The Zugspitze site of the former Fraunhofer Institute for Atmospheric Environmental Research (Fraunhofer-Institut für Atmosphärische Umweltforschung, IFU; now the Karlsruher Institut für Technologie, Institut für Meteorologie und Klimaforschung, IMK-IFU) is located at the northern rim of the Alps (47.421∘ N, 10.986∘ E). A detailed description of the topography was given by Reiter et al. (1986). The observations of trace gases, meteorological parameters, and 7Be were made at the summit of Zugspitze (2962 ma.s.l.) or temporarily close to the summit at 2932 m (1989–1994 or during a shorter period, depending on the parameter of the measurement programme). Not all the quantities have been measured over the entire period since 1970; some measurements were abandoned earlier, such as SO2 or CO2. The data are available in the data archive of the IMK-IFU as half-hour averages and supplemented by statistical products such as daily, monthly, and annual means. For the work presented here, we use ozone, 7Be, RH, and carbon monoxide. We completed the data archive for the Zugspitze summit until the end of 2011, based on files with evaluated data found on Hans-Eckhart Scheel's computer. No evaluated version of the data for the final year, 2012, was found. The revised scientific analysis is based on the methods for identifying stratospheric intrusions described by Trickl et al. (2010).

Since 2001, atmospheric in situ measurements have taken place at the Global Atmosphere Watch (GAW) observatory at the UFS research station on the southern face of Zugspitze, which is operated by the German Umweltbundesamt (UBA, i.e. Federal Environment Agency; 47.417∘ N, 10.979∘ E; air inlet at 2671 ma.s.l.). The calibration of the UBA instrumentation is routinely performed and verified as part of the GAW quality assurance standards. The instruments are controlled daily and serviced on all regular workdays.

Due to the small altitude difference, in principle, the stratospheric influence at UFS should not differ much from that at the summit. However, one should consider that, during the warm season, the boundary layer formation may prevent intrusions from descending by much below 3000 m. Thus, even a local orographic influence can be significant and lead to differences due to different upslope winds advecting the air from lower altitudes (Sect. ). However, the trends should behave similarly, and the results for UFS can serve as a tool to extrapolate the results for the summit.

Trajectories have been used to verify deep stratospheric intrusions. Many case studies have been made with FLEXTRA trajectories and the particle dispersion model FLEXPART (Stohl and Trickl, 1999; Trickl et al., 2003, 2010, 2011, 2014). ETH Zürich provided 4 d forecasts on a daily basis, using the LAGRANTO model (Wernli, 1997; Wernli and Davies, 1997), until the end of the ozone lidar measurements in early 2019 (e.g. Zanis et al., 2003; Trickl et al., 2010, 2020a). In cases with subsidence periods exceeding 4 d, HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory; Draxler and Hess, 1998; Stein et al., 2015; https://www.ready.noaa.gov/HYSPLIT_traj.php, last access: 23 July 2023) 315 h backward trajectories were calculated. We prefer the “reanalysis” mode that, despite having a moderate spatial resolution, has better explained our observations (e.g. Trickl et al., 2020a).

Linear regression analyses were done by applying a program developed for spectroscopic studies (e.g. Trickl and Wanner, 1984; Trickl et al., 1993, 1995). This program includes individual weights for the data points, error propagation, and recalibration of the standard deviations based on a χ2 analysis. In addition, more than 20 programs were developed to convert data formats, to fill gaps in the data archive, to calculate the percentiles of 7Be and RH, and for data filtering. For sliding arithmetic averages, we used the tool provided by the Origin graphics program to prepare the figures. This tool mostly generates rather realistic boundary conditions for the smoothing of the data set at both ends.

A clear rise in the ozone level during the relevant time period is a good indicator of stratospheric air. The intrusion in Fig.  is a good example. However, pronounced ozone peaks are not always the case. There are several factors that can lead to less pronounced signatures. Of course, aged intrusions undergo mixing with tropospheric air, which could be verified by humidity measurements by the UFS water vapour lidar or radiosondes. As mentioned, we found that just the lowermost layer above the tropopause starts to subside (Fig. 18 in Trickl et al., 2014). This is mostly the case in winter and sometimes leads to just a small rise in ozone (Trickl et al., 2020a). Finally, there are cases in which just an edge of the intrusion layer hit the station, which can be associated with lower peak ozone. Small rises in ozone in the Zugspitze time series are hard to distinguish from the mostly rather variable concentrations around an intrusion period. Thus, we exclude ozone from the list of parameters used for identifying STT. This is justified by the thorough analysis of Trickl et al. (2010).

As discussed by Parrish et al. (2020), there is just a small average relative decrease of 2.6 % between the ozone values at the summit and the lower-lying UFS station. Therefore, we use the UFS data to extend the time series beyond 2011. The question is how well the results for the stratospheric influence at UFS match those for the summit during the period of overlap. In any case, even if there is a discrepancy, then at least the trend for the extended period of observation can be judged.

Fabian and Pruchniewicz (1977) published ozone values for the period 1970 to 1977. These values were rejected by Tarasick et al. (2019) and are not included in our analysis. Instead, we use an extrapolated constant ozone mixing ratio for the period 1970 to 1977, as justified in Sect. .

In general, short ozone data gaps are filled during the analysis by the respective monthly average.

Carbon monoxide has both natural and anthropogenic sources (e.g. Duncan et al., 2007) and is of importance in tropospheric chemistry because of its reaction with the OH radical (Logan et al., 1981). The Zugspitze CO mixing ratio displays a pronounced seasonal cycle, with a maximum around April and a minimum from summer to autumn.

CO has been used to identify polluted air masses (Scheel, 2003). It is interesting that the highest CO mixing ratios have been observed in fronts (Rudolf Sládković, personal communication, around 1996). A strong rise in the CO level in a front is frequently associated with polluted air picked up over industrial regions to the northwest. In Fig. , this frontal passage took place on 20 July 2001 but with just a moderate increase in CO and NOy.

As confirmed in Fig. , CO is not a good tracer of stratospheric air hitting the summit. Trickl et al. (2014, 2016) explain this by pointing to the fact that intrusions emerge from the lowermost layer of the stratosphere, where obviously the descent to stratospheric values (20 to 40 ppb; e.g. Zahn et al., 1999; Fischer et al., 2000; Pan et al., 2004; Hegglin et al., 2009; Vogel et al., 2011) is not yet pronounced.

Trickl et al. (2014) determined a very small positive CO trend in intrusions for 1900 to 2005. The CO trend outside the intrusions during that period is slightly negative, indicating an improving air quality.

Here, we repeat the analysis based on one of the revised filtering criteria and extend it to 2020 by including the UFS data that start in 2009.

As strongly elevated values of the 7Be activity concentration (mBqm-3) are indicators of stratospheric intrusions (Reiter et al., 1983), this tracer has been important for flagging ozone data with respect to stratospheric influence. Its half-life time of 53.42 d ± 0.1 d (Huh and Liu, 2000) is rather suitable for studies of STT.

The 7Be threshold used at IFU in studies until 2000 is 8 mBqm-3 (Sládkovič and Munzert, 1990), which was adopted by Stohl et al. (2000). Elbern et al. (1997), in their analysis of Wank and Zugspitze, applied variable thresholds, given by a predefined increase in the standard deviation of the values against the running monthly mean, for the species used. In a study of stratospheric intrusions at Monte Cimone (Italy), Cristofanelli et al. (2006) employed both the fixed value of 8 mBqm-3 and a dynamic threshold based on running monthly means. By comparing trajectory-based predictions of stratospheric air intrusions for the period 2001–2005, Trickl et al. (2010) showed that the 7Be criterion could be weakened to a threshold of 5.5 mBqm-3 or perhaps less during that period.

There are two drawbacks that limit the specificity of 7Be for STT. As mentioned, one is the small concentration of the isotope in intrusions, which necessitates sampling over 24 h in the apparatus used. Second, 7Be is not only produced in the stratosphere but also in the upper troposphere, with an estimated global average contribution of about 33 % (Table 3 in Lal and Peters, 1967). This fraction does not apply for our latitude of about 47.5∘ N, where just 23.4 % is obtained from Fig. 16 found in Lal and Peters (1967). It is even smaller at the higher altitudes of typical source regions relevant for the observations of descending stratospheric air at the northern rim of the Alps. For example, from the same figure, we estimate a constant tropospheric fraction of just 10 % at 70∘ N and higher latitudes. This low fraction is highly advantageous for our analysis, since it reduces its uncertainty.

The primary production mechanism of stratospheric tracers such as 7Be, 10Be, and 14C is spallation or neutron capture by cosmic rays or solar wind (Lal and Peters, 1967; Herbst et al., 2017). The atmospheric production of these isotopes is modulated by the solar magnetic field, solar wind, and the geomagnetic field strength. For the period covered here, the influence of nuclear testing can be ruled out, since the last atmospheric nuclear test took place on 29 September 1969 (in China; https://en.wikipedia.org/wiki/Nuclear_weapons_testing, last access: 23 July 2023), given the short lifetime of the 7Be isotope.

Figure  shows the full 7Be time series from 1970 to 2006 and the gliding 90 and 365 d arithmetic averages. An almost steady increase is visible after 1976. Reiter (1973a b, 1979) and Reiter and Littfaß (1977) point out the importance of solar flares in the 7Be data. However, the averages shown are obviously not strongly correlated with the frequency of solar flares (https://www.ngdc.noaa.gov/stp/space-weather/solar-data/solar-features/solar-flares/index/flare-index/, last access: 23 July 2023) and annual sunspot numbers (http://www.sidc.be/silso/DATA/SN_y_tot_V2.0.txt, last access: 23 July 2023), which are also displayed in Fig. . The increase in 7Be since 1977 has not correlated with the frequency of solar flares or the number of sunspots per year. It seems that the solar activity decreases in the new millennium, but 7Be does not diminish. We conclude that the observed 7Be values are reasonable proxies for STT. However, it makes sense to use an additional tracer such as RH for identifying intrusions.

Indeed, Herbst et al. (2017; Fig. A1) found that the global production of 10Be in the atmosphere for 1960 to 2015 predominantly consisted of a constant term plus an 11-year modulation, with an amplitude of just roughly 20 % of the constant level.

Since 7Be is attached to aerosols, it is subject to washout, which might mask the original stratospheric signature. However, Reiter et al. (1971) found that washout is relatively small for air directly transported downward to 3000 m from the tropopause due to clear-air conditions nicknamed “7Be weather” (Eisele et al., 1999). For low-lying stations, the tropospheric lifetime is estimated as 35 d, including washout (Bleichroth, 1978).

The role of the attachment of 7Be by the persistent aerosol layer in the lower stratosphere or the tropopause region could have been particularly pronounced between 1980 and 2000 due to major volcanic eruptions (Jäger, 2005; Trickl et al., 2013). However, in Fig. , there is no evident positive correlation of the 7Be data with the extreme eruptions of El Chichón (1982) and Mount Pinatubo (1991). It is unclear where the 7Be atoms are attached to aerosol.

The seasonal cycle may be influenced by the weather conditions. A good example is the year 1970, with a winter minimum of 7Be. That winter was characterized by never-ending snowfall, resulting in 4 m of snow in a neighbouring valley by the beginning of spring.

In Fig. , we show four examples of annual distributions of 7Be at time intervals of 10 years. The daily data are reorganized in time, according to the growing specific activity, as necessary for calculating percentiles. For all 4 years, the maximum number of measurement days is less than 365 or 366 d, as reflected by the changing position of the rise to the highest values. The curves are rather smooth. For later years, their slopes become steeper than at the beginning of the measurements in the 1970s. Based on the above information on the stratospheric fraction of 7Be, we also mark the positions of 80 % and 90 % of 7Be during the respective years, which is needed for the analysis described in Sects.  and .

Figure  shows the series of annual percentiles for the entire period from 1970 to 2006. The highest values, also representing the highest specific activities in single intrusions, change as a function of time. This behaviour is also seen for smaller values. Thus, it is reasonable to use the percentiles as thresholds for the stratospheric origin of an air mass. Scheel (2003) already replaced the 8 mBqm-3 threshold with the 85th percentile. We now use the 65th percentile, approximately corresponding to 5.5 mBqm-3, right after the year 2000.

Stratospheric air may arrive at the Zugspitze summit with much longer travel times than the 2 to 15 d found for detectable intrusions in our previous analyses. This component can no longer deliver a similarly clear signature in the data, with the exception of 7Be (Fig. ). For the longer travel times, in principle, the limited lifetime of the isotope must be taken into consideration, but this is a difficult task in the absence of information on the respective transport path and time. In addition, we would, in principle, need to know the stratospheric fraction at the source and not at the receptor site.

The signature of the air influenced by stratospheric influx also comprises a pronounced decrease in the humidity during the respective episode. Stohl et al. (2000) have discussed the advantages and drawbacks of the parameters of specific humidity and relative humidity. In conclusion, RH is preferred, and it may well serve to help identify stratospherically influenced air, at least for a fixed altitude. Indeed, Trickl et al. (2010), based on trajectory forecasts and backward trajectories, found a very high probability of identifying intrusion layers reaching the Zugspitze summit if, additionally, a minimum RH < 30 % was fulfilled in a given layer. However, in principle, it is not an unambiguous stratospheric air tracer. A combination of an RH criterion with a 7Be criterion is the best choice.

Figure  shows monthly percentiles of the Zugspitze RH from 1978 to 2011. An obvious drying of the lower free troposphere during that period is indicated. However, deeper subsidence in more recent years must also be taken into consideration.

The monthly minimum RH can be as low as 1 % but is higher between 1985 and 1997. This suggests the presence of three phases, perhaps with different sensors. As pointed out above, DWD data are used for the period 1970 to 1997. Indeed, according to DWD listings, there was a change in sensors on 13 March 1986 (from a psychrometer to a MIRIAM-TDH system). This suggests that the higher winter minimum in 1985 must be accidental. The step no longer exists for the 5th and higher percentiles, as seen in the averaged percentiles in Fig. .

For the data filtering, we correct the data during the period from 13 March 1986 to the end of 1997 by applying the following formula: 1RHc=RH-ΔRHRH-100ΔRH-1008. ΔRH=5.0 (all in %). The exponent was estimated by a gradual rise from the lower values in order to fulfil the approximate disappearance of RHc above 30 % RH. This formula significantly modifies the RH values for low RH by -0.43 %, -1.26 %, and -3.24 % RH for 30 %, 20 %, and 10 % RH, respectively.

In general, the Zugspitze relative humidity is dominated by high values (mostly 100 %). This amplifies the detection capability for subsiding air masses that are associated with clear-weather conditions. The explanation is the frequent formation of an orographic updraft and cloud formation at and above this isolated high mountain.

Despite this step, it is obvious that there is also a pronounced downward trend at higher percentiles. Figure  shows 25-month averages for several percentiles, roughly representing a 1-year time resolution (e.g. Trickl et al., 2020b).

The humidity measurements at UFS by DWD show higher offsets of low RH. Between 2001 and late 2011, the RH minima at UFS are about 7 % RH (not shown here). Afterwards, the monthly minima are 3 % RH. This suggests the use of a different type of humidity sensor, but we do not have clear information on an instrument change. Since we do not have information on the local RH background at UFS, we do not correct for the 3 % RH offset. However, in Eq. () we apply ΔRH = 4.45 % until 31 July 2011 (Sect. ) in order to shift the RH minima to 3 %. In Fig. , we show the corresponding smoothed RH percentiles for UFS and exclude the years before 2009, where an unrealistic rise in the curves is seen that would cause confusion in the figure. The most important message is that afterwards no trend is observed at all in most curves.

An elevated minimum RH in an intrusion layer seems to indicate the mixing of the stratospheric air with the surrounding tropospheric air during the descent to the Alpine summits. The consequence would be a severe problem for the quantification of the stratospheric component of the Zugspitze ozone. Fortunately, this not the case. In four of our earlier papers (Trickl et al., 2014, 2015, 2016, 2020a), we documented that deep intrusions are much drier than what is measured on average at the Zugspitze sites (see Sect. ). We hypothesized an instrumental wet bias of up to 10 % RH for the dew point mirror instrument that has been used at the summit under dry conditions since 1998. However, the elevated minimum RH values are not restricted to this instrument (Fig. ), which calls for a different explanation.

The lowest minima, down to 1 % to 2 %, mostly occur during the cold season and nighttime. As a matter of fact, the UFS DIAL revealed general deviations from the measurements at the summit during warm and convective conditions (Vogelmann and Trickl, 2008). We conclude that orographic transport takes place in a shallow surface layer (Fig. 5 in Carnuth and Trickl, 2000) that influences the humidity measurements at the summit station and UFS but not in the lidar measurements that probe the humidity outside of this surface layer. The Zugspitze summit could act like a chimney, where directly rising slope winds are at least slightly focussed into a dry stratospheric layer, which could be the reason for the positive bias under warm conditions. Also, the evaporation of moisture from the slopes and terraces around the stations could contribute to the elevated humidity minima.

Not only shallow slope winds can influence the humidity; a daytime upvalley flow (“valley wind”) can also make a contribution that is lifted to altitudes about 1 km above the surrounding summits in the late morning hours (Carnuth and Trickl, 2000; Carnuth et al., 2002; Yuan et al., 2018).

Local mixing of the dry stratospheric air masses with air streaming upward from the boundary layer is much more likely to influence the results at UFS, which might explain the higher minimum values. Yuan et al. (2019) examined the diurnal cycles of CO2 at three Zugspitze sites, UFS, a window in a tunnel above UFS, and the summit station, and found significant differences in the orographic influence. Of course, the probability of an intrusion to overlap fully with the station is lower at UFS than at the summit due to the limited penetration of the dry layers towards lower altitudes.

We conclude that, in most cases, the orographic admixture of humid air at the summit masks the true very low humidity level in the stratospheric layers. Thus, our analysis of STT should be rather realistic.

Lidar measurements at IMK-IFU and UFS have not been made throughout the year and around the clock. However, a large number of intrusion cases have been studied and allow us to draw important conclusions.

A good example of an intrusion case analysed with both the UFS water vapour DIAL and the Zugspitze summit in situ data is given in Figs. 10 and 11 in Trickl et al. (2016). The lidar figure shows the descent of the extremely dry layer exhibiting a stratospheric-type water vapour minimum mixing ratio (about 0 to 50 ppm or RH ≪ 1 %) across the Zugspitze summit, and the station time series verifies the elevated ozone up to 73.3 ppb. The Zugspitze RH minimized at 7.2 %; i.e. it is substantially higher than suggested by the lidar measurements (see Sect. ).

A complete downward passage of an intrusion layer across the summit station was verified with the two DIAL systems in many other measurement series. However, the lidar measurements also show examples of intrusions that do not descend to altitudes below 3000 m, mostly during the warm months. Sometimes, these layers persist over several days at a rather constant altitude, which is typically around that of the Zugspitze summit.

Intrusions do not necessarily lead to a pronounced rise in ozone (Trickl et al., 2014), especially during the cold season (Trickl et al., 2020a). The upper panel in Fig.  shows lidar measurements of ozone from 3 to 7 October 1997 at intervals of 1 h. The intrusion structures below 4 km are not as clearly visible as in other colour-coded images presented by us. The small rise in ozone is verified in the lower panel in Fig.  by the corresponding Zugspitze time series that exhibits just four 60 ppb peaks residing on a 50 ppb background. On 4 October, the overlap with the intrusion is limited, however, resulting in some rise in 7Be. On the following 2 d, 7Be is substantially higher, but the RH minima are just slightly below 30 % (see Sect. ). It is important to mention that the RH minima in the Munich, Hohenpeißenberg, and Stuttgart radiosonde data on these days are much lower and range between 5 and 15 %, sometimes at just slightly higher altitudes. Also, on the other side, to the southwest, the Innsbruck radiosonde verifies much drier conditions. HYSPLIT backward trajectories initiated at the Zugspitze summit show long descents from high altitudes over more than 10 d, in part from Siberia. Mixing with the surrounding tropospheric air is likely during this long travel of the layer and explains why the RH is not smaller than 5 %. Please note that the two most-pronounced RH dips in Fig.  occurred around noon, when the orographic wind system maximizes. This could be the reason for the rather high minimum RH.

Again, we conclude that the intrusion air masses fully hitting the Zugspitze summit are much more stratospheric than suggested by the in situ RH measurements. This was established from many years of comparing the in situ humidity measurements with lidar and sonde profiles. However, a bias of unknown magnitude can be introduced by intrusions just partly overlapping with the summit (i.e. with the humidity minimum located above 3 km). Such a bias cannot be determined from the available data alone. However, these cases are more likely to occur in summer, when just a few cases are registered per month.

In a first step, we determined the monthly and annual average number of intrusions. In Fig. , we present the number of intrusions per month, based on the 7Be criterion (January 1970 to April 2006). We exclude short events of ≤ 2 h. A linear least-squares fit of the data yields a moderate rise in the intrusion count of 0.0456 a-1 (standard deviation 0.028 a-1). Figure  displays the same for the RH criterion (slope of 0.0474 a-1 ± 0.020 a-1) and the regression line for the 7Be criterion from Fig. . It is interesting that the slopes of the two regressions are almost equal, which is explained by the fact that most intrusion cases are the same for both criteria, at least if the minimum RH stays below 30 %. The elevated relative standard deviations are caused by the strong variability in the monthly averages that served as the input data of the least-squares fit. For the 7Be criterion, the calculated count grows from 3.57 (1970) to 5.21 (end of 2005), i.e. by 46 %. For the RH criterion, the growth is 2.56 (1970) to 4.17 (2006) and 4.44 (2012), i.e. by 63 % and 73 %, respectively.

The red curves in Figs.  and  represent the sliding ± 12-month averages, which means a single-year temporal resolution, following the definition in a VDI guideline (VDI, 1999; see Iarlori et al., 2015, Leblanc et al., 2016, and Trickl et al., 2020b, for other definitions). The residual noise is explained by the fact that the arithmetic average is not a perfect frequency filter.

As already shown for the years 2000 to 2004 by Trickl et al. (2010), there is a summer minimum and a winter maximum in the monthly intrusion count. This is verified now for the entire period of the Zugspitze measurements.

The overall result is influenced by the occasionally aperiodic behaviour of the monthly counts before 1995 and is, thus, not shown here.

The average duration of intrusions seems to be almost free of trend, which is, however, masked by variations, particularly for the RH criterion. The average duration of the intrusions for the 7Be criterion in a given year maximizes in winter (25 h ± 5 h around 1980 and 30 h ± 8 h around 2005) and stays at 12 h ± 4 h in summer. A winter maximum of the average duration with 40 to 60 h (7Be criterion) or 45 h (RH criterion) occurred between 1990 and 1995. This maximum does not appear in the corresponding annual average.

In Fig. , we show the monthly and annual averages of ozone in direct intrusions for the RH criterion and the ± 12-month sliding arithmetic averages for both criteria. There is a pronounced positive trend of the annual peak STT ozone. The annual averages exhibit an almost periodic variation, with a period of roughly 7 years. The relative increase in the average STT ozone from 1978 to 2011 exceeds that for the intrusion count (Sect. ). This means that the average amount of ozone transported in individual intrusions has also increased over the years.

The seasonal cycle of the monthly averages is somewhat clearer in its structure. This allows us to present the ozone averages in the direct intrusions for January–February and June–July in Fig. . The winter maxima for the two filtering criteria do not differ by too much. However, the summer minima for the 7Be are higher by roughly 50 % than those for the RH criterion, due to the wet bias for low RH. The equations for the four regression lines are (in ppb) as follows: -9.276(1.81)×102+y×4.713(0.91)×10-1, January–February, 7Be criterion;-1.083(0.86)×102+y×5.587(4.30)×10-2, June–July, 7Be criterion;-9.183(1.94)×102+y×4.668(0.97)×10-1, January–February, RH criterion;and-3.858(5.88)×101+y×2.044(2.95)×10-2, June–July, RH criterion. y is the year, and the numbers in parentheses are the respective standard deviations. In order to guide the eyes, we also fitted a third-order polynomial to the winter data for the RH criterion. The four parameters are as follows: P0=7.345×106,P1=-1.106×104,P2=5.555,P3=-9.298×10-4.

The relative standard deviations of the four parameters are as high as about 0.45 each. This reflects the strong year-to-year variability in the data.

It is obvious that the increase in the STT ozone mostly took place during the cold season. This observation will be further discussed in Sect. . The new analysis for the second half of the 1990s exceeds, during the cold season, the values of the FLEXPART analysis for the years 1995 to 1999, as presented by Trickl et al. (2010) in their Fig. . We tentatively ascribe this fact to the excessive mixing scheme in the model (Trickl et al., 2014).

Although high-accuracy ozone data do not exist before 1978, we attempted to estimate the situation back to 1970 by assuming a constant average ozone mixing ratio (Fig. ). This assumption is justified by the results for the nearby Hohenpeißenberg (distance of about 41 km) sonde measurements for 1970 to 1977, as evaluated by Claude et al. (2001), for 700 mbar (about 3000 m). Claude et al. (2001) published an almost-constant average mixing ratio, which justifies our choice. However, the Hohenpeißenberg mixing ratio before 1978 is about 41 ppb. This value is higher than expected from the extrapolation of the Zugspitze measurements to earlier years, which is estimated to be 36.25 ppb.

As mentioned in the Introduction, the measurements of Fabian and Pruchniewicz (1977) yield somewhat low ozone mixing ratios. In Fig. , we display the monthly mean values for 1970 to 1975, which have been graphically reconstructed from their Fig.  as crosses.

Figures  (monthly averages plus a few ± 12-month averages) and  (annual averages for all quantities) also show the curves for the indirect stratospheric contribution, calculated with Eq. (), for assuming 80 % and 90 % of the 7Be being produced in the stratosphere. The annual averages are listed in Table . The values for the total STT ozone confirm the analysis by Scheel in 2005 rather well (Fig. 1 in Trickl et al., 2020a), considering that Scheel assumed a stratospheric contribution of just 66.7 % (i.e. the global average). The overall stratospheric contribution derived now is 12 or 14 ppb for the 1970s and 19 or 24 ppb around 2005 for the 80 % or 90 % threshold, respectively. The series for the direct intrusions in the 1970s can be nicely extended for the years after 1978. The rise of 7Be in the early 1970s is not reproduced by the STT ozone. The difference in the total ozone mixing ratio and the stratospheric contributions is an estimate of the tropospheric burden. Most importantly, the tropospheric contribution, calculated as the difference in the full annual average mixing ratio and the estimated total stratospheric contribution, does not exhibit a positive trend after 1990, which is the period of improving air quality. In any case, the ozone trend after 1990 is not negative as one could expect from the reduction in European emissions (see the Introduction and Sect. ).

Despite the good performance of the results for the direct intrusions before 1978, the analysis of the indirect stratospheric contributions required a slight adjustment. The analysis showed that the influence of the seasonal cycle cannot be neglected when estimating the indirect stratospheric contribution. We, thus, added an artificial sinusoidal seasonal cycle with an amplitude of 8 ppb (Fig. ). We realized that the “indirect” values for 1970 and 1971 were 2 to 3 ppb higher than those obtained from a sliding 12-month average for these 2 years. Therefore, we corrected for this bias and also modified the unrealistic O3/7Be calibration factor that did not agree with the (rather constant) factor for 1978 to 2006.

In order to extend the analysis beyond 2005 or 2011, respectively, we filtered the UFS values. No 7Be measurements have been performed at UFS. Here, we just apply the RH criterion. In Fig. , we present the results for direct intrusions for the years 2002 to 2020. We also include the smoothed 1978–2011 results for the summit. There is an obvious difference between the two stations between 2002 and 2008. This difference is that it looks rather high with regard to the small altitude difference of only 0.3 km. We found that the RH percentiles are comparatively high during that period (Sect. ) and concluded that there were problems with the RH measurements of the DWD at UFS during the early phase. This could explain the low ozone mixing ratios for direct STT before 2009.

We multiply the UFS average by 1.1 to improve the agreement between UFS and the summit between 2009 and 2011 (dark green curve in Fig. ). It is obvious that the increase in the stratospheric influence on the Zugspitze ozone came to an end at around 2003. The slightly negative trend in Fig.  during the first decade of the new millennium is confirmed, but it comes to an end in 2010. Apart from oscillations in the ozone from direct intrusions, the contribution from direct STT stays rather constant between 2005 and 2020.

The monthly averages of the measurements at both the summit and UFS, together with the sliding ± 12-month averages, are also displayed in Fig. . The UFS ozone values are slightly lower than those obtained at the summit. Scheel and Ries determined an average difference from April 2002 to June 2008 of 0.82 ppb (Fig. 1 in Zellweger et al., 2011). The difference is almost outside the combined uncertainty level for both sides. It is reasonable to conclude that this difference is caused by a lower stratospheric influence at UFS.

In Fig. , we evaluate the amplitude of the seasonal cycle of the overall Zugspitze ozone. The ±12-month averages were subtracted from the monthly averages. There has been an obvious decrease in the amplitude of the seasonal cycle since the late 1980s. We applied linear least-squares fits to the annual maxima and minima, which resulted in the following: 2.0467(0.63)×102-y×9.738(3.16)×10-2 (maxima, in ppb),-2.235(0.27)×102+y×1.073(0.14)×10-1 (minima, in ppb).

Again, y is the year, and the number in parentheses is the standard deviation. In order to obtain a reasonable result, we enhanced the a priori error bars of a few obvious outliers in the input data (such as the dry summer of 2003). For the period 1988 to 2021, we obtain a relative amplitude decrease by 29 % and 35 %, respectively. This means a considerable reduction in the air pollution at this near-background site.

Trickl et al. (2014) show, in their Fig. , the behaviour of the annual average of Zugspitze carbon monoxide for air inside and outside the intrusion layers. The analysis was now repeated for the modified filtering criteria. Figure  shows the results for the RH criterion for 1990 to 2011; here the average monthly contributions are seen. In addition, we include the ± 12-bin sliding averages for both criteria. The slightly positive trend obtained by Scheel (Trickl et al., 2014) for CO in the direct intrusions is confirmed in the revised analysis. This becomes obvious if one connects the maxima or the minima (respectively) of the smoothed curves. Also, the negative trend for the complementary data is confirmed. From 1990 to 2011, the averaged CO outside direct intrusions dropped from about 127 to about 93 ppb. During the early 1990s, the amplitude of the seasonal cycle was clearly higher than later on, which is in agreement with the reduction in the European air pollution during that decade.

We also analysed the UFS CO data in the same way. Since the CO data are preliminary for some years, we do not show the results here. For the time being, we slightly renormalized the UFS data with the summit CO. The decrease in the averaged corrected complementary mixing ratios (not fully tropospheric) intensifies after 2011. By the end of 2020, a roughly estimated value of 72 ppb was reached (i.e. 56 % of the highest value in 1990). This means that there has been a substantial improvement in the tropospheric air quality.

In contrast, the monthly mean CO attributed to direct intrusions stays rather constant after 2011, at about 17 ppb. Thus, the slight rise seen in the summit data from 1990 to about 2005 does not continue, which is similar to the behaviour found for ozone. As stated earlier (Trickl et al., 2014), we speculate that there is an Asian contribution in the tropopause region that is being fed by the frequent offshore warm conveyor belts over the western Pacific (Stohl, 2001). This contribution could lead to a growth of carbon monoxide (or at least prevent a strong decrease). In addition, the roles of biomass burning (e.g. Fromm et al., 2010) or air traffic must be considered.