Impact of stratospheric intrusions and intercontinental transport on ozone at Jungfraujoch in 2005: comparison and validation of two Lagrangian approaches

. The particle dispersion model FLEXPART and the trajectory model LAGRANTO are Lagrangian models which are widely used to study synoptic-scale atmospheric air ﬂows such as stratospheric intrusions (SI) and intercontinental transport (ICT). In this study, we focus on SI and ICT events particularly from the North American planetary boundary layer for the Jungfraujoch (JFJ) measurement site, Switzerland, in 2005. Two representative cases of SI and ICT are identiﬁed based on measurements recorded at Jungfraujoch and are compared with FLEXPART and LAGRANTO simulations, respectively. Both models well capture the events, showing good temporal agreement between models and measurements. In addition, we investigate the performance of FLEXPART and LAGRANTO on representing SI and ICT events over the entire year 2005 in a statistical way. We found that the air at JFJ is inﬂuenced by SI during 19% (FLEXPART) and 18% (LAGRANTO), and by ICT from the North American planetary boundary layer during 13% (FLEXPART) and 12% (LAGRANTO) of the entire year. Through intercomparsion with measurements, our ﬁndings suggest that both FLEXPART and LAGRANTO are well capable of representing SI and ICT events if they last for more than 12 h, whereas both have problems on representing short events. For comparison with in-situ observations we used O 3 and relative humidity for SI events. As parameters to trace ICT events we used a combination of NO y /CO and CO, however these parameters are not speciﬁc enough to distinguish aged air masses by their source regions. Moreover, a sensitivity study indicates that the agreement between models and signiﬁcantly on the threshold values


Introduction
Tropospheric ozone (O 3 ) is known as the most important air pollutant of summer smog causing harm to human health and ecosystem (e.g., Brunnekreef and Holgate, 2002).It strongly determines the oxidizing capacity of the troposphere and acts as the third most relevant greenhouse gas in terms of anthropogenic radiative forcing at the Earth's surface (IPCC, 2007).For these reasons, tropospheric O 3 concentrations have been an important subject of atmospheric research over the last decades.Tropospheric O 3 is known to have two sources: chemical production by the oxidation of volatile organic carbons (VOCs) and carbon monoxide (CO) in the presence of nitrogen oxides (NO x =NO+NO 2 ) (e.g., Levy, 1971;Chameides and Walker, 1973;Crutzen, 1974), and transport from the stratosphere (e.g., Junge, 1962;Danielsen, 1968).Many studies have tried to determine the contribution of O 3 from the stratosphere to tropospheric O 3 (Cristofanelli et al., 2003;James et al., 2003a;Hsu and Prather, 2005;Bourqui, 2006;Brioude et al., 2007).Even if it is currently thought that the greatest contribution to tropospheric O 3 originates from photochemical production (Staehelin et al., 1994;Yienger et al., 1999), the contribution from the stratosphere cannot be neglected (Roelofs and Lelieveld, 1997).
Recent studies have also suggested that cross-tropopause mass flux may increase during the 21st century as a result of climate change (Sudo et al., 2003;Collins et al., 2003).According to these studies, the long term enhancement of stratospheric O 3 could significantly modify the tropospheric O 3 budget.Ordóñez et al. (2007) suggested that the positive O 3 trends in the lower free troposphere over Europe during the 1990s were likely due to changes in lowermost stratospheric O 3 concentrations and associated changes in O 3 flux from the stratosphere.Tarasick et al. (2005) examined Canadian ozonesonde data and found a strong correlation between O 3 changes in the lower stratosphere and in the troposphere at the Canadian ozonesonde stations, suggesting that at least part of the changes in the troposphere results from changes in the lower stratosphere.All these findings indicate that stratospheric intrusions (SI) which transport O 3 -rich air into the troposphere is one of the most important processes that impacts the tropospheric O 3 budget.
The lifetime of O 3 in the free troposphere ranges from a few days to several months (e.g., Liu et al., 1987), which allows its transport over distances of intercontinental and hemispheric scales.Transport of O 3 and O 3 precursors may thus have impacts on O 3 concentrations found downwind of industrialized regions of the Northern Hemisphere, where most of the O 3 precursors (VOCs and NO x ) are emitted (Berntsen et al., 1999;Jacob et al., 1999;Jonson et al., 2001;Stohl, 2001;Wild and Akimoto, 2001;Wild et al., 2004).In particular, transatlantic transport of O 3 is raising a lot of interests because of the relatively short distance between North America and Europe (Wild and Akimoto, 2001).For example, Li et al. (2002) reported that anthropogenic emissions from North America led to an additional 20% of violations of the European council ozone standard in summer 1997 over Europe.A more recent study using the GEOS-CHEM global transport model also indicated that North American emissions contribute 11% to the tropospheric O 3 annual average total burden over Europe while European sources only contribute 9% (Auvray and Bey, 2005).
Trajectory models are widely applied to investigate atmospheric transport processes such as SI and intercontinental transport (ICT).The Lagrangian analysis tool LAGRANTO (Wernli and Davies, 1997) allows the identification of air mass source regions.It has been applied to study SI (e.g., Cristofanelli et al., 2003;Gerasopoulos et al., 2005) and ICT (e.g., Guerova et al., 2006;Li et al., 2005) events.The Lagrangian particle dispersion model FLEXPART (Stohl et al., 2005), which simulates the transport and dispersion of tracers by calculating the trajectories of a multitude of particles parameterizing turbulence and deep convection, has also been used in previous studies to investigate SI (e.g., Stohl et al., 2000;Zanis et al., 2003) and ICT (e.g., Huntrieser et al., 2005).FLEXPART was validated with data from three largescale tracer experiments in North America and Europe (Stohl et al., 1998), and it has been used previously in many case studies of trace gas transport (Stohl and Trickl, 1999;Forster et al., 2004;Spichtinger et al., 2001), in a climatology of ICT (Stohl et al., 2002a;Eckhardt et al., 2003) and in simulating SI (Cristofanelli et al., 2003;James et al., 2003a,b).
A comparison between FLEXPART and LAGRANTO has been discussed in previous studies on stratosphere-totroposphere exchange (STE) flux (e.g., Cristofanelli et al., 2003;Bourqui, 2006).Bourqui (2006) found that purely trajectory-based approaches yield lower-bound estimates for STE flux because they capture the large-scale transport but not small-scale turbulent mixing.In contrast, particle dispersion models with parameterized turbulence led to slightly larger estimates.Nevertheless, it is claimed that both models in Bourqui's study accurately identified regions subject to STE and provided realistic quantitative estimates.
To our knowledge, a statistical inter-comparison of the performance of FLEXPART and LAGRANTO on representing SI and ICT events has not been documented up to now.In this study, FLEXPART and LAGRANTO were run for the year 2005.We evaluate the models' performance on capturing SI and ICT events by comparing the model simulations with measurements at the Jungfraujoch (JFJ) observatory in the Swiss Alps.Moreover, an intercomparison between FLEXPART and LAGRANTO is discussed.It needs to be clarified that in this paper we discuss ICT only from the North American planetary boundary layer (PBL), which transports air pollutants contributing considerably to the European O 3 budget (Auvray and Bey, 2005).
In Sect. 2 we describe the data set and model set up.In Sect.3, we compare model simulations with measurements using two case studies.Section 4 presents a systematic comparison between models and measurements from a statistical point of view.The sensitivity of the comparison to variations of threshold values is determined in Sect. 5. Finally, a general discussion is given and conclusions are drawn in Sect.6.

Measurements
The JFJ station is located at 7.98 • E and 46.55 • N, 3585 m above sea level on the main crest of the Bernese Alps, Switzerland.Trace gas measurements at JFJ are performed within the Swiss Air Quality Monitoring Network (NABEL) and are also part of the Global Atmosphere Watch (GAW) program of the World Meteorological Organization (WMO).Due to its elevated altitude, JFJ samples more than 50% free tropospheric air, particularly in winter (Zellweger et al., 2000).Continuous trace gas measurements are performed by the Swiss Federal Laboratories for Materials Testing and Research (Empa) and additional meteorological parameters are routinely measured by the Swiss Federal Office of Meteorology and Climatology (MeteoSwiss).The methods used for routine chemical trace gas measurements are described in detail by Zellweger et al. (2000).In our study, 1-hourly measurements of NO y , CO, O 3 , and relative humidity (RH) covering the entire year 2005 were used.

FLEXPART
To simulate the influence of SI and ICT at the JFJ site, FLEXPART (version 6.2) (Stohl et al., 2005) was employed, and was driven by model-level data from the European Center for Medium-Range Weather Forecasts (ECMWF) with a temporal resolution of 3 h (analyses at 00:00, 06:00, 12:00, 18:00 UTC; forecasts at 03:00, 09:00, 15:00, 21:00 UTC), and 61 vertical hybrid levels.All relevant fields were interpolated onto a latitude/longitude grid with horizontal resolution of 1 • ×1 • .We conducted forward simulations of longrange and synoptic-scale transport of anthropogenic CO tracers from the North American PBL and of stratospheric O 3 tracers, respectively.Although FLEXPART forward mode is not the optimal setup for simulation for a single receptor site, we adopted it in this study.Indeed, the forward approach not only allows to investigate the influence of SI and ICT events on JFJ point site, but also to evaluate the geographic variability in the surrounding area of JFJ caused by SI and ICT events.
Anthropogenic CO emissions of North America in 1 • ×1 • resolution were taken from the EDGAR Version 3.2 inventory Fast Track 2000 (Olivier et al., 2005), which follows the EDGAR 3.2 method (Olivier and Berdowski, 2001;Olivier et al., 2002).The tracer was purely inert in the model, meaning that chemical degradation was ignored.The particles were released between the surface and 150 m above the ground level at a constant rate according to the EDGAR emission distribution for North America.The total number of the particles treated by the model was approximately 950 000.The total particles were not randomly distributed and mainly transported coherently according to the synopticand meso-scale wind fields, therefore are believed to be sufficient for the purpose of this study.Indeed, some selected case studies confirm the smoothness of the trace concentrations over the JFJ target area.In a particular grid cell, the number of released particles was proportional to CO emissions in the cell related to the total CO emissions in North America.Since transatlantic transport often takes place in the time range of one week (Li et al., 2005), in our simulation the passive CO tracer was allowed to be advected by the winds for 10 days, after which it was removed from the simulation.The lifetime of CO is in the time scale of a month or longer, and the simulation does not account for background CO concentrations but only for enhancements over the CO background due to emissions during the last 10 days.The simulations started on 20 December 2004 in order to fulfill the 10 days spin-up of the model.Mixing ratios of CO were produced separately on a 3-hourly basis on a 3-D grid with horizontal resolution of 1 • ×1 • , and 25 vertical levels.
In this study, the 2 potential vorticity (PV) unit (PVU, 10 −6 Km 2 kg −1 s −1 ) surface was used as the extratropical tropopause, which separates the chemically and dynamically very distinct stratosphere from the troposphere.This definition mimics reasonably well the thermal tropopause (Holton et al., 1995) and has the advantage of being based upon the dynamically significant conservation of PV in adiabatic, frictionless flow.To simulate SI impact with FLEXPART, a passive O 3 tracer was continuously released in the stratosphere within the model boundaries (60 • W-60 • E, 20 • N-80 • N), and was allowed to advect with ECMWF winds.This model domain is sufficiently large for simulating the intrusions discussed in our study, but it is worth noting that the defined domain tends to underestimate the intrusions frequency since the intrusions occurring outside this domain are missing in the resulting climatology.In order to obtain a rough indication of the validity of our simulations in this respect, we used calculated backward trajectories to estimate its restriction.According to the analysis based on backward trajectories, 63% of the trajectories originating in the stratosphere crossed the tropopause within the domain.Nevertheless, Stohl et al. (2000) studied the influence of stratospheric intrusions on ozone concentrations in the alpine region using FLEXPART with an even more restricted domain (50 • W-50 • E).They suggested that although the restricted domain underestimated the influence of older stratospheric intrusions (>4 days), it adequately covers direct intrusions which are most likely to produce clear signatures in the measurements.
A statistical relationship between O 3 and PV (M O 3 =M air ×PV×C×48/29) was used to determine the stratospheric O 3 tracer concentration, where C=60×10 −1 PVU −1 is the ratio between the O 3 volume mixing ratio and PV (in units of PVU) in the stratosphere (Stohl et al., 2000).The ratio 48/29 is used for the conversion of volume mixing ratio to mass mixing ratio where 48 and 29 represent the molecular mass of O 3 and mean molecular mass of ambient air, respectively.As for ICT, the simulation started on 20 December 2004 and ended on 31 December 2005.During the model run, approximately 3 million particles were released, which were distributed proportional to the air density in accordance with the above formula.Each particle was tagged with a clock, which was set to zero when it remained in the stratosphere, and started to count once the particle entered into the troposphere.The clock was reset to zero if the particle returned to the stratosphere.If the particle stayed in the troposphere exceeding its assigned age class of 10 days, it was removed from the simulation.O 3 fields were produced every 6 h as 6-h averages at 1 • ×1 • grid cells.The grid cells were 500 m thick between the surface and 10 km, 1000 m thick from 10 km to 15 km and 5000 m thick from 15 km to 20 km.The output O 3 fields only provide an estimation of the O 3 enhancement relative to the tropospheric background O 3 due to the transport from the stratosphere.

LAGRANTO
Ensembles of 10-day backward trajectories were calculated for every 6 h of the year 2005 using LAGRANTO.ECMWF analysis winds (with a temporal resolution of 6 h, horizontal resolution 1 • ×1 • and 61 vertical hybrid levels) were used to drive the model.Because a single trajectory is often not a reliable estimate of an air parcel's path (due to lack of coherence of the flow), it is necessary to consider an ensemble of trajectories in order to obtain statistically meaningful results and to evaluate the coherence of the air flow.Each ensemble consists of a reference trajectory (centered on JFJ at a mean pressure of 650 hPa, 7.98 • E and 46.55 • N) and 4 horizontally 0.5 • displaced trajectories.

Identification of SI events
Besides having high O 3 concentrations, stratospheric air is characterized by low water vapor content.Thus low humidity levels can indicate stratospheric air in a tropospheric environment.Relative humidity (RH) is considered as the best humidity measure to detect SI.Specific humidity is not an appropriate measure because of its high variation with season, which can lead to an artificial bias in the evaluation of seasonal SI frequency (Stohl et al., 2000).Thus, in this paper RH in conjunction with O 3 was used to objectively identify SI events from the measurements.Enhancements over the 10-day running mean O 3 value (M 10 O 3 ) and 50% RH are taken as thresholds and are applied to O 3 and RH measurements, respectively.If O 3 values are continuously more than 10% above M 10 O 3 and RH values below 50% for a time period of at least 6 h, SI events are identified by measurements.The time span of the period indicates the duration of the SI event, and is referred henceforth as T D .The duration T D allows to classify the single events: we refer to events (either SI event or later on ICT event) with T D >24 h as long events, those with T D falling into the range between 12 and 24 h as medium events, and finally those with T D ≤12 h as short events.
To identify a SI event from FLEXPART simulations, 10 ppb is used as threshold value of O 3 tracer.Accordingly, a FLEXPART SI event is identified if the O 3 tracer exceeds the threshold value.Based upon the analysis of LA-GRANTO trajectories ensemble, we objectively identify a LAGRANTO SI event by the following approach.A stratospheric residence time can be defined for an ensemble of trajectories by the sum of the residence times of the single ensemble members.In these terms, a LAGRANTO identified SI event is characterized by a non-vanishing stratospheric residence time.To simplify our description, we introduce the new parameter T R which represents the ensemble residence time.A LAGRANTO identified SI event at JFJ therefore means that T R must be larger than zero at every time step during a time span of T D , where T D is the duration of the model defined SI event, and again is used to separate long, medium and short events.

Identification of ICT events
CO has a lifetime on the order of weeks to months (Seinfeld and Pandis, 1998), whereas the NO y lifetime is on the order of several days.NO y is removed from the troposphere mainly by wet deposition.During the transport from the North American PBL over the North Atlantic Ocean to JFJ, most of the NO y is removed in the free troposphere, whereas CO remains more or less constant, resulting in a relatively low NO y /CO ratio and enhanced CO concentration compared to the background CO.According to Stohl et al. (2002b), the NO y /CO ratio drops to values below 0.01 for air masses older than 4 days if the original ratio of NO x /CO in the North America was assumed to be 0.16.Zellweger et al. (2003) suggest that the strongly aged air masses measured at JFJ are always accompanied with the lowest NO y /CO ratios, if undisturbed free tropospheric condition occur.In our study, we chose the NO y /CO ratio along with measured CO as parameters to identify ICT from measurements.
To identify ICT events from the measurements, background CO under the undisturbed free tropospheric situation is calculated using the method of partitioning JFJ data according to different meteorological situations as described by Zellweger et al. (2003).They studied the influence of different meteorological situations on JFJ data, and split them into ones with and without impact of local polluted air: for autumn/winter (September to February) 68% and for spring/summer (March to August) 40% of JFJ data were measured under the meteorological situations of undisturbed free troposphere implying without impact from local PBL air.In this study, we assume the same frequencies of meteorological situations and take into account that measurements are characterized by significantly enhanced NO y values (not shown here) when under the influence of the local PBL air (Zellweger et al., 2003).Hence we obtain the free tropospheric situations through removing the measurements with the highest 32% NO y values from September to February and the highest 60% NO y values from March to August.Using the remaining measurements, we calculate the monthly CO mean value which then is used as background CO concentration.A measured ICT event is then defined to have a measured CO higher than the corresponding background CO concentration and NO y /CO lower than its 30% monthly percentile.
An ICT event is identified from FLEXPART simulations if the CO tracer exceeds a threshold value of 5 ppb.Analogously, each ICT event is then marked with its duration T D .A similar approach is used to identify ICT events from the LAGRANTO trajectories ensemble.T R in this case means the ensemble residence time within the North American PBL.An ICT event is identified from LAGRANTO trajectories if T R is larger than zero.

Case studies
In order to illustrate our methodology, we first present two cases of SI and ICT, respectively, which were selected according to the measurements and using the criteria introduced in Sect. 2. (not shown).RH significantly decreased from 17 February 06:00 UTC to 18 February 07:00 UTC, and fluctuated at low levels until 19 February 06:00 UTC.Since the stratosphere shows much higher O 3 concentrations and is much dryer than the troposphere, this signal is indicative for an SI event.However, note that O 3 concentrations around 60 ppb as measured at JFJ during this event indicate significant mixing with tropospheric air since O 3 concentrations in the stratosphere are much higher.By means of the criterion described in Sect.2.4, we identify a measured SI event which sustains for 29 h from 17 February 06:00 UTC to 18 February 11:00 UTC.Accordingly, the 29-h SI event is classified as a long SI event.

SI event on 15-20 February
FLEXPART O 3 tracer values from 15 February 00:00 UTC to 20 February 00:00 UTC show peaking values during the period from 17 February 06:00 UTC to 19 February 00:00 UTC, which covers the period of the measured SI event, indicating good temporal agreement between the FLEXPART simulation and the measurements.O 3 tracer values during the peak period were significantly higher than the threshold value of 10 ppb.According to our terminology, the O 3 tracer indicates a FLEXPART long SI event because of the duration T D >24 h.
It should not be ignored that at 16 February 06:00 UTC, although the combination of O 3 increase and RH decrease did not fulfill the selected criteria for a SI event, changes of O 3 and RH were nevertheless discernible.This stratospheric signature at 16 February 06:00 UTC was also captured by the FLEXPART simulation.As seen in Fig. 1, a relatively narrow peak occurs exactly at 06:00 UTC in the O 3 tracer.
Figure 2 presents LAGRANTO backward trajectories (left) from 18 February 00:00 UTC to 18 February 18:00 UTC.We only present the trajectory ensembles which showed contact with the stratosphere.At least one member of the trajectories ensemble crossed the tropopause and descended to JFJ during the period, which indicates evident stratospheric origin of part of the air.According to the definition of a LAGRANTO SI event, a medium SI event was identified which persists for T D of 18 h.Taking the different time resolutions between the measurements and model output, as well as the bias due to numerical treatment into account, the LAGRANTO SI event well matches with the measured SI event even though its shorter T D .The right panel in Fig. 2 shows the trajectories ensemble started at 16 February 06:00 UTC.Although the reference trajectory stayed in the troposphere, two displaced trajectories had contact with the stratosphere.The T R in the stratosphere at 16 February 06:00 UTC was 162 h, which indicated the air masses arriving at JFJ had clear stratospheric influence.Since the T D is less than 12 h, it is identified as a short SI event.This LAGRANTO short SI event temporally coincides with the FLEXPART short SI event, as well as with considerable O 3 increase and RH decrease in the measurements.This indicates good agreement between LAGRANTO simulation and the measurements, as well as between FLEXPART and LA-GRANTO simulations.

ICT event on 2-4 July 2005
Figure 3 presents a 5-day series of CO, NO y , NO y /CO values during the period from 1 July 00:00 UTC to 5 July 00:00 UTC, along with FLEXPART CO tracer values.CO showed an evident increase starting from 1 July 23:00 UTC while NO y significantly decreased.From 1 July 23:00 UTC to 2 July 15:00 UTC, CO values were almost always higher than background CO and NO y /CO ratios were found to be always lower than the 30% monthly percentile.According to the method described in Sect.2.5, an ICT event with T D of 16 h is thus identified from the measurements.
Except for an offset of several hours, the FLEXPART simulation agrees well with the measurements: the CO tracer showed an evident enhancement from 2 July 03:00 UTC to 4 July 21:00 UTC, which is larger than the threshold value of 5 ppb adopted for this study.We therefore identify it as a FLEXPART ICT event, which is consistent with the www.atmos-chem-phys.net/9/3371/2009/Atmos.Chem.Phys., 9, 3371-3383, 2009 ICT event deduced from the indicators measured at JFJ.It is worth to note that FLEXPART generated an ICT event with a much larger CO increase, probably because the decay of CO tracer is not integrated in the model simulation, which is however not negligible in summer.
Figure 4 shows the LAGRANTO trajectories ensemble from 2 July 06:00 UTC to 3 July 06:00 UTC, which depicts the transport history of air masses from North America to JFJ.The trajectory ensemble indicates the air arriving at JFJ during this period originated from North America (top panel); the air parcels stayed over North America for about one week, and were transported across the North Atlantic Ocean to JFJ afterwards.The vertical distribution of the trajectories which had contact with the North American PBL (here taken as the levels below 800 hPa) during their 10-day history are shown in the bottom panel.The air masses were above North America first descended to the North American PBL, afterwards the air was quickly uplifted and transported across the Atlantic Ocean.The trajectories of the air masses show a coherent transport pattern, which is typical for transport in Warm Conveyor Belts (WCB) (Eckhardt et al., 2004).At every time step during the period from 1 July 06:00 UTC to 3 July 06:00 UTC, at least one member of the trajectory ensemble had contact with the North American PBL, meaning T R in the North American PBL was above zero.Therefore, a LAGRANTO ICT event with T D of 24 h was identified.This event agrees well with the FLEXPART and the measured ICT event despite a few hours offset.

General comparison
In the previous section it was shown that individual SI and ICT events, as identified by transport simulations (FLEX-PART and LAGRANTO), agreed well with measurement based identifications of such events.The same approach is now applied to one year (2005) of model simulations in order to obtain a statistical evaluation of the models' performance.
Figure 5 presents the one-year time series of measured O 3 and RH, FLEXPART O 3 tracer and LAGRANTO diagnosis  of SI events in 2005.According to FLEXPART, the O 3 tracer was found to be higher than its threshold value of 10 ppb for 19% of the time in 2005.According to LAGRANTO, 18% of the time in 2005 air masses arriving at JFJ had contact with the stratosphere in their 10-day backward history.To be more specific, during 4% of the entire year, contact with the stratosphere took place within 4 days before the air masses arrived at JFJ.For the remaining 14%, the contact with the stratosphere took place between 4-10 days prior to arrival at JFJ.The results are listed in Table 1.
Figure 6 presents time series of measured CO, NO y /CO ratio, FLEXPART CO tracer and LAGRANTO diagnosis of  ICT events in 2005.Applying the criterion of Sect. 2 to FLEXPART and LAGRANTO, the influences of the intercontinental transport from the North American PBL on JFJ were quantified, respectively.FLEXPART yields a frequency of 13% for ICT impact at JFJ, which is well comparable to the frequency derived from LAGRANTO simulations of 12%.The results are also listed in Table 1.

Model comparison of SI and ICT events
Based upon measured O 3 and RH and according to the criterion used in Sect.3, we identified 103 measured SI events, which include 4 long events, 19 medium events and 80 short events (see Table 2).
A quantitative assessment of the models' performance was accomplished using the following approach: if during the period of a measured SI event, at least one SI event is identified with the model simulation, we consider the model well captures the measured SI event.We call this approach "forward comparison".By means of this approach, we found that FLEXPART captured 3 of 4 long SI events, 12 of 19 medium SI events, but only 27 of 80 short SI events.Correspondingly, LAGRANTO captured 4 long SI events, 16 of 19 medium SI events and 41 of 80 short SI events.Although in total FLEXPART and LAGRANTO captured only 41% and 59% of 103 measurement SI events, they captured 65% and 87% of measured long and medium SI events, respectively.This indicates that under the current criteria both models are more capable of capturing long and medium measured SI events.In addition, it also shows that under the chosen thresholds, LAGRANTO is somewhat superior to FLEXPART for capturing measured SI events.In the measurements 143 ICT events were identified in 2005 using NO y /CO ratios and measured CO as thresholds (see Sect. 2.5).These ICT events were further split into three groups according to the length of T D , as shown in Table 3.They consist of 8 long, 21 medium and 114 short events.After applying an analogue approach of forward comparison, it was found that FLEXPART captured 1 of 8 long, 13 of 21 medium and 34 of 114 ICT short events.On the other hand, LAGRANTO captured 4 of 8 long, 10 of 21 medium and 39 of 114 short ICT events.Compared to modelled and measured SI events, the corresponding agreement of ICT events is less convincing.This might partly be due to the larger uncertainties involved in determining ICT events from measured CO and NO y /CO.Air masses originating from other polluted source regions having traveled long enough in the troposphere before arriving at JFJ, are also expected to exhibit relatively low NO y /CO ratios and CO enhancement.Even so, the results show that both FLEXPART and LA-GRANTO are more capable on capturing long and medium ICT events than short ones except for FLEXPART on long ICT events.

Confirmation of modeled SI and ICT events by measurements
In a second step, comparison between model identified events and the measured events was conducted in a reverse way, namely "backward comparison".We firstly collected SI events identified with FLEXPART O 3 tracer values and LAGRANTO trajectory ensembles, respectively.Applying the same criteria as in Sect.3, we found 121 FLEXPART SI events including 16 long, 21 medium and 84 short events, and 177 LAGRANTO SI events including 1 long, 20 medium and 156 short events.The detailed results are displayed in Table 4.The approach used in backward comparison is analogue to the forward one: a model SI event is assumed to be confirmed by the measurements if at least one measured SI event (independent of its length) is found during the period of the model SI event.Comparing all FLEXPART SI events with measurements, we found that 75% of 16 long, 62% of 21 medium and 23% of 84 short events were confirmed.Correspondingly, for LAGRANTO we found 60% of 20 medium and 34% of 156 short events confirmed by the measurements.Astonishingly, the only long event was not confirmed by the measurements.However the measured NO y concentration during this long SI event was found to be significantly elevated, which might be due to the mixing with uplifted polluted air from the planetary boundary layer, causing increased O 3 titration, thus to large extent weakening the stratospheric signature.Additionally it can be also just mixing and dilution of the stratospheric air with PBL air with lower ozone concentration which may cause weakening of the stratospheric signature.The results of confirmed events in percentages are shown in brackets in Table 4.
We present the results of the backward comparison on ICT events in Table 5.In order to highlight the ICT from the North American PBL, we considered only the trajectories which have contact with the North American PBL and have no contact with the European PBL.Therefore, we could to a large extent diminish the impact of European PBL NO yrich air on NO y measurements, and solely focus on the impact of long-range transport from the North American PBL on trace gas composition at JFJ.In the year 2005, 118 LA-GRANTO ICT events were identified, which include 1 long, 19 medium and 98 short events.Comparison with measurements shows that 100% of long, 47% of 19 medium and 43% of 98 short events were confirmed by measurements.As for FLEXPART, applying the method described in Sect.2.5 to FLEXPART CO tracer, we identified 80 ICT events including 14 long, 19 medium and 47 short events.In comparison with measurements, it was found that for FLEXPART, 71% long, 68% of 19 medium and 36% of 47 short events were confirmed by measurements.

Sensitivity study
In this section, we investigated the sensitivity of the agreement discussed in Sect. 4 between FLEXPART or LA-GRANTO and the measurements to varying threshold values of the control parameters.With respect to SI identification of the measurements, the threshold value of O 3 was changed from 10% M 10 O 3 to 7.5% M 10 O 3 and 5% M 10 O 3 , respectively.The RH threshold value was separately changed from 50% to 40%, and then to 60%.The threshold value of NO y /CO was changed to 20% and 40% of its monthly percentile.For the SI and ICT identification based upon FLEXPART simulation, 15 ppb and 10 ppb as threshold values were additionally applied to the FLEXPART O 3 tracer and CO tracer.For LA-GRANTO SI and ICT events, new residence times T R (12 and 24 h) in the stratosphere and in the North American PBL were applied.With every alteration of the above mentioned threshold values, the calculations of Sect. 4 were repeated.The results of the agreement between models and measurements are displayed as percentages in Fig. 7 for SI events and in Fig. 8 for ICT events.In both figures, the upper panel shows the results of the foward comparisons, the lower one the results of the backward comparisons.
The general pattern in Figs.7 and 8 show that the agreement between FLEXPART or LAGRANTO and measurements depends significantly on the applied threshold values of the control parameters.Although an apparent change on the agreement was found when changing the threshold values, the results were still in line with the conclusion drawn in Sect.4, suggesting that both FLEXPART and LAGRANTO are well able to capture long and medium SI and ICT events, however less favorable on capturing short SI and ICT events.The figures also show that generally better agreement was achieved when using less strict threshold values.Few exceptions will be discussed in the following.
Figure 7 (top panel) shows when increasing the threshold value of O 3 from 10 to 15 ppb, the response on the agreement between FLEXPART and measurement was unchanged for long and medium SI events, meaning all measured long and medium SI events correspond to FLEXPART O 3 tracer of above 15 ppb.Moreover, when solely increasing the threshold value of T R in the stratosphere from 0 to 12 h and then to 24 h, agreement between LAGRANTO and measurements on long SI events also showed no response to the alteration, indicating that all measured long SI events correspond to large stratospheric T R (>24 h) characterized in the trajectories ensemble.This finding suggests the measured long SI events were consistent with significant FLEXPART O 3 elevation and large T R in the stratosphere as obtained from the analysis with LAGRANTO trajectories.Hence, there is a good agreement between FLEXPART or LAGRANTO and measurements, as well as between FLEXPART and LAGRANTO on the identification of significant long SI event.
Figure 8 (top panel) shows generally modest agreement between FLEXPART or LAGRANTO and measurements in identifying ICTs, supporting the conclusion from the statistical point of view that a combination of NO y /CO and CO is still not suitable to specifically identify ICT from the North American PBL.Note that the agreement between LAGRANTO and measurement on the measured long and medium ICT events was reasonably stable when changing the threshold value of NO y /CO from 20% to 30%, and then to 40% of its monthly percentile.Figure 8 (bottom panel) shows that the agreement between LAGRANTO and measurements was not changed if we altered either the threshold value of NO y /CO from 20% to 30%, or 40% of its monthly percentile or increased T R in the North American PBL from 0 to 12 h and 24 h.This suggests that LAGRANTO long ICT events were characterized with long residence times in the North American PBL (>24 h), which is consistent with significant ICT impact in the measurements as the temporal corresponding NO y /CO ratios were always below the 20% monthly NO y /CO percentile.
The impact of regional air pollution levels on our results was also considered.As described in Sect.2.5, 68% in winter/spring and 40% in summer/autumn of meteorological conditions indicate free troposphere at the measurement site.These rules, used to calculate the CO background, were derived for 1997/1998 and might not hold for the year under investigation due to the natural year-to-year variability.Therefore, sensitivity experiments were performed with 60%, 70% and 75% in winter/spring and 35%, 45% and 50% in summer/autumn to calculate the background CO concentraion.No significant difference was found on the results (not shown here).
Finally we made an attempt to explore the sensitivity regarding backward trajectory length.For this purpose LAGRANTO 20-day backward trajectories were calculated based on ERA interim data.The analyses based upon the newly calculated trajectories show that using 20-day trajectories, we were able to identify twice as many SI and ICT diagnoses when comparing with 10-day trajectories.However this won't change our conclusion drawn previously, since the influence of SI and ICT events with transport time over 10 days will possibly not be reflected evidently in the measurements due to the enhanced mixing and dilution throughout the transport process.In addtion, note that the purpose of our study is to compare the performance of two models on representing SI and ICT events, the models' capability on capturing all events is not our primary focus.

Conclusions
Measurements, the particle dispersion model FLEXPART and the trajectory tool LAGRANTO were used independently to identify stratospheric intrusion (SI) and intercontinental transport (ICT) events occurring at the high mountain measurement site Jungfraujoch (JFJ) in 2005.Our particular interest was devoted to evaluate the performance of the two models by intercomparing with the measurements.This study represents the first attempt to investigate the models' capability of representing SI and ICT events in a statistical way.
Comparison between measurements and models was conducted reversibly.SI events were firstly selected based upon measurements and then compared with FLEXPART and LA-GRANTO simulation.Our results show that both FLEX-PART and LAGRANTO are capable of capturing most of the evident SI events (duration longer than 12 h), however encounter severe difficulties in capturing unconspicuous SI events (duration shorter than 12 h).Low agreement between models and measurements on short SI events points out that the transient nature and realistic representation of these events is associated with a higher degree of uncertainty.Furthermore, our results seem to suggest that more evident SI events were identified with FLEXPART than with LA-GRANTO.This is probably because of the different implication of the age class of 10 days in the FLEXPART simulation and 10-day backward time in LAGRANTO.In FLEXPART, 10 days indicate the longest residence time of the particles in the troposphere after crossing the tropopause, while 10 days in LAGRANTO simulation include the entire transport time of air parcel in the troposphere and stratosphere.Concerning ICT events, the comparison between measurements and model output is more difficult because no combination of field measurements are known to us to specifically identify North American ICT events.However, it looks as if both models (FLEXPART and LAGRANTO) are better at representing evident ICT events than less obvious ones.Nevertheless, note that the 10-day duration of the FLEXPART simulations and LAGRANTO calculations has restrictions.Both models are believed to be capable of capturing the fast (evident) ICT events which are most likely to be reflected in the measurements, however are unlikely able to identifiy the slow ICT events especially in summer when atmospheric transport is slower.In addition, the comparison between FLEXPART and LAGRANTO on predicting SI and ICT events show very good agreement, which is not suprising since both models use ECMWF input data.The two models are therefore not independent of each other.
A sensitivity study reveals that generally less strict threshold values resulted in better agreement between simulations and measurements.For every alteration of employed threshold values, the change in the agreement between models and measurements on representing SI events were within 20%.When adopting less strict threshold values, although a higher percentage of measured ICT events were captured by the models, the agreement between measurements and models was however still low.In summary, our analysis shows that particle dispersion and trajectory tools are capable to capture well long-lasting, significant SI and ICT events, but encounter problems when considering short events.

Figure 1
Figure 1 presents the 5-day series of O 3 and RH measured at JFJ from 15 to 20 February 2005, showing enhanced O 3 concentrations with continuously more than 10% of M 10 O 3

Fig. 1 .
Fig. 1.Measurements and tracer model results for the period between 15 February 00:00 UTC and 20 February 00:00 UTC showing FLEXPART's stratospheric O 3 tracer (thick solid line) and the measured O 3 (thin solid line) and RH (thin dash line).

Fig. 6 .
Fig. 6.Measured CO (top: grey) and NO y /CO ratios (bottom: grey) with colored background representing matching with FLEXPART (red), LAGRANTO (blue) and both models (yellow) ICT diagnostics.Dashed black lines indicate monthly background CO (top) and 30% monthly quantiles of NO y /CO (bottom).

Fig. 7 .
Fig. 7. Sensitivity of the agreement on SI events between FLEX-PART simulation (black) or LAGRANTO simulation (red) and the measured SI events in forward comparison (top) and in backward comparison (bottom) to varying threshold values of FLEXPART O 3 tracer, LAGRANTO T R , measured O 3 and RH.

Table 1 .
Frequencies of the influences of stratospheric intrusion and intercontinental transport from the North American PBL on JFJ in 2005 according to FLEXPART and LAGRANTO, respectively.

Table 2 .
Agreement (in number of events and in percentage) between FLEXPART and LAGRANTO simulations and the measured SI events at JFJ in 2005 according to residence time T D in the stratosphere.

Table 3 .
Agreement in percentage between FLEXPART and LA-GRANTO simulations and the measured ICT events at JFJ in 2005 as a function of T D in the North American PBL.

Table 4 .
Agreement in percentage (shown in bracket)between models identified SI events and the measurements at JFJ in 2005 as a function of T D of the event in the Stratosphere.

Table 5 .
Agreement in percentage (showing in the bracket) between FLEXPART and LAGRANTO simulations and the measured ICT events at JFJ in 2005 as the function of T D in the North American PBL.