Continuous measurements of total column ozone (TOC) over the continental United States began in the early 1960s. Individual measurements were also made earlier at some sites, but more sporadically and mainly in conjunction with the International Geophysical Year 1957.

The backbone of the World Meteorological Organizations (WMO) ozone monitoring network is the Dobson ozone spectrophotometer, an instrument developed in the 1920s specifically for high-accuracy measurements of total column ozone (e.g., Dobson, 1957, 1968). The concept of the Dobson measurement is the differential absorption of ozone at selected wavelengths in the solar ultraviolet spectrum. Two pairs, where one spectral range absorbs light more strongly than the other, are combined to minimize the effect of aerosol interference on the measurements. Measurements made using the direct solar beam are used to determine the TOC based on Lambert–Beer law, while measurements made with the scattered light from the zenith are converted to a TOC value based on the statistics of quasi-simultaneous measurements of both direct sun and zenith.

Operational instrument calibration is maintained by monthly tests with reference and discharge lamps, plus regular intercomparison with two standard instruments: D083 (world primary standard) or D065 (world secondary standard). The calibration of the primary standard is maintained by Langley Plot Campaigns at the National Oceanic and Atmospheric Administration's Earth System Research Laboratory Mauna Loa Observatory (Hawaii).

In the contiguous USA, column ozone has been measured routinely at five observational sites (Bismarck, ND; Boulder, CO; Caribou, ME; Wallops Island, VA; Nashville, TN) since the 1960s. In this study we analyze the total ozone records from these five sites spanning from the 1960s through 2012. A detailed overview on geographical location and information on record length, data completeness and time series properties of the individual station records is provided in Fig. 1.

There is now a broad consensus that long-term trends in total ozone are driven primarily by changes in the atmospheric concentration of ODSs (WMO, 2011). Nevertheless, active research in the field has shown that besides ODSs, several other processes have significant influence on total ozone changes and variability. The 11-year solar cycle, the quasi-biennial oscillation (QBO) and volcanic eruptions are among the most prominent explanatory variables often used to describe the influence of atmospheric variability on column ozone (WMO, 2011). At midlatitudes other dynamical features also show a significant influence on column ozone on seasonal and interannual timescales. In particular, synoptic-scale meteorological variability, described by, for example, the NAO (e.g., Appenzeller et al., 2000; Orsolini and Doblas-Reyes, 2003; Rieder et al., 2010a, 2011; Frossard et al., 2013) and climate modes such as ENSO (e.g., Rieder et al., 2010a, 2013; Brönnimann et al., 2004), has been shown to significantly influence ozone variability and trends.

In the present study we focus particularly on the influence of atmospheric dynamics on variability and trends in column ozone over the USA. To this aim we use a set of indices describing ENSO and NAO modes on a seasonal basis in the statistical analysis. For ENSO we are using the seasonal NINO3.4 index provided by NOAA's Climate Prediction Center – available at http://www.cpc.ncep.noaa.gov/data/indices/3mth.nino34.81-10.ascii.txt. For the NAO we are using the PC-based index (time series of the principal components (PC) of the leading empirically determined orthogonal function (EOF) of seasonal sea level pressure anomalies over the Atlantic sector (20∘ N–80∘ N, 90∘ W–40∘ E)) provided by the NCAR/UCAR climate data center – available at https://climatedataguide.ucar.edu/climate-data/hurrell-north-atlantic-oscillation-nao-index-pc-based.

Recent work has introduced concepts of statistical extreme value theory (EVT) into the field of total ozone research (Rieder et al., 2010a, b, 2011, 2013; Frossard et al., 2013; Fitzka et al., 2014). Here we build on these methodologies to analyze events of extremely low and high ozone (termed ELOs and EHOs, respectively) in the US long-term total ozone records.

The generalized Pareto distribution (GPD) is a commonly used distribution in the framework of extreme value theory (e.g., Davison and Smith, 1990; Ribatet et al., 2009) because it arises as the natural distribution for the exceedance of a random variable (here total ozone) over a threshold. Below we briefly describe the modeling procedure for values over a threshold. Note that the modeling for values below a threshold is precisely the same; the only thing to be done is to negate the values and apply exactly the same procedure as for values above a high threshold as min(xi) = -max(-xi).

The GPD, which is the limiting distribution of exceedances over a threshold, is defined as Fx=1-1+ξx-uσ-1ξ,σ>0,x>u,1+ξx-uσ>0, where x are daily data (here total ozone), u is the threshold value, and σ and ξ are the scale (a measure of the spread of the distribution of x) and shape (which is determining the shape of the distribution rather than shifting it, as u does, or shrinking/stretching it, as σ does) parameters, respectively.

Threshold values u are determined on a monthly basis and interpolated to daily values following the procedure described by Rieder et al. (2010b). Thresholds for ELOs and EHOs, as well as long-term monthly mean values for the five US sites, are shown in Fig. 2.

Here the well-known seasonal cycle with ozone minima in fall and maxima in spring, as well as the latitudinal dependence of total ozone mean values and thresholds (i.e., higher TOC at northern sites (Bismarck and Caribou) due to transport of ozone-rich air from high latitudes), is visible.

Following Rieder et al. (2010b), total ozone observations at the individual US sites are categorized into three groups (see Eqs. 2–4): ELO=xt:xt<uLOWt,EHO=xt:xt>uHIGHt,NEO=xt:uLOWt≤xt≤uHIGH(t), where x(t) is the column ozone amount at a given day; uLOW(t) and uHIGH(t) are the thresholds for low and high ozone on a given day; and ELO, EHO, and NEO denote days with low, high, and non-extreme ozone, respectively. The frequency of low- and high-ozone events is discussed below in the context of dynamical features (see Sect. 4.1).

Seasonal trend decomposition of time series based on LOESS (LOcally wEighted Scatterplot Smoothing) (e.g., Cleveland et al., 1990) decomposes a data record (here ozone) into seasonality, trend, and residual components. When applied to the US column ozone data it returns a well-known picture: (i) a strong seasonal cycle with maxima in spring and minima in winter/fall, in accordance with the understanding of the influence of the Brewer–Dobson circulation on column ozone (transport of ozone-rich air towards northern midlatitudes during boreal winter); (ii) a negative trend component dominated by the influence of ODSs on column ozone; and (iii) a highly variable residual component representing the effects of local-scale meteorology on column ozone. An example of an STL decomposition for the site in Boulder, CO, is shown in Fig. S1 in the Supplement of this article. For a more detailed description of the STL procedure we refer the interested reader to the paper of Cleveland et al. (1990), which describes the method.

STL-trend components represent smoothed TOC residuals after the seasonal cycle is removed and are thus not reliable measures for TOC trend analysis. Statistical trend analysis in this manuscript is solely based on linear regression analysis (see Sect. 4.4). Here we utilize STL because the resulting trend component provides a more detailed picture of the seasonal and interannual variability in the overall TOC time series compared to, for example, a simple linear trend component, and thus is suitable for the secondary assessment of “fingerprints” of NAO and ENSO events as long-term time series variability is preserved.

In Sect. 3.1 we described the classification of total ozone observations into days with extremely low, extremely high, and non-extreme ozone. Here we focus on the frequency distribution of the extremes and analyze the influence of ENSO and NAO events on column ozone at the five US ozone monitoring sites. As the general features are very similar among the individual sites, we mainly show the results for Boulder and Caribou, which give the envelope of column ozone observations over the USA, in the main body of the paper. For convenient reference, illustrations for other sites are available in the Supplement for this article.

In Figs. 3 and 4 we plot the observed frequency of ELOs, NEOs, and EHOs as time series for Boulder and Caribou (the results for the remaining sites are shown in Figs. S2–S4). Next, we turn to fingerprints of atmospheric dynamics in the frequency distribution of EHOs and ELOs at these sites.

First, we turn to the North Atlantic Oscillation, the leading mode in the Atlantic sector, which influences the direction and intensity of the tropospheric jet stream (e.g., Orsolini and Limpasuvan, 2001) and represents the main driver of the interannual variability in storm tracks during the cold season (e.g., Lau, 1988). A positive NAO phase leads to lower ozone over Europe and the USA and higher ozone over the Labrador Sea and Greenland, and vice versa for a negative NAO phase (see Rieder et al., 2010b, and references therein, as well as Frossard et al., 2013, for a spatial representation of NAO influence on column ozone at northern midlatitudes).

The North Atlantic Oscillation in its negative phase (NAO index < -1, marked with blue (winter) and light blue (spring) dots in Figs. 3 and 4) leads to higher column ozone over the United States during winter and spring. Conversely, a fingerprint of the NAO in its positive phase (NAO index > 1, marked with red (winter) and orange (spring) dots in Figs. 3 and 4) is seen as lower ozone over the USA in the individual station records. Over the study period (1963–2012), the NAO was in a strongly positive phase (NAO index > 1) 16 times during winter and 12 times during spring. While all but one of the wintertime events are captured in the frequency of ELOs at multiple sites, 2 out of the 12 spring events remain undetected (1982 and 1994) at any of the five US sites (see Fig. 6). The correlation analysis confirms the relationship between column ozone and the NAO phase (see Table 1).

In both winter and spring the NAO index correlates negatively with the frequency of EHOs and positively with the frequency of ELOs, indicating an increase (decrease) in the frequency of high-ozone events during a negative (positive) NAO phase and vice versa for low-ozone events, manifested also in the seasonal means.

It has been noted that the NAO has tended towards a more positive phase in recent decades (e.g., Hurrell, 1995; Thompson and Wallace, 2000), concomitant with a strengthening of the northern polar vortex. Nevertheless, a strongly negative NAO phase (NAO index < -1) is found 10 times during winter and 5 times during spring (roughly half the rate of positive phase events) in 1963–2012. Out of the 10 wintertime events, 8 are discernible in the frequency distribution of extreme total ozone, with the winters of 1973/74 and 2010/11 being absent. The missing NAO fingerprint in winter 2010/11 is not surprising given the unusual dynamical conditions in the Arctic in this year, which led to a particularly strong stratospheric polar vortex and record Arctic ozone losses (e.g., Manney et al., 2011), counteracting the dynamic enhancement of column ozone due to the NAO. Out of the five springtime negative NAO events, three can be identified in (most of) the five US monitoring sites, with 2005 and 2006 missing. As was the case in 2011, the particularly cold conditions in the Arctic spring of 2005 likely contribute to the missing NAO fingerprint. Differences in the occurrence and detection frequency of NAO fingerprints are captured in the correlation analysis, with overall stronger correlations during winter than spring. This is further explored in Fig. 7, which shows the fraction of EHOs and ELOs during wintertime positive and negative NAO events: more (fewer) EHOs during NAO- (NAO+) phases and vice versa for ELOs.

The NAO fingerprints identified in the US column ozone records are in broad agreement with those for European sites and satellite data. Appenzeller et al. (2000) were among the first to report on the influence of the NAO on column ozone over Europe, based on their analysis of the world's longest total ozone record: Arosa, Switzerland. Rieder et al. (2010a) extended these investigations toward low and high ozone values and Rieder et al. (2011) documented the influence of the NAO in its positive (reduced column ozone, reduced frequency of high-ozone events) and negative (increased column ozone, increased frequency of high-ozone events) phases for five European ground-based sites in 1970–2010. These authors report a similar number of detected fingerprints and occasional misses at individual sites due to local effects. Frossard et al. (2013) extended investigations to larger spatial scales by analyzing the NIWA assimilated total ozone data set in 1979–2007. These authors report that the fingerprint of the NAO is of similar spatial extent for both mean values and ozone extremes but that the magnitude of influence on total ozone is larger for extremes than mean values. These results are in broad agreement with those presented here for the US long-term ozone records, documenting the significant influence of the NAO on column ozone variability throughout northern midlatitudes.

Next we turn to the El Niño–Southern Oscillation. Warm ENSO events are triggered by a high contrast between tropical and extratropical Pacific sea-surface temperatures, which are known to affect midlatitudes (in particular the North Pacific) via changes in the Hadley cell and Rossby wave generation (e.g., Trenberth, 1998; Alexander et al., 2002). During warm ENSO events, the meridional circulation in the stratosphere leads to enhanced ozone transport from the tropics to middle and high latitudes and a warmer lower stratosphere, both of which tend to increase midlatitude ozone (Rieder et al., 2013, and references therein).

The warm ENSO phase (El Niño, NINO3.4 index > 0.7) is, as expected, associated with higher ozone over the USA during winter/spring, visible in the frequency distribution of the extremes. During the study period, moderate to strongly positive ENSO events were recorded 11 times during winter and 4 times during spring. Most wintertime events (except those in 1983, 1992, and 1995) and springtime events (except 1983 and 1992) can be identified in the frequency distribution of ozone extremes. The absence of ENSO fingerprints in the remaining 3 years is consistent with their occurrence immediately after the two major volcanic eruptions of the last century (El Chichón in 1982 and Mt. Pinatubo in 1991), when the effects of the volcanic eruptions (enhanced ozone depletion on sulfate aerosols) would have masked the dynamical signal. As was the case for the NAO, the ENSO results for the US sites are in good agreement with findings for European sites and satellite data (e.g., Rieder et al., 2010a, 2013), illustrating the importance of ENSO in modulating column ozone at northern midlatitudes. The correlation analysis between ENSO and column ozone (or the frequency of EHOs and ELOs) is less conclusive then for the NAO, probably because of the rather small number of strong ENSO events.

At all sites we find a more consistent presence of fingerprints of NAO and ENSO in extreme values of column ozone (upper panels in Figs. 3 and 4) than in its mean values (lower panels of Figs. 3 and 4). While the extremes show a pronounced response (increasing or decreasing frequency) to the prevailing ENSO and NAO phases, the mean values often do not show large differences compared to neighboring years without ENSO or NAO events. This is particularly evident for NAO+ events, where about twice as many events were detected in the frequency distribution of the extremes than in the seasonal mean values.

Here we contrast the findings of the EVT-based analysis with results from the STL-decomposition approach. In Fig. 5 we show the anomaly of the trend components of the STL decomposition for the two selected US sites – Boulder and Caribou – as above in the EVT analysis (the results for the remaining sites are shown in Fig. S5). While the overall trend curves show a steady decline that is most pronounced in the 1980s and 1990s, as expected from the strong negative influence of ODSs on column ozone (e.g., WMO, 2011), there is also a large degree of interannual variability in these curves. This variability is not related to seasonality in the ozone field, since the seasonal component has been removed from the data prior to the trend computations within the STL procedure.

As for the EVT analysis we now identify fingerprints of ENSO and NAO events in the STL trend component. The colored vertical bars in Fig. 5a and c mark positive and negative winter- and springtime NAO events. The analysis of the STL trend component shows that positive NAO events are associated with a decreasing tendency in the trend curve, and thus with lower column ozone, while negative NAO events are associated with an increasing tendency in the STL trend component, and thus with higher column ozone. Figure 5b and d show the corresponding results for warm ENSO events, which tend to enhance column ozone. The good agreement between the results in Fig. 5 with those from the EVT analysis (Figs. 3 and 4) provides further evidence of the significant influence of strong NAO and ENSO events on column ozone variability over the continental USA.

Despite the overall similarity in trends and patterns of variability, it is important to note that fingerprints of individual NAO and ENSO events are not always found at all five stations analyzed. Figure 6 provides a summary of all major ENSO and NAO events over the 1963–2012 time period and their detection (or absence) in the individual station records. Solid squares in Fig. 6 mark fingerprints detected, while open squares mark “absent” fingerprints at individual sites. The majority of ENSO and NAO events are detected at all five US total ozone monitoring sites, but some individual events are not discernible at individual (or multiple) sites such as the negative NAO event of spring 1996. The absence of individual fingerprints is not too surprising given the large spatial distance between individual sites and their regional location (see Fig. 1). The occasional masking of large-scale ozone variability by localized synoptic-scale meteorology (e.g., the influence of the subtropical jets and localized tropopause variations) is associated with the regional patterns of advection and convergence or divergence that are related to changes in tropospheric and stratospheric pressure systems as has been previously reported for regions other than the USA (e.g., Koch et al., 2005; Mäder et al., 2007; Wohltmann et al., 2007).

Direct correlations of daily TOC between sites are rather inconclusive due to the difficulty in accounting for local meteorological effects at a station or temporal lags between stations due to transport. Unfortunately, vertical investigations are limited by the absence of vertically resolved ozone profiles at most of the stations (except for Boulder, CO). In addition, seasonal comparisons between years with fingerprints and without are restricted to a small sample size (i.e., a few missing fingerprints on a site basis). Nevertheless, a comparison of cumulative distribution functions (CDFs) on a site basis between neighboring years with and without fingerprints reveals the absence of high- or low-ozone events associated with the NAO or ENSO (see Fig. S6 in the Supplement). Thus, instead of individual effects, we quantify the overall contribution of extremes to seasonal mean column ozone by calculating the influence of ELOs and EHOs at each site.

Several studies have linked the occurrence of multiple tropopauses to Rossby wave breaking events along the subtropical jet (Homeyer and Bowman, 2013, and references therein) and to associated tropospheric intrusions (e.g., Pan et al., 2009). Climatological maxima in multiple tropopause occurrence have been linked to observed changes in vertical profiles of satellite-observed trace gases that are consistent with air from the tropical tropopause layer being drawn into the region between the two tropopauses; specifically, climatological ozone mixing ratios in midlatitude multiple tropopause regions are substantially lower than those in regions with a single tropopause (Schwartz et al., 2015). Schwartz et al. (2015) estimated that in Northern Hemisphere winter midlatitudes, when multiple tropopauses are most common, climatological ozone values can be as much as 20 % lower than they would be without multiple tropopauses.

These results are consistent with the observed association of lower column ozone with multiple tropopauses (e.g., Castanheira et al., 2012; Mateos et al., 2014). Mateos et al. (2014) also noted more common occurrence of such tropospheric intrusion events during NAO positive phases, suggesting a role for dynamical modes such as NAO and ENSO in modulating multiple tropopause occurrence and thus their corresponding effects on ozone.

In addition, there is a maximum in multiple tropopause occurrence frequency over the USA in winter and spring, extending poleward from the region where upper tropospheric jets are most common (Manney et al., 2014). Boulder, Nashville, and Wallops Island are near the latitude of maximum multiple tropopause occurrence just poleward of the subtropical upper tropospheric jet, while Bismarck and Caribou are at the northern edge of the region of enhanced multiple tropopause activity (Manney et al., 2014) and are thus less frequently affected by processes in multiple tropopause regions.

The absence of individual fingerprints on a site basis and their underlying cause is of general interest but beyond the spatial and climatological scope of the presented study. Nevertheless, further analysis (including vertical information from sounding profiles) is suggested for future site-specific analysis addressing effects of local dynamics on column ozone variability.

In this section we turn the focus to column ozone trends at the five US Dobson sites.

To analyze the influence of extremes (both low and high) on ozone trends we contrast linear trends for the entire observational time series (i.e., all observational data included) with trends for time series with extremes removed. We focus on two main time periods: 1970–1995, with almost linearly increasing ODSs, which includes the peak in ozone depletion (following the Mt. Pinatubo eruption), and 1996–2010, which extends from the maximum in ODSs (∼ 1996/1997) to current conditions. Results for each site during winter and spring are given in Table 2. During the 1970–1995 period, with almost linearly increasing ODSs and significant midlatitude ozone losses in the early 1980s and 1990s (following the El Chichón and Mt. Pinatubo eruptions), ozone trends vary between -2.8 and -4.8 % per decade among the sites and seasons (Table 2). All sites except Caribou show larger negative trends in spring than in winter, consistent with results from European midlatitude sites (e.g., Rieder et al., 2010b, 2011). We argue that this qualitative difference between Caribou and the other US sites is determined by geography. Caribou is the northernmost US monitoring site and is thus more frequently affected by transport of air masses out of the Arctic polar regions in winter and spring than the other stations; such Arctic air may, in particularly cold winters, carry the signature of chemical ozone depletion. The more southerly sites are usually most strongly influenced by midlatitude ozone-rich air masses (e.g., Manney et al., 2014), though they may also show effects of transport of low-ozone air from low latitudes and accompanying troposphere-to-stratosphere exchange.

Comparing the entire observational records with those with extremes removed, we find that trends are only about half as strong in the latter case. This is particularly interesting as no statistically significant trend (at a 95 % level) is found for the magnitude of EHOs or ELOs over 1970–1995. The individual time series show the well-known pattern of large interannual variability but no robust increase (or decrease) in the average magnitude of the extremes themselves. Thus the influence of extremes on seasonal mean column ozone (see below) can be understood as a function of their occurrence frequency, driven by chemical ozone depletion and dynamics.

Turning now to the more recent past, i.e., 1996–2010 (Table 2), we find positive trends at most sites, an anticipated result since stratospheric chemistry in this period is impacted by slowly but steadily declining ODSs. The key interest in the trends for 1996–2010 is thus not the sign of the trends but their significance. Observational and modeling studies suggest that chemical ozone depletion ceased to increase around the turn of the century (e.g., WMO, 2011), but whether significant ozone recovery has started is still undetermined. Positive trends at the majority of sites indicate that ozone has stopped declining over the USA, particularly during winter, suggesting that chemical depletion may have ceased (Table 2). Nevertheless, since the trend estimates over the 15-year period of 1996–2010 are not significant at the 95 % level (see p values in Table 2), there is no clear evidence that significant ozone recovery has started yet. As was the case for 1970–1995, the trends are much smaller (by a factor of 2–3) if extremes are removed from the records. Again we investigate whether significant changes occurred in the extremity of ELOs and EHOs, and, as for the 1970–1995 period, we find large interannual variability in the magnitude of lows and highs (driven by dynamics and chemistry) but no significant trends at rigorous test levels (i.e., 95 %).

Discriminating the effects of the individual dynamical proxies on column ozone is difficult because (i) fingerprints for multiple proxies are found in several years (e.g., a strongly positive NAO and a warm ENSO phase) and (ii) the occurrence frequency of the individual fingerprints is highly variable. Also correlations between sites are rather noisy on daily timescales (local effects) and seasonal comparisons between years with fingerprints and without are restricted to a small sample size (i.e., too few missing fingerprints on a site basis). Thus, instead of individual effects, we quantify the overall contribution of extremes to seasonal mean column ozone by calculating the influence of ELOs and EHOs at each site:

IELOs=Ms-MsELOexMs⋅100,IEHOs=Ms-MsEHOexMs⋅100, where IELOs(IEHOs) is the influence of extremely low (high) total ozone on seasonal mean column ozone (Ms) in percent, and MsELOex(MsEHOex) is the seasonally averaged column ozone with ELOs (EHOs) excluded from the time series. In Figs. 8 and 9 we show the influence of ELOs and EHOs on winter and spring column ozone, respectively. While EHOs are the dominant influence in the early and late parts of the station records, the time period from 1980 to 2000 is dominated by the influence of ELOs, consistent with the nearly linear increase in ODSs and their importance to column ozone changes in this time period. During this period of strong ozone depletion, however, individual years still show a net positive effect of the extremes on seasonal mean column ozone, highlighting the importance of dynamical factors, such as warm ENSO events (e.g., spring 1986 and 1998), on column ozone variability. The influence of ELOs and EHOs on seasonal mean ozone is bounded by about ±5 %, thus about as large as the overall long-term trend values given in Table 2.

Next we analyze the pattern correlation of the net contribution of the extremes (i.e., IELOs+IEHOs) among individual sites. The seasonal pattern correlations among individual sites are summarized in Table 3. Pattern correlations are highest for neighboring sites, i.e., Boulder–Bismarck and Nashville–Wallops Island. Caribou is again an exception in this respect, with a seasonally dependent correlation with the other sites. During winter, the correlation at Caribou is highest with the eastern sites (Nashville and Wallops Island), while during spring the correlation is highest with the western sites (Bismarck and Boulder), suggesting that Caribou is under the influence of the same air masses as the eastern/western sites during different seasons. Examinations of the overall correlations for each decade of the record (not shown here) indicates that the correlation between Boulder and Nashville (particularly in winter) increases with time (i.e., it is higher in the 1990s and 2000s than in the 1970s and 1980s), while the correlation between Nashville and Wallops Island slightly decreased in recent decades (in both winter and spring), consistent with Boulder and Nashville being more frequently influenced by similar air masses in recent years. Recent operational changes at Caribou, resulting in reduced sampling frequency, significantly affect the correlation between Caribou and Wallops Island, thus highlighting the importance of continuous and frequent ozone observations for both trend analysis and assessment of relationships between measurements at different sites.