transport of stratospheric O3 to the troposphere is at a maximum in spring [Tang et al., 2011], while photochemical O3 production maximizes in summer. Hence, the observed shift of the O3 maximum concentration to earlier in the year may imply that the relative contribution from the stratosphere is increasing. Strong correlations between observed O3 anomalies in the lower stratosphere and, for example, European mountain sites, support this hypothesis, although no trend in air mass fluxes across the tropopause was found [Ordóñez et al., 2007]. Changes in the large-scale circulation in the troposphere may have also played a role. For example, an increased prevalence of stronger westerly flow over the North Atlantic, related to changes in the North Atlantic Oscillation, from the late 1980s up to the late 1990s may have led to increased O3 in northwestern Europe in the winter and spring [Pausata et al., 2012]. These changes may result in enhanced transport of pollution from North America to Europe at this time of year [Creilson et al., 2003] due to more active frontal transport [Eckhardt et al., 2003]. They may also be related to increased downward transport of stratospheric O3 [Sprenger and Wernli, 2003]. However, although certain studies suggest significant stratospheric influence, even at surface sites [e.g., Hess and Zbinden, 2013], other analyses show larger contributions from upwind regions compared to the stratosphere (e.g., Cui et al.  find that only 4–8% of air masses at Jungfraujoch have strong stratospheric influence). This suggests that enhanced, more efficient intercontinental transport of O3 and its precursors in winter and spring may have contributed to changing seasonality.
 Second, changing emissions and subsequent photochemical production within the troposphere may also have played a role in the shift of the O3 seasonal cycle. Over the past two decades, global anthropogenic emissions of O3 precursors have remained approximately constant in magnitude [Granier et al., 2011], but their spatial distribution has changed significantly. North American and European emissions have generally decreased, while eastern Asian emissions have increased. The satellite data presented in Figure 3 demonstrate that this effective shift of emissions is not only between continents, but also from more northerly to more southerly latitudes. This may have led to an earlier onset of significant photochemical production, supported by observations of earlier O3 maxima at southerly Japanese sites [Tanimoto et al., 2005]. Earlier spring-time photochemical O3 production over Asia may also be leading to increased peroxyacetyl nitrate production that could lead to increased ozone production during transport and/or over downwind continents. These factors coupled with more efficient springtime vertical and westward transport of Asian pollution and more efficient O3 production in lofted air masses [Wild and Akimoto, 2001] may explain the increase in springtime O3 concentrations in more recent years over North America and Europe. Significant increases in springtime free tropospheric O3 over North America, particularly in air masses originating in East Asia, have been reported by Cooper et al. [2010, 2012]. Multimodel apportionment studies also support enhanced springtime contributions from Asian emissions to European lower tropospheric O3 over the last 2 decades (Oliver Wild et al., manuscript in preparation, 2013).
 In contrast, within Europe and North America, decreasing O3 precursor emissions have led to reductions in peak summertime ozone concentrations at surface continental sites [Royal Society, 2008; Cooper et al., 2012]. At the same time decreasing NOx emissions have resulted in less loss of O3 via NO titration in winter and early spring [e.g., Jonson et al., 2006], manifesting itself in positive wintertime O3 trends at European and North American sites [e.g., Wilson et al., 2012; Parrish et al., 2012; Cooper et al., 2012]. It is worth noting that stronger seasonal cycle changes are seen in unfiltered Mace Head data (not shown) most likely driven by reductions in European NOx, especially in winter. Hence, reduced summertime photochemical production and reduced winter/early spring photochemical removal of O3 at more northerly European and North American latitudes, combined with increased springtime photochemical production of O3 at more southerly Asian latitudes, may synergistically act to yield an earlier seasonal O3 maximum.
 Third, changes in other processes linked to climate change or natural variability may have also played a role. Variations in temperature, water vapor, and natural emissions, driven by meteorological conditions, contribute to surface O3 variability on interannual or multiannual time scales, although there are large differences between factors found to be important in different studies [e.g., Hess and Mahowald, 2009; Pozzoli et al., 2011]. While the impacts of climate change due to increases in anthropogenic emissions on the O3 seasonal cycle are currently unknown, some studies [e.g., Stevenson et al., 2012] suggest that climate change between preindustrial times and present day has affected Northern Hemisphere lower tropospheric O3 by just a few ppbv.
 Without detailed global chemical model simulations that can accurately reproduce the observed O3 seasonal cycle, including its magnitude, changing phase, and spatial variation, it is not possible to definitively test these hypotheses, determine which is dominant, or perhaps to discover more realistic ones. Reproducing the seasonal cycle of tropospheric O3 at different locations, and its temporal evolution, will provide useful benchmarks of model performance leading to improved causal attribution and tropospheric O3 predictions as the atmosphere is progressively more affected by climate change. The use of tracers in long multiannual simulations, representing CO emissions over different regions, or the stratospheric ozone flux into the troposphere, for example, can be used to discriminate between O3 changes due to emissions or transport from the stratosphere. Runs with fixed emissions or meteorology can be used to tease out the effects of emissions versus transport pathway changes. As a first step, it is planned to examine the evolution in the seasonal cycle over the 60 years of model simulations performed as part of the Chemistry-Climate Model Initiative (see http://www.igacproject.org/CCMI ).
 This observed shift in the O3 seasonal cycle has implications for O3 pollution control policies based on exceedances above ambient ozone concentration thresholds impacting human health and vegetation (e.g., 35 and 40 ppbv, respectively, in Europe [Royal Society, 2008]). Populations may be exposed to higher concentrations for a greater time during the year resulting in increased health effects, and crops and forests may be susceptible to damage by higher concentrations in the early part of the growing season.
 The authors are grateful to P.G. Simmonds and T.G Spain for providing the Mace Head data, to A.J. Manning for sorting the Mace Head data into baseline and nonbaseline observations, to U.S. National Park Service for providing Lassen NP data. D. Parrish acknowledges support from NOAA's Climate Change and Health of the Atmosphere Programs. K.S. Law acknowledges support from the European Union Framework 7 (FP7) IP project PEGASOS (FP7-ENV-2010/265148). We acknowledge the free use of tropospheric NO2 column data from the GOME and SCIAMACHY sensors from www.temis.nl.