European source and sink areas of CO 2 retrieved from Lagrangian transport model interpretation of combined O 2 and CO 2 measurements at the high alpine research station Jungfraujoch

. The University of Bern monitors carbon dioxide (CO 2 ) and oxygen (O 2 ) at the High Altitude Research Station Jungfraujoch since the year 2000 by means of ﬂasks sampling and since 2005 using a continuous in situ measurement system. This study investigates the transport of CO 2 and O 2 towards Jungfraujoch using backward Lagrangian Particle Dispersion Model (LPDM) simulations and utilizes CO 2 and O 2 signatures to classify air masses. By investigating the simulated transport patterns associated with distinct CO 2 concentrations it is possible to decipher different source and sink areas over Europe. The highest CO 2 concentrations, for example, were observed in winter during pollution episodes when air was transported from Northeastern Europe towards the Alps, or during south Foehn events with rapid uplift of polluted air from Northern Italy, as demonstrated in two case studies. exchange variations in O 2 concentrations at Jungfraujoch correlation between CO 2 and APO (Atmospheric Potential Oxygen) deviations a seasonally varying background


Introduction
The present-day concentration of atmospheric carbon dioxide (CO 2 ) continues to rise. Ongoing emissions of CO 2 and other greenhouse gases will influence the global climate system during the next decades and centuries. It is therefore essential to gain further insight into the processes involving CO 2 exchanges among the atmosphere, the biosphere and the ocean in order to estimate the carbon sources and sinks and to comprehend the global carbon cycle. For that reason CO 2 and other atmospheric compounds are constantly monitored at several locations around the world. Long-term records obtained with high precision measurements are used to study the evolution of the changes in time and to budget the sources and sinks of atmospheric CO 2 . Atmospheric oxygen observations (often measured as the ratio of oxygen to nitrogen, i.e. O 2 /N 2 and expressed as δO 2 /N 2 in relation to a standard ratio) can be used as a complementary carbon cycle tracer which provides additional information on the partitioning of anthropogenic CO 2 between the terrestrial biosphere and the ocean (Keeling, 1988;Keeling and Shertz, 1992;Battle et al., 2000;Van Der Laan-Luijkx et al., 2010).
Published by Copernicus Publications on behalf of the European Geosciences Union. Systematic measurements of atmospheric O 2 have begun in 1989 by means of flask sampling at three sites (Keeling and Shertz, 1992). Since then, the flask-sampling network for atmospheric O 2 measurements has been extended over the globe (Bender et al., 2005;Manning and Keeling, 2006;Tohjima et al., 2008) and later on also included continuous on-site measurements (Manning, 2001;Kozlova et al., 2008;Valentino et al., 2008;Popa et al., 2010;Thompson et al., 2009).
The analysis of the correlation between CO 2 and O 2 is especially useful because CO 2 and O 2 are inversely coupled in terrestrial biospheric exchange processes (photosynthesis and respiration) with a rather constant stochiometric ratio of about −1.1 mol O 2 per mol CO 2 (Severinghaus, 1995). O 2 fluxes between ocean and atmosphere are mainly driven by the biological and the solubility pump, whereas ocean's CO 2 uptake is dominantly driven by the carbonate buffering system (Sabine et al., 2004). In the air-sea exchange between O 2 and CO 2 due to photosynthesis and respiration of marine biosphere (biological pump), the stochiometric ratio has been determined to be −1.4 mol O 2 /mol CO 2 (Takahashi et al., 1985) clearly deviating from the terrestrial biospheric ratio. The seasonal variations in the O 2 /N 2 ratio of air are partly caused by the warming and cooling of the ocean (solubility pump), also called ocean thermal effect. Typically, O 2 has a low solubility in water and exchanges very fast between the sea surface and the atmosphere with a timescale of a few weeks (Keeling et al., 1993). As oceans temperature rises in spring and summer, gases become less soluble and are transferred from the ocean to the atmosphere. In wintertime, seawater cools and the effect is reversed (Keeling and Shertz, 1992;Bender et al., 1996).The ocean is a net source of O 2 at the sea surface during spring and summer, because more organic matter is produced in the euphotic zone than is consumed by respiration. Net O 2 production begins in spring, when phytoplankton has enough light to bloom. Therefore surface seawater becomes supersaturated in O 2 and consequently a flux of O 2 towards the atmosphere is initiated which is reinforced by the thermal outgassing. Part of the organic matter produced during the phytoplankton photosynthesis accumulates in the ocean mixed layer as particulate or dissolved organic carbon (POC or DOC). In fall and winter when the sea surface temperatures cool down, the mixed layer deepens bringing O 2 under-saturated waters to the surface causing a demand of O 2 from the atmosphere and thus decreasing the atmospheric O 2 concentration. This decline continues until the following spring when a new cycle starts (Sverdrup, 1953;Siegel et al., 2002).
On the other hand, during the combustion of fossil fuels O 2 is consumed and CO 2 produced with a ratio of −1.39 mol O 2 /mol CO 2 . This value is the global consummation weighted average of the ratios of the different fuels, like −1.95 for gas fuels, −1.44 for liquid fuels, −1.17 for solid fuels and −1.98 for gas flaring (Marland et al., 2009). Therefore, by knowing the stochiometric ratios of O 2 -CO 2 exchanges, one can separate the CO 2 uptake into land and ocean components. A sensitive tracer for oceanic air-sea gas exchange is the Atmospheric Potential Oxygen (APO= O 2 /N 2 +1.1 CO 2 /0.20946, with CO 2 =CO 2 −365 ppm) as proposed by Stephens et al. (1998). Variations in APO (seasonal and latitudinal differences) are sensitive to CO 2 and O 2 air-sea exchanges plus a residual due to the fossil fuel, but APO is invariant with respect to land biospheric processes (Stephens et al., 1998).
Regarding the fossil fuel influence, our results are compared with the anthropogenic CO 2 emissions distribution in Europe according to the Emission Database for Global Atmospheric Research (EDGAR) emission inventory for the year 2005 (EDGAR v4.0, 2005, http://www.pbl.nl/en/ themasites/edgar/index.html). According to the EDGAR emission map (Fig. 1), Germany (especially the Ruhr area), Belgium, the Netherlands, and Poland at the Northern side of the Alps as well as the Po valley (Northern Italy) at the Southern rim of the Alps are the major source regions of anthropogenic CO 2 in Europe. It is noteworthy that the Po valley is one of the most well-known source regions for air pollutants observed at Jungfraujoch (Reimann et al., 2004. These potential source regions are the most densely populated and industrialized countries in Europe. An exception in this regard is France due to it's high proportion of nuclear energy. The emissions per capita are highest in the Benelux states and also tend to be high in the most northern countries (heating) and Mediterranean countries (cooling systems), while emissions in the European part of Russia are only about half of those in Western Europe (Schulze et al., 2009).
In this paper, we present a record of 4.5 yr (January 2005-June 2009) of continuous measurements of CO 2 and O 2 at the High Altitude Research Station Jungfraujoch in the Swiss Alps, which is an important measurement site in Europe to monitor atmospheric composition particularly in the lower free troposphere. Our main focus is on investigating, by means of Lagrangian Particle Dispersion Model (LPDM) simulations with FLEXPART (Stohl et al., 2005), the origin of short-term deviations of CO 2 and O 2 from background concentrations at Jungfraujoch and on understanding their seasonally changing characteristics. Many studies have been carried out on local and long range transport and on identifying source and sink regions of polluted air reaching Jungfraujoch (e.g. Seibert et al., 1994Seibert et al., , 1998Forrer et al., 2000;Henne et al., 2005a;Cozic et al., 2008;Balzani Lööv et al., 2008;Tuszon et al., 2011). However, only few studies have done this in a systematic way using a statistical approach in which multiyear trace gas observations are interpreted by means of a large number of trajectories or LPDM results, which has the significant advantage that errors in individual transport simulations tend to cancel out. Examples for this approach are the studies by  studying the origin of NO x , O 3 and CO at several sites in the Alps and those by Balzani-Lööv et al. (2008) and   Cui et al. (2009) investigating background concentrations and the role of stratospheric intrusions and intercontinental transport, respectively. Here, we follow such a statistical approach to classify air masses with distinct O 2 and CO 2 signatures with respect to their origin. Rather than clustering air masses according to transport histories, i.e. trajectory clustering (Moody and Galloway, 1988;Henne et al. 2008) or LPDM clustering (Paris et al., 2010;Hirdmann et al., 2010), the classification is based on identifying specific regions with the CO 2 -O 2 correlation and then analyzing the air mass origin for each class. Similar to the study by Lewis et al. (2007), this approach puts more weight on the trace gas observations as compared to the transport modeling, which is likely better suited to identify different source and sink processes. However, while Lewis et al. (2007) performed an objective clustering of air masses based on simultaneous observations of multiple species, our approach is to focus on two closely interrelated species only which precluded such a cluster analysis.
2 Site description, experimental and model methods

The Jungfraujoch research station
The High Altitude Research Station Jungfraujoch (7 • 59 2 E and 46 • 32 53 N) is located on the Southern side of a rocky crest in the Bernese Alps, Switzerland, at an altitude of 3580 m a.s.l. (Sphinx Observatory). It is situated on a saddle between the Jungfrau peak (4158 m a.s.l.) in the west and the Mönch peak (4099 m a.s.l.) in the east. Therefore, the wind is channeled locally in north-south direction. In winter and summer more than 70 % of the wind direction is from northwest. North is also the prevailing wind direction in the other seasons. Due to its location at high elevation and its central position in Europe the site is particularly suited to observe the free tropospheric background air above Europe (Schwikowski et al., 1995;Cozic et al., 2007;Reimann et al., 2008;Lanz et al., 2009). Nevertheless, several studies on aerosol and trace gas composition (Lugauer et al., 1998;Forrer et al., 2000;Coen et al., 2004;Reimann et al., 2004;Li et al., 2005) pointed out that Jungfraujoch is frequently influenced by direct transport from the planetary boundary layer through convection and local thermal wind systems mainly during summer, when vertical transport is more active. In addition to these rather local influences, frontal uplift of air from the European boundary layer followed by quasi-horizontal transport to Jungfraujoch episodically affects the station in all seasons of the year. Therefore greenhouse gas measurements at Jungfraujoch can capture both advected air from the polluted boundary layer as well as air from the lower free troposphere. Moreover, the permanent research station is easily accessible by railway all year round. Local emissions are usually low, since all heating is electrical and waste is transported back to the valley.
The central role of the research station Jungfraujoch is documented by the fact that it was selected as one of the Global Atmosphere Watch stations (GAW) within the framework of WMO activities. In addition, Jungfraujoch is an alpine site within the Network for the Detection of Atmospheric Composition Change (NDACC) and acts as a station within the Swiss national air pollution monitoring network (NABEL) and the European Monitoring and Evaluation Program (EMEP).

Continuous measurements of CO 2 and O 2 at Jungfraujoch
The Climate and Environmental Physics Division of the University of Bern is monitoring atmospheric CO 2 and O 2 at the High Altitude Research Station Jungfraujoch since 2000 by means of weekly flasks sampling. At the end of 2004 a continuous measurement system for in situ analysis of CO 2 and O 2 was installed in order to get an optimal data coverage to better detect short-term fluctuations in CO 2 and O 2 concentrations. Carbon dioxide is measured by a commercial NDIR infrared analyzer (S710 UNOR, SICK MAIHAK) while oxygen is measured in two ways: paramagnetically, utilizing the paramagnetic properties of oxygen molecules (at the Jungfraujoch Parox 1000 paramagnetic sensor, from MBE AG, Switzerland is used) and electrochemically, by means of fuel cells. We use Max-250 fuel cells manufactured by Maxtec, USA (Sturm, 2005;Valentino et al., 2007Valentino et al., , 2008Uglietti et al., 2008). The CO 2 results are reported on the WMO mole fraction scale (XWMO 2007) whereas O 2 values are reported on our internal oxygen scale since no international and common calibration scale exists yet. The difference between the Bern internal scale and SIO scale was set to −556 per meg after measuring several times three standard air cylinders (CA07043, CA07045, CA07047) filled and determined at the Scripps Institution of Oceanography of California, La Jolla. The continuous measurement system at Jungfraujoch is calibrated using calibration gases contained in 50 l cylinders from CARBAGAS, Switzerland. All standard cylinders are placed horizontally in order to avoid thermal and pressure fractionation of O 2 /N 2 .
It is indeed known that for instance thermal fractionation of O 2 relative to N 2 inside high pressure gas cylinders resulting from small ambient temperature changes can lead to drifts in δO 2 /N 2 values for the delivered air on the order of 10-20 per meg over the liftime of the cylinder (Keeling et al., , 2004(Keeling et al., , 2005Langenfelds et al., 2005).
Every calibration gas is passing through the entire system alternatively following an automated sequence written in LabView. We use a high CO 2 calibration gas (430 ppm), which correspond to the low O 2 calibration gas (−1300 per meg), a low CO 2 calibration gas (356 ppm) corresponding to the high O 2 calibration gas (35 per meg) and the working gas which should have CO 2 and O 2 concentrations in the range of the air sample values. The long-term reproducibility of the measurements was calculated from a standard gas which is normally measured twice a day in the same way as the sample air (every gas is measured for 6 min). Data allocation is on one second basis but the first 242 s were skipped for the data processing to avoid the abrupt transient changes occurring after the switching between different gases, whereas the next 115 s were averaged. The remaining 3 s were skipped to exclude pressure pulse influences from the valve switching. The standard error of the measurements was computed during a period of one month and the resulting reproducibility for CO 2 was determined to be less than ±0.1 ppm (around 0.06-0.08 ppm) and for O 2 to be in the range of ±8 per meg. The CO 2 value matches the target precision of 0.1 ppm. For O 2 , on the other hand, the value is still higher than the target precision of 5 per meg for O 2 (Uglietti et al., 2009).
The CO 2 measurements are part of the Swiss GCOS (Global Climate Observing System) office program supporting long-term records. The online data are archived through IMECC (preliminary dataset, https://ramces.lsce.ipsl.fr/ index.php?option=com content&view=article&id=36). The CO 2 data record presented in this study have been submitted to the WDCGG database by the May 2011. The flask data are only available through the CarboEurope database.

Lagrangian transport simulations for qualitative allocation of sources and sinks
Trajectory models are a common tool in environmental and atmospheric science to study transport of air masses in the atmosphere. Trajectories describe the pathway of individual air parcels and are typically calculated based on 3-dimensional wind data from a meteorological model (Stohl et al., 1995(Stohl et al., , 1999Stohl and Seibert, 1998). A significant shortcoming of trajectories is that they describe atmospheric transport only in a simplified way neglecting sub-grid scale turbulent and convective motions. Nevertheless, the statistical analysis of a great number of back trajectories from receptor sites has turned out to be a valuable tool to identify sources and sinks of atmospheric trace substances . Lagrangian particle dispersion models (LPDMs) are increasingly being used in atmospheric science to study the transport and dispersion of air masses in the atmosphere (Lin et al., 2011).
As opposed to simple trajectory models, LPDMs also account for turbulent motion and convection. We employed the FLEXPART LPDM (Stohl et al., 2005) in backward (receptor-oriented) mode to establish the relation between potential sources and the receptor location Jungfraujoch (Seibert and Frank, 2004). FLEXPART describes the evolution of a dispersing plume by simulating the transport of "particles", i.e. infinitesimally small elements of air, by the 3-dimensional grid-resolved wind and by sub-grid scale turbulent and convective motion. As meteorological input for the model 3-hourly fields (alternating between 6 hourly analyses and short forecasts) from operational European Centre for Medium Range Weather Forecast (ECMWF) were used. For the period January 2005 to February 2006, meteorological fields were available at 1 • × • horizontal resolution and 60 vertical levels. In February 2006, ECMWF switched to a higher-resolution model (T799 spectral resolution, 91 levels). From February 2006 onwards, we therefore additionally prepared ECMWF input fields at much higher resolution (0.2 • × 0.2 • , 91 levels) for a domain covering parts of Europe (4 • W-16 • E-39-51 • N). This allowed us to perform nested FLEXPART simulations at high resolution over Europe and at lower resolution outside. At the coarse resolution of 1 • × 1 • the ECMWF model surface at Jungfraujoch is only at 1300 m, thus more than 2 km below the true altitude. At the high resolution of 0.2 • × 0.2 • the Alpine topography is better resolved and the model surface is at 2170 m. For each 3 h time interval, 50 000 particles were released from Jungfraujoch and tracked 5 days backward in time. For each simulation, a footprint was calculated which is a gridded representation of the residence time of the particles in a layer above the model surface (0-100 m in our case), and can be interpreted as a map of the sensitivity of a receptor to a source of unit strength (Seibert and Frank, 2004). Each footprint was then associated with the corresponding 3 h  average of the CO 2 and O 2 measurements. Particles were released from 3000m which is a compromise between the model surface altitude and the real station height and is based on a comparison of specific humidity measurements at Jungfraujoch with vertical profiles acquired from routinely launched soundings in Payerne over the flat Swiss Plateau (46.8 • N and 6.9 • E), which showed best agreement around 3000 m. The best release altitude for mountain stations is a matter of ongoing debate. In the study of , for example, the best results were obtained for particles released from Jungfraujoch close to the model surface. However, their results, which were based on a different LPDM and simulations at higher resolution using a mesoscale model, are not directly applicable here. Since our study focuses on qualitative assessment of source/sink regions rather than on quantitative emission estimation, we do not consider the choice of release altitude as being critical. It should also be noted that the first version of our manuscript (Uglietti et al., 2011) has been based on trajectory calculations at lower resolution in which the model surface at Jungfraujoch was more than 800 m lower than in the present simulations. Nevertheless, the results had been quite similar.
The total footprint, i.e. the sum of all footprints from January 2005 to June 2009 is presented in Fig. 2. Not surprisingly, the influence from the western direction is enhanced in comparison to the east due to the prevailing westerly winds . The maximum residence time and thus the maximum sensitivity to source/sink processes is close to the receptor point Jungfraujoch and decreases rapidly with increasing distance (approx. with the squared distance). The footprints also help to understand the general pattern of atmospheric transport to Jungfraujoch. Air masses from the PBL are typically transported to the site by meteorological processes generating substantial upward motion such as fronts, Foehn events or thermal up-lifting by convection during anticyclonic periods in summer. These events together with associated atmospheric tracer concentrations can be used to assess emission areas from the European continent.
Transport patterns and source/sink regions associated with air masses with distinct CO 2 and O 2 characteristics (e.g. very high CO 2 levels during anthropogenic pollution episodes) are visualized in this study by means of "relative footprints". These are obtained by dividing the total footprint of a given set of observations with concentrations above or below a threshold by the overall total footprint shown in Fig. 2, after first scaling both footprints by the respective number of data points. The relative footprint is directly proportional to the Potential Source Contribution Function (PSCF) proposed in other studies (Zeng and Hopke, 1989;Scheifinger and Kaiser, 2007) in which this concept was applied to trajectory rather than LPDM simulations. The only difference is the scaling by the number of footprints (or trajectories) which makes the relative footprint independent of the number of points selected in a given subset. The relative footprint indicates where a selected subset of air masses has spent relatively more (or less) time close to the surface than the "average air mass" observed at Jungfraujoch. The strong radial pattern of absolute footprints with a peak at Jungfraujoch as seen in Fig. 2 is eliminated by the use of relative footprints or PSCF .

CO 2 and O 2 time series
Hourly means of CO 2 and O 2 online records from Jungfraujoch are presented in Fig. 3 and the corresponding mean seasonal cycles in Fig. 4. Atmospheric oxygen and carbon dioxide reveal the typical behaviour observed at mid-latitude sites of the Northern Hemisphere. The CO 2 winter maximum is reached in March, when the photosynthetic sink sets in to take up CO 2 and to release O 2 . Then the CO 2 mixing ratios decline rapidly during the May-July growing season and reaching a minimum in August. Afterwards, the CO 2 level rises slowly during September-November because of ecosystem respiration and approaches the maximum to start the seasonal cycle again. O 2 is not exactly six-months out of phase with CO 2 . It reaches maximal values in the period end of August until the beginning of September (one months later than the CO 2 minimum) and minimal ones in March. This offset in timing of seasonal extremes between CO 2 and O 2 is due to the ocean that affects O 2 stronger than CO 2 . Thermal CO 2 outgassing is opposite in sign to the marine photosynthesis and respiration CO 2 flux thus leading to only a small seasonal variation in CO 2 air-sea exchange in addition to the dampening effect of the ocean's carbonate chemistry.
On the contrary, the thermal effect and the biological activity act synchronously for the O 2 air-sea exchange, reinforcing the O 2 seasonal cycle (Keeling and Shertz, 1992;Bender et al., 1996). Moreover the O 2 air-sea exchange takes place on shorter times (few weeks) compared to the CO 2 air-sea exchange (1 yr) (Keeling et al., 1993).
The CO 2 and O 2 time series at Jungfraujoch can be separated in three components: first the long term trend covering variations on timescales of years to decades. Second, the mean seasonal component which represents the variations related to the biospheric cycle of photosynthesis and respiration and to a lesser extent to marine CO 2 and O 2 exchanges. Finally a short term component including hourly to weekly variations mainly associated with meteorological variability and corresponding shifts in transport patterns.

Background corrected data
Since our LPDM simulations only account for the recent history (past five days) of the air measured at Jungfraujoch a background corrected record is calculated. The background was computed using an iterative procedure: 30 days running means and their standard deviations were calculated. All data points above and below 2 σ (red band in Fig. 3) were then rejected and the procedure was repeated until convergence (in particular 26 % of CO 2 values and 32 % of O 2 values were rejected). The resulting background was then subtracted from the original data producing the background subtracted records of CO 2 , O 2 and APO as shown in Fig. 5. This record represents the third, short-term, component of variability mentioned above, and is used in the following to identify the source and sink regions of CO 2 and O 2 for high frequency anomalies.

Seasonal cycle
The mean seasonal cycle for the Jungfraujoch in situ measurements (displayed in Fig. 4) was derived from the background data by subtracting first the long-term trend (for every background value the corresponding trend value was subtracted). The trend was calculated according to the method of Nakazawa et al. (1997), using a smoothing cubic spline as stated also in Valentino et al. (2008) and Uglietti et al. (2008). The resulting values were then averaged over the monthly means of the different years (Morimoto et al., 2003). The difference between the CO 2 summer minimum and the winter maximum (seasonal amplitude) is 8.7 ± 0.2 ppm, while the opposite variation in the O 2 signal is 80 ± 13 per meg and in APO 23 ± 15 per meg, respectively. These amplitudes are smaller compared to other sites in Europe (CarboEurope atmospheric measurement sites) because Jungfraujoch is at a high altitude where the dilution effect plays an important role, diminishing the amplitude. Moreover Jungfraujoch is less influenced by the terrestrial biosphere (especially areas covered by forests) compared to the other remote sites in Europe (i.e. tall towers).

Long term and short term variations
The trends were calculated first by subtracting the mean seasonal cycle (a fit curve with two harmonics was computed as seasonal cycle by means of the method described by Nakazawa et al., 1997) from the observations and then computing the linear trend for every year. The online record (Fig. 3) is relatively short in time (January 2005-June 2009) to draw conclusive statements about the long term trend. Nevertheless we calculated a mean annual CO 2 growth rate, by computing the linear trends year by year and not using the smoothing cubic spline as for the entire trend line. The resulting mean annual CO 2 growth rate corresponds to 1.89 ppm yr −1 which is in agreement with the global growth rate (1.91 ppm yr −1 ) calculated by the IPCC for the decade 1995 to 2005 (Forster et al., 2007). Our investigations focussed on the short-term variations as displayed in Fig. 5. In the following section, model experiments are presented for data interpretation of classified short-term variations based on a CO 2 and O 2 , or APO.

Model experiments
In order to better understand the daily to weekly CO 2 and O 2 variations with regard to their underlying processes and source areas, it is helpful to cross-plot them as done in Fig. 6 (O 2 versus CO 2 (left) and APO versus CO 2 (right)). In the following we focus on the interpretation of specific subsets of points in the scatter diagram: extreme events of high CO 2 are first investigated in Sect. 4.1 in the form of case studies. A more statistical analysis is then presented in Sect. 4.2 for large clustered subsets of APO versus CO 2 . In each case the interpretation of the data is supported by an analysis of air mass origin by means of relative footprint maps.

Extreme events of high CO 2
A number of particularly high CO 2 concentrations (>10 ppm above average) were recorded at Jungfraujoch as documented in Fig. 5. In the following we focus on two high CO 2 concentration events showing very different relative footprints.

Case 1
The first was observed in the period from 24 to 27 February 2005. According to the in situ measurements CO 2 concentrations began to increase in the morning of 24 February reaching a maximum of 395.75 ppm at around 21:00 UTC and decreased from the morning of 25 February.  Generally, carbonaceous aerosols (BC) are emitted by incomplete combustion processes such as fossil fuel combustion, wood burning, charcoal combustion or natural fires (Stoffyn et al., 1997;Castro et al., 1999). As reported in previous studies (Baltensperger et al., 1991; occasionally large aerosol and BC peaks attributed to long-range transport events (e.g. forest fires) occur at Jungfraujoch. In general BC concentrations at Jungfraujoch exhibit a seasonal cycle with lower values in winter and a maximum in summer, due to the injection of air from the more polluted planetary boundary layer during summertime (Weingartner et al., 1999;Henne et al., 2005b).
In the framework of the GAW aerosol monitoring program several measurement campaigns have been conducted at Jungfraujoch in the period 2004-2005 to characterize the physical and chemical properties of tropospheric aerosols. On 25 February the concentrations of the chemical species analyzed (sulfate, nitrate, ammonium, organic matter, BC) were exceptionally high for the winter season and more reminiscent to typical summer values (Cozic et al., 2008).
The relative footprint map (Fig. 7) for the considered period shows high residence time of air masses above Eastern Europe and Germany. Actually MeteoSwiss reported for that period an unusual phase of cold weather conditions with repeated surges of arctic air hitting the Alps associated with light snowfall recorded at several weather stations on the Northern side of the Alps. The Southern side of the Alps was protected from these cold fronts and experienced almost no precipitation (Bader and Schlegel, 2005). This persistent weather situation appears to have induced large accumulation Fig. 7. Relative footprint map of trajectories associated with exceptionally high CO 2 concentrations (extreme event, case 1) recorded with the continuous measurements system at Jungfraujoch between 24 and 27 February 2005. Air masses originate from northeastern Europe. Colors correspond to the ratio of residence times over specific grid-cells relative to the average residence time of all trajectories, with yellow to red colors indicating above average and blue colors below average residence times. E.g. a value of 10 denotes that the selected trajectories had a 10 times higher probability to have passed over a particular grid cell than any arbitrary trajectory released from Jungfraujoch of trace gases in these air masses as they moved from the north passing over heavily industrialized countries before they were lifted by the frontal activity and finally transported to Jungfraujoch.

Case 2
The second high concentration event discussed here occurred between 21 and 24 November 2007. Also in this case the high values were recorded with both techniques, i.e. the continuous measurement system and the flask sampling. The maximum concentration, recorded in the morning of 23 November around 08:00 UTC, was 397.4 ppm. The concentration started to increase during the afternoon of 21 November and returned to the mean concentration of 384 ppm in the morning of 24 November.
The corresponding relative footprint (Fig. 8, left) shows air masses originating from the Atlantic Ocean and North Africa and passing with high residence times over the Mediterranean Sea.
Nevertheless vertical transport to Jungfraujoch during south Foehn events, which is typically associated with precipitation and convection along the southern flank of the Alps, is not well resolved by the ECMWF model at the coarse resolution of only 1 • ×1 • as used here. We therefore also analyzed trajectories based on much higher resolution (7 km×7 km) wind fields of the COSMO model of MeteoSwiss for this day. They reveal rapid vertical ascent from the Po Valley and the Italian peninsula within only one day before reaching Jungfraujoch (not shown).
The high CO 2 concentrations recorded for these days were also associated with relatively high CO values (as observed by the NABEL network). These high trace gas concentrations could be explained by advection from the Po valley due to a typical south Foehn event. However, air reaching measurement sites at the Northern edge of the Alps often originates from altitudes of about 2000 m a.s.l. in the south and not necessarily from low altitude in the boundary layer (Seibert, 1990;Campana et al., 2005). Consequently, there are south Foehn events which transport highly polluted air from the Po Basin to the Jungfraujoch and others with no sign of pollution when the origin was probably above the boundary layer .
In this study case it is very likely that the high CO 2 concentrations are caused by contact of these air masses with polluted air above the Po valley before arriving at the Jungfraujoch (Forrer et al., 2000).
Indeed Meteo Swiss (Fig. 8, right) reported for those days (21-23 November 2007) the formation of a depression over the Bay of Biscay which has sparked an inflow from the southwest and south with the addition of moist air into the Southern rim of the Alps. This has established a classical Foehn situation with substantial precipitation amounts on the Southern side of the Alps (Bader and Croci-Maspoli, 2007). Moreover the analysis of wind speed and directions recorded at Jungfraujoch by Meteo Swiss confirms the south flow showing that the wind direction changed abruptly on the 18 from north to south until 25 of November, when the wind direction changed back to north.
In contrast to case 1 the BC measurements at Jungfraujoch did not show any anomalous values during this event. It is likely that, in contrast to CO 2 and CO, the aerosols have been washed out by precipitation. This increase of CO 2 without a simultaneous increase of BC may thus be interpreted as a typical indicator of south Foehn events because during such events aerosols tend to be washed out by precipitation.

Origin of air masses for different CO 2 and APO concentrations subsets
This section investigates the influence of different air mass origins on CO 2 and O 2 concentrations at Jungfraujoch in a more statistical way, by analyzing the combined footprints of a a large number of observations corresponding to specific subsets of points in the APO-CO 2 scatter plot. Each subset is represented by a certain range of CO 2 and APO values. Note that the advantage of using APO instead of O 2 is due to the definition of APO that guarantees a subtraction of a mean terrestrial biosphere influence and therefore allows a better distinction of ocean and terrestrial biosphere influences on CO 2 than CO 2 vs. O 2 . The selection of subsets is illustrated in Fig. 9, left panel. We only select subsets of points with outstanding characteristics. The selection is somewhat subjective, and it is a compromise between representing clear (outstanding) signals in trace gas concentrations and including a sufficiently large number of points to obtain robust patterns in the corresponding trajectory footprints.
The box in the middle (Fig. 9, left panel, subset 1) corresponds to background values. All points are within the 2σ bands of CO 2 (±2.71 ppm) as shown in Fig. 3 (red band) and APO (±37 per meg). The other five subsets are defined as follows: points with enhanced CO 2 and reduced APO (Fig. 9, left panel, subset 2); points with low CO 2 and low APO (subset 3); points with low CO 2 but enhanced APO (subset 4); points with high APO but normal CO 2 (subset 5); points with regular CO 2 and low APO (subset 6). In the following all selected subsets defined above are discussed. The occurrence of specific subsets follows a seasonal cycle as indicated in Fig. 9, right panel. Air masses enhanced in CO 2 are preferentially observed in winter, those depleted in CO 2 in summer. Associated with the seasonal cycle in air-sea exchange is a seasonal cycle in APO with a tendency of negative anomalies occurring in spring and positive anomalies in autumn. The fraction of points per subset is: Subset 1: 62 %; Subset 2: 2.0 %; Subset 3: 2.9 % (1.7 % if restricted to summer months); Subset 4: 8.9 %; Subset 5: 1.7 %; Subset 6: 11.6 %. Remaining points not belonging to a subset: 11 %. Quite a large fraction is thus in the background cluster, which is actually explaining that that most of the time the Jungfraujoch observes undisturbed background air.

The background cluster (subset 1)
By applying the back trajectories analysis to background values (all data within ±2σ ), the relative footprint map shown in Fig. 10   (blue colours), indicating that these air masses had less than average contact with the European boundary layer. This is expected for background values and demonstrates the success of the classification based on FLEXPART simulations for this class. There seem to be somewhat more air masses classified as background in both spring and autumn (mainly October), possibly because these are transition seasons for the biosphere with neither large emissions nor uptake. There is also a north-south gradient as documented by the colours. Background air masses from the north tend to have been even less in contact with the PBL than those from the south. This is understandable from a meteorological point of view since advection of background air from the north is typically associated with anticyclonic flow and therefore subsidence. Such conditions appear to be ideal to observe undisturbed free tropospheric air.

Enhanced CO 2 and reduced APO and O 2 (subset 2)
The high CO 2 and low O 2 concentrations recorded at Jungfraujoch occur generally in wintertime because of increases due to the lower biospheric activity (more O 2 is used for respiration) and higher anthropogenic CO 2 emissions in Europe due to the burning of different fossil fuels (solid, liquid or gas) (Rotty, 1987), thus enhancing the decrease of atmospheric oxygen. In wintertime the boundary layer is mostly located below the altitude of Jungfraujoch and the high CO 2 concentrations cannot be explained by thermal uplift of polluted air masses from the valleys surrounding Jungfraujoch. These high concentrations are thus likely associated with air masses which had resided in the polluted boundary layer for extended periods of time and were then transported to the free troposphere probably in connection with frontal activity. The left panel of Fig. 11 represents all CO 2 pollution events which mostly but not exclusively occurred between October and April. Since the near-surface residence time of air masses reaching Jungfraujoch is generally about 26 % lower in winter than in summer a winter only map was calculated (Fig. 11, right) in which the total wintertime footprint served as reference for the computation of the relative footprint. This figure emphasizes the enhanced residence times of these air masses over the eastern parts of Europe (Poland and Germany). Thus the high CO 2 concentrations may be explained by the passage of air masses above polluted areas in those countries. Therefore, from the EDGAR emission map (Fig. 1) these areas can be considered as potential source regions for polluted air masses reaching Jungfraujoch. The elongated, curved patterns of enhanced residence times (yellow to red areas in Fig. 11, right) suggest that transport associated with such events typically followed a cyclonic path supporting the hypothesis of the importance of frontal lifting. In both case studies presented earlier, the upward transport to Jungfraujoch was indeed connected with fronts.

Low CO 2 and low APO (subset 3)
Air masses with low CO 2 and comparatively low O 2 are almost exclusively observed in summer and describe air masses advected from Eastern Europe (Fig. 12). The low CO 2 concentration is an effect of plant CO 2 uptake, but O 2 in these air masses is not enhanced as would be expected from photosynthesis. Therefore, a strong oxygen sink is needed. To our knowledge four options may be considered for this sink: (i) fossil fuel consumption, (ii) stratosphere-troposphere exchange (STE), (iii) atmosphere-ocean exchange and (iv) oxidation of soil organic material. Option (i) is rated as improbable considering that the percentage of fossil fuel emissions ( Fig. 1) compared to the natural carbon exchange is low its influence on the oxygen budget is limited despite the higher carbon oxidation ratio (−1.39 mol O 2 /mol CO 2 ). To option (ii), it has recently been reported that O 2 /N 2 ratios are enriched with increasing altitude (Ishidoya et al., 2008) associated with decreased CO 2 concentrations (Gurk et al., 2008). This behaviour is explained by altitude dependent air age. Hence a reduced STE would lead to an apparent O 2 sink and CO 2 source due to delayed upward transport of ground-based emission signals. If this option would be the major cause then ozone measured at Jungfraujoch should express lower values as reaction to this STE reduction, yet ozone shows a summer maximum. Therefore, this option seems also improbable. Option (iii): Atmosphere-ocean exchange is important for atmospheric O 2 concentrations. In summer, however, we would expect increased outgassing from the ocean due the increased temperature. This is exactly opposite (neglecting the marine biosphere activity) to what is needed. Furthermore, air masses for subset 3 are dominantly originating over the continent with an actively growing biosphere (low CO 2 , high O 2 ). Yet, a slight timing difference of the temperature induced ocean source for O 2 in summer compared to the terrestrial biosphere O 2 source could lead to an ocean-continent gradient. The observed low CO 2 , O 2 and APO values for subset 3 might be the result of such a gradient with advected air masses from the continent that have not been in contact with the ocean surface for an extended period of time (weeks) and therefore have O 2 (APO) values which are lagging behind the normal 8030 C. Uglietti et al.: European source and flow patterns Fig. 13. Relative footprint map corresponding to low CO 2 but high APO and O 2 (subset 4). See Fig. 10 for further details seasonal cycle observed at Jungfraujoch (Fig. 4). Note that there is a significant increase in O 2 as well as APO from June to August such that this delay would indeed lead to anomalously low APO making this option highly probable. Finally, option (iv) might be acting in parallel since dry summer soils consume more oxygen in order to oxidise soil organic material or released methane from wetlands than it is produced from photosynthesis (Pezeshki and Delaune, 2002;Cuna et al., 2008). Whether this option can play a major role is difficult to judge due to the sparse information about this process of soil oxidation.

Low CO 2 and high APO (subset 4)
As shown in Fig. 4, the average CO 2 background concentration decreases during summertime due to photosynthesis whereas the background O 2 increases (ocean outgassing, land and ocean biosphere contribution). The subset of points with low CO 2 and high APO would thus be consistent with air masses influenced by CO 2 uptake by the biosphere combined with increased ocean outgassing. The footprint map generated from this cluster (Fig. 13) indeed consists predominantly of summer events, but a high residence time over the the North Atlantic is not evident. Rather, these air masses tend to frequently originate from high latitudes with somewhat enhanced residence time over Scandinavia, the Baltic countries and the East Sea. The low CO 2 values are in line with the fact that these regions have comparatively low fossil fuel CO 2 emissions (see CO 2 emission map, Fig. 1) but significant plant/forest-covered areas. The reason for the enhanced APO in these air masses, however, is not clear and needs further investigation.

High APO (corresponding to high O 2 ) and normal CO 2 (subset 5)
The subset with regular CO 2 and high APO values (corresponding also to high O 2 values, subset 5) is associated with air masses advected from the Western North Atlantic Ocean, in particular during late autumn, winter into early spring (from November to April). The corresponding relative footprint map is shown in Fig. 14. The main provenance of air masses is from the ocean, thus supporting the view of marine biospheric activity being responsible for high APO and oxygen concentrations measured at Jungfraujoch. To support our view of marine air masses being responsible for high APO (and high O 2 ) concentrations at Jungfraujoch, we looked at the monthly maps of chlorophyll-a for the investigated period (January 2005-June 2009) available from the GlobColour project (http://www.globcolour.info/). The regions around Europe with highest marine photosynthesis rates are the Baltic Sea and the Atlantic Ocean in the vicinity of the coasts of Mauritania and Morocco. The maximum concentration of chlorophyll-a along the Western African coast is around 5-10 mg m −3 and occurs in general between February and April. Therefore, it can be argued that the high O 2 concentrations measured at Jungfraujoch could be influenced by the marine biospheric activity especially in the Atlantic Ocean and particularly in spring as shown in the footprint map.

Regular CO 2 and low APO (subset 6)
Most of the events associated with regular CO 2 and low O 2 (low APO) concentrations were observed in wintertime. The relative footprint map displayed in Fig. 15 shows a relative high residence time of air masses over the Mediterranean Sea and over Northern Africa. A possible mechanism for our observation could be an uptake of atmospheric oxygen by up-welling waters in the (Eastern) Mediterranean Sea under-saturated in oxygen (Kress and Herut, 2001). This is in line with a deepening of the thermocline during winter time bringing deeper waters to the surface. Indications that such O 2 uptake might be occurring at least periodically is given in a recent publication (Van Der Laan-Luijkx et al., 2010).  understand the processes responsible for their variations and to identify main source and sink regions. For this purpose, footprints (maps of air residence time near the surface) were computed for each 3-hour measurement interval based on 5-days backward simulations using the Lagrangian particle dispersion model (LPDM) FLEXPART driven by meteorological data from ECMWF.

Continuous
In general, footprints associated with particularly high CO 2 concentrations show transport of air masses from North Eastern Europe (Poland and Germany) or passing over densely populated and heavily industrialized areas like the Ruhr area in Germany and the Po Valley in Northern Italy. Transport from the Po Valley is mainly connected to south Foehn events. A detailed footprint analysis of clustered background corrected CO 2 and APO values was performed in order to trace source and sink areas as well as underlying processes resonsible for observed short-term variations. A selection of six clusters based on a scatter-plot of APO versus CO 2 allowed not only a spatial distinction (major source and sink areas) but also seasonal differentiation (winter, summer). Short-term variations with low CO 2 values, occuring in spring and summer, exhibit clearer characteristics than those with medium or high CO 2 (autumn, winter and spring). Additional work is needed to fully understand this results. Finally, this study demonstrates the great use of LPDM simulations to support the classification of air masses and to explain short-term fluctuations in the CO 2 and O 2 observation records. The addition of more atmospheric trace gases, such N 2 O, CO and CH 4 would further constrain the influence of air mass transport on measured concentrations. In future studies it would also be desirable to use wind fields at even higher resolution than used here to better resolve the flow pattern over the complex topography of the Alps.
Stations Jungfraujoch and Gornergrat for supporting our research, in particular we are thankful to the custodians at Jungfraujoch who took the flask samples and were looking after the online instrument.
Edited by: M. Heimann