Fourteen months of on-line measurements of the non-refractory submicron aerosol at the Jungfraujoch (3580 m a.s.l.) – chemical composition, origins and organic aerosol sources

. Chemically resolved (organic, nitrate, sulfate, ammonium) data of non-refractory submicron (NR-PM 1 ) aerosol from the ﬁrst long-term deployment (27 July 2012 to 02 October 2013) of a time-of-ﬂight aerosol chemical speciation monitor (ToF-ACSM) at the Swiss high-altitude site Jungfraujoch (JFJ; 3580 m a.s.l.) are presented. Besides total mass loadings, diurnal variations and relative species contributions during the different meteorological seasons, origin and sources of organic aerosol (OA) are discussed. Backward transport simulations show that the the from regions lowest were from apportionment for each the The contribution of local primary organic aerosols was subtracted from the mass spectra before calculation of the fractions. It is noted, that the increased width of the distribution in winter and autumn is due to smaller signal-to-noise ratio during these seasons.


Introduction
Many environmental topics which are becoming increasingly important in the future are closely linked to aerosols, e.g.climate change on global (Charlson et al., 1992;Lohmann and Feichter, 2005;Boucher et al., 2013) and regional scales (Ramanathan et al., 2001;Stott, 2003;Shindell and Faluvegi, 2009), or impacts on human health (Seaton et al., 1995;Laden et al., 2000;Cohen et al., 2005;Pope and Dockery, 2006) and Earth's ecosystem (Mahowald, 2011).Therefore a good knowledge of aerosol chemical and physical properties in the atmosphere is an essential scientific task.In situ measurements of these properties provide the basis for a deeper understanding of the aerosol influence on the topics mentioned above.
Knowledge of the aerosol concentrations and properties in the lower FT is important due to its impact on cloud formation and to validate model results from the FT.The high-altitude research station on the Jungfraujoch (JFJ) at 3580 m a.s.l. in the Swiss Alps allows the investigation of free tropospheric aerosol year round due to its elevated location.Nevertheless, vertical transport of polluted air masses from the planetary boundary layer (PBL) is observed occasionally especially in summer (see Lugauer et al., 1998;Henne et al., 2004Henne et al., , 2010)).However, suitable approaches have been applied in the past to discriminate between PBLinfluenced conditions and free tropospheric air masses (see Sect. 2.5).Due to its well-developed infrastructure and unique location many decades of scientific measurements of gas-phase constituents (Zander et al., 2008;Pandey Deolal et al., 2012), meteorology (Appenzeller et al., 2008) and atmospheric aerosols (Baltensperger et al., 1997;Collaud Coen et al., 2007;Cozic et al., 2008;Bukowiecki et al., 2015) have been performed there.Given the absence of local vegetation and small scale of local emissions, the JFJ is an ideal location for the measurement of pollution arriving from all over Europe (Reimann et al., 2008;Uglietti et al., 2011).
An important property of atmospheric aerosol particles is their chemical composition providing insight into their origin and chemical evolution pathways.Aerosol mass spectrometry provides a tool for measuring the chemical composition with high-time resolution as opposed to the collection of filter samples, which are often prone to sampling artefacts.The application of the aerosol mass spectrometer (Jayne et al., 2000, AMS, Aerodyne Research Inc., Billerica, MA, USA) has produced a large database of aerosol chemical composition around the planet (see e.g.Zhang et al., 2007;Jimenez et al., 2009;Crippa et al., 2014).However, because of the complexity and cost of the instrument, most AMS measurements cover only a few days to weeks of sampling.Exceptions are a 1-year continuous AMS measurement in London by Young et al. (2015) and a 3-year data set from Mace Head by Ovadnevaite et al. (2014).
To facilitate long-term mass spectrometric measurements, the aerosol chemical speciation monitor (Ng et al., 2011c, ACSM) was developed.Long-term measurements are essential to capture long-term trends, inter-season variations or other effects taking place on longer timescales.Furthermore, increasing data coverage allows for a more reliable determination of the typical aerosol conditions at a specific site.The ACSM provides a valuable and robust tool for measuring chemical composition of submicron aerosol with time resolutions on the order of minutes to hours on a long-term basis.Several long-term and medium-term ACSM data sets have been presented in the literature: 2 years near Paris (France) (Petit et al., 2015), 1.5 years in central Oklahoma (Parworth et al., 2015), 1 year in Zurich (Switzerland) (Canonaco et al., 2013(Canonaco et al., , 2015)), near Johannesburg (South Africa) (Tiitta et al., 2014;Vakkari et al., 2014), near Barcelona (Spain) (Ripoll et al., 2015;Minguillón et al., 2015), and in the Southeastern United States (Budisulistiorini et al., 2015), between 1.5 and 3.5 months in Tijuana (Mexico) (Takahama et al., 2013), on Crete (Greece) (Bougiatioti et al., 2014), in Atlanta (USA) (Budisulistiorini et al., 2013), in Beijing (China) (Sun et al., 2012) and in Santiago de Chile (Chile) (Carbone et al., 2013).In addition, a large database of long-term ACSM data sets (including the data set presented in this manuscript) will be made available in the framework of the European ACTRIS (Aerosols, Clouds, and Trace gases Research In-fraStructure) project (see stations overview at http://www.psi.ch/acsm-stations/acsm-and-emep-stations) via the EBAS database hosted by the Norwegian Institute for Air Research (NILU, http://ebas.nilu.no/).
In the present manuscript 14 months of data from the first long-term deployment of a time-of-flight aerosol chemical speciation monitor (ToF-ACSM, Fröhlich et al., 2013) are presented.The ToF-ACSM provides the advantage of increased sensitivity compared to the ACSM version using a quadrupole mass spectrometer, making it more suitable for remote locations with low expected mass concentrations like the JFJ.The high time resolution of 10 min allows for a yearround detailed analysis of chemical composition of aerosol, which to our knowledge to date is unique for the high Alpine environment and was made possible only by the development of the ToF-ACSM.Diurnal as well as seasonal cycles are discussed together with the influence of injected air from the PBL and the influence of different source regions on the chemical composition.Furthermore, a closer look at the chemical evolution of the organic aerosol fraction and its sources is presented.

Measurement site
The Sphinx observatory is part of the High Alpine Research Station Jungfraujoch, which is integrated into several measurement networks: amongst others there are the international Global Atmosphere Watch (GAW) programme of the World Meteorological Organisation (WMO), the Swiss National Monitoring Network for Air Pollutants (NABEL) and the European ACTRIS infrastructure project.It is located at an elevation of 3580 m a.s.l.(07 • 59 ′ 02 ′′ E, 46 • 32 ′ 53 ′′ N) on an exposed rock formation on the saddle between the two mountains Jungfrau (4158 m a.s.l.) and Mönch (4107 m a.s.l.).The surrounding topography leads to a channelling of local winds in north-westerly or south-easterly directions (see e.g.Hammer et al., 2014).In the north-west the terrain descends steeply towards the Bernese Oberland with Lake Thun and Lake Brienz, while in the opposite (southeast) direction the descent is less steep over the Valaisian Great Aletsch glacier and the Rhône Valley.While the upper two floors of the Sphinx observatory are reserved solely for scientific purposes the lower levels are accessible to tourists who are the only small but ever growing (866 000 visitors in 2014) source of local pollution.Monthly average outside temperatures are below freezing year round.Appenzeller et al. (2008) reported about 10-20 annual days with minimum daytime temperatures above 0 • C, with an increasing trend (50 % increase for the investigated period between 1961 and 2005).
Measurements took place between 27 July 2012 and 2 October 2013, spanning all seasons (see Fig. 1a and b) and totalling more than one year.Recorded concentrations are in agreement with previous studies investigating chemical composition at the JFJ, making the findings representative of the JFJ aerosol.A recent review compiling the results of 20 years of aerosol measurements at the JFJ was published by Bukowiecki et al. (2015).All data are reported in UTC, which is local time minus 1 h in winter and local time minus 2 h in summer (daylight saving time), respectively.

Instrumentation
The ToF-ACSM (Aerodyne Inc., Billerica, MA, USA and Tofwerk AG, Thun, BE, Switzerland) which is described in detail in Fröhlich et al. (2013), is applied for the chemical composition measurements.Like the quadrupole ACSM (Q-ACSM, Ng et al., 2011c), the ToF-ACSM is based on Aerodyne AMS technology but is smaller, less costly, and optimised for long-term deployment with minimal maintenance.In comparison to the Q-ACSM, the ToF-ACSM features improved mass-to-charge resolution ( M M = 500, enabling separation of some isobaric components in the mass spectrum using peak fitting if desired) and, more importantly for deployments at remote sites, improved sensitivity.Additionally, the ToF-ACSM provides the possibility to span a larger m/z range (up to m/z ≈ 300 Th) than the Q-ACSM where the technical limitation lies around m/z 150.At the JFJ, data were averaged to 10 min (corresponding to an organic aerosol detection limit of 63 ng m −3 according to Fröhlich et al., 2013).
The deployment at the JFJ represented the first field test of the ToF-ACSM instrument.Optimisation and upgrading of the instrument and software continued throughout the deployment, especially during the first half of the year, resulting in occasional breaks in operation (cf.Fig. 1a).Even including these planned downtimes, the ToF-ACSM measured for 73 % of the entire deployment and 97 % during the last 5 months (after the final configuration was reached).
The ToF-ACSM provides quantitative mass spectra of non-refractory (NR) submicron (PM 1 ) aerosol.The operational principles are described in detail elsewhere (Fröhlich et al., 2013) and summarised here in four key steps: separation of particles from gas-phase molecules, flash vaporisation, ionisation of the resulting vapour molecules, and sort-ing of the consequent ions by time-of-flight mass spectrometry.Separation, focusing and PM 1 size cut of the particles is achieved by combination of an aerodynamic lens and a skimmer cone.Because of the low ambient atmospheric pressures at the JFJ (average 655 hPa) the instrument inlet was modified by replacing the standard critical orifice (d = 100 µm) with an orifice that has a diameter of d = 130 µm.This keeps the inlet mass flow rates comparable to those obtained during standard operation at lower elevations.Flash vaporisation of non-refractory species takes place at T = 600 • C on a heated porous tungsten plate and ionisation is performed by electron ionisation at E kin = 70 eV.
Ion fragments are attributed to the main constituents of aerosol particles, organics, nitrate (NO − 3 ), sulfate (SO 2− 4 ), ammonium (NH + 4 ) and chloride (Cl − ) according to the fragmentation table-based analysis developed by Allan et al. (2004).At the JFJ, chloride concentrations were always below the 10 min detection limit of the ToF-ACSM (4 ng m −3 ) and hence chloride is omitted in the following discussion.
All necessary calibrations as detailed in Fröhlich et al. (2013) (signal-to-mass (mIE), NO − 3 relative ionisation efficiency (RIE) of sulfate and ammonium, flow rate) were performed on a monthly to bimonthly schedule.Calibrated RIE values were 3.2 ± 0.4 for NH + 4 and 0.6 ± 0.1 for SO 2− 4 .The mean values were applied throughout the data set.For organic molecules, the standard RIE of 1.4 was applied (Alfarra et al., 2004;Canagaratna et al., 2007).
Previous short-term measurements with ACSM and AMS at the JFJ showed a particle collection efficiency (CE) close to unity (Lanz et al., 2010;Jurányi et al., 2010;Fröhlich et al., 2013).The same CE = 1 was applied to the data of this study.This expectation was confirmed by plotting total mass concentration measured with the ToF-ACSM plus equivalent black carbon (eBC) from optical absorption measurements vs. the total mass estimated from a scanning mobility particle sizer (SMPS; for a detailed set-up description, see Jurányi et al., 2011;Herrmann et al., 2015), which resulted in distributions around the 1 : 1 line during all seasons (see Fig. S1 in the Supplement).For that analysis, periods with high concentrations of refractory particles (e.g.Saharan dust events -see Sect.2.6 for method of separation -or construction work) were excluded from the correlation.To convert SMPS number size distributions into mass, a density ρ part was estimated according to the relative densities of the main species measured by the ToF-ACSM at the JFJ: SO 2− 4 and organics (together almost always > 80 % of total PM 1 ).Densities of 1.8 g cm −3 for sulfate and 1.3 g cm −3 for organics (experimentally determined by Cross et al., 2007) were applied.The high CE close to unity is uncommon as typically a CE around 0.5 is reported for AMS/ACSM instruments because a fraction of NR particles is lost due to bounce off the vaporiser before flash vaporisation.According to Matthew et al. (2008) this bounce depends on particle phase and is reduced if particles are either wet, acidic or containing high fractions of  nitrate.All three conditions are not fulfilled at the JFJ.The reason for the discrepancy is still unknown but it is noted that compared to other atmospheric measurement sites the JFJ exhibits extreme ambient conditions (pressure, temperature).
The ToF-ACSM was connected to the total inlet (see e.g.Weingartner et al., 1999;Henning et al., 2003) which is also used for the aerosol monitoring instrumentation of the GAW programme on site.Heating of the inlet to indoor temperatures (T = 25 • C) avoids build-up of ice and ensures dry aerosol without the use of additional drying equipment due to the always low outside temperatures at the JFJ (annual mean temperature: T mean = −7 • C, highest monthly mean temperature in July: T mean, July = −1.2• C).A bypass flow of 50 cm 3 s −1 in addition to the 2.4 cm 3 s −1 flow of the ToF-ACSM makes particle losses in the inlet line (d = 5 mm; l ≈ 2 m) negligible.
Furthermore, data from a nephelometer (Model 3563, TSI inc., Shoreview, MN, USA), a condensation particle counter (CPC, Model 3772, TSI inc., Shoreview, MN, USA), and an aethalometer (AE31, Magee Scientific, Berkeley, CA, USA) operated at the JFJ in the framework of the GAW activities were used in the context of this study.The default mass attenuation cross section for aerosol on the filter matrix (MAC atn value) from the AE31 manual of 16.6 m 2 g −1 at 880 nm (no loading correction) was applied to determine eBC.MAC atn accounts for instrumental error and optical properties of the aerosol.The latter varies between sites and aerosol type (for details see Weingartner et al. (2003) and Zanatta et al. (2015)).For the determination of the boundary layer influence on the air masses at the JFJ, data from a 222 Rn analyser built by the Australian Nuclear Science and Technology Organisation (ANSTO) and operated by the University of Basel (for more details see Griffiths et al., 2014), NO y data measured by chemiluminescence (CraNO x , Eco Physics, Dürnten, Switzerland, Pandey Deolal et al., 2012) and CO data from a cavity ring-down spectrometer (model G2401, Picarro Inc, Santa Clara, CA, USA, Zellweger et al., 2012) of the NABEL network were used.

Bilinear factor analysis with the multilinear engine (ME-2)
Positive matrix factorisation (PMF, Paatero and Tapper, 1994;Paatero, 1997) facilitates disentanglement of bulk mass spectral time series into several contributing factors by minimisation of the quantity Q, consisting of the sum of the squares of the error-weighted residuals of the model.Especially for the organic part of the aerosol mass spectra recorded with AMS/ACSM instruments this method is very successful (Lanz et al., 2007) and has been applied to a great number of studies around the world (Zhang et al., 2011).
In conventional PMF analyses, rotational ambiguity coupled with limited rotational controls can impede clean factor resolution.In contrast, the multilinear engine implementation (ME-2, Paatero, 1999;Paatero and Hopke, 2009) enables efficient exploration of the entire solution space by the application of specifically designed penalties or restrictions to mathematically viable but physically unrealistic or ambiguous solutions.A simple approach utilises the so-called a value (Lanz et al., 2008;Brown et al., 2012;Canonaco et al., 2013;Crippa et al., 2014).Here a priori information is introduced by fixing one or more anchor spectra from previous campaigns/analyses with varying degree of freedom.The freedom is regulated with the a value which can vary between 0 and 1 and stands for the percentage each m/z signal of the final solution spectra may differ from the anchor.0 means no deviation allowed; 1 means 100 % deviation allowed.The source finder (SoFi, Canonaco et al., 2013) tool version 4.9 for Igor Pro (WaveMetrics Inc., Lake Oswego, OR) was used for ME-2 input preparation and result analysis.

Back dispersion (BD) clustering
For the full measurement period, the history of air masses arriving at the JFJ was calculated using backward simulations of the Lagrangian particle dispersion model FLEX-PART (v9.01, Stohl et al., 2005).Taking into account mean flow, deep convection, and turbulence, the model calculated the backward trajectories (10 days) of 50 000 particles released every 3 h at the location of the JFJ.The model was driven off-line with meteorological data from the European Centre for Medium Range Weather Forecast (ECMWF) at a vertical resolution of 91 levels (138 levels after 25 June 2013) and a horizontal resolution of 0.2 • ×0.2 • for the Alpine area (rest: 1 • ×1 • ) between 4 • W-16 • E and 39-51 • N. Since in the Alps many terrain features are not well resolved by a 0.2 • × 0.2 • resolution, a starting height for the released particles of 3000 m a.s.l. was used, which lies between the real elevation of the JFJ station (3580 m a.s.l.) and the model ground at ∼ 2500 m a.s.l.Previous studies (Keller et al., 2011;Brunner et al., 2012) reported best model performance with this selection.
FLEXPART provided emission sensitivities (footprints) with a horizontal resolution of 0.1 • × 0.1 • , covering the Eu-ropean region shown in Fig. 5).The surface footprints were then categorised according to the clustering technique detailed in Sturm et al. (2013) andPandey Deolal et al. (2014).Here, seven categories were distinguished and can be identified by their main source region, i.e. the region with largest intensity of surface influence (see caption of Fig. 5).

Characterising air masses: influence of the planetary boundary layer
Although the JFJ's elevation above the surrounding low lands is well above typical PBL heights (see e.g.Seidel et al., 2010;Chan and Wood, 2013), frequent transport of polluted air masses from the PBL can be observed, as is the case for many mountainous sites (Rotach and Zardi, 2007).This transport is mainly driven by three processes (Zellweger et al., 2003;Forrer et al., 2000): dynamically by deflection of synoptic winds over mountainous terrains (e.g.Foehn winds), vertical turbulent mixing over flat terrain and advection to the JFJ (e.g.thunderstorms) or thermally by anabatic mountain winds.Especially during the warm season the thermal vertical transport drawing in PBL air from a radius up to ∼ 80 km (Weissmann et al., 2005) leads to frequent increases of aerosol concentrations during the afternoon hours (Henne et al., 2005).This thermal transport results in the creation of a dynamically decoupled injection layer (IL) (Nyeki et al., 2000).
In the literature, several methods for the separation of FT air from IL air have been described.In the context of this study three methods were selected and applied: 222 Rn concentrations as tracer (Griffiths et al., 2014), the ratio of the sum of oxidised nitrogen species (NO y ) to carbon monoxide (CO), i.e.NO y / CO as tracer (Zellweger et al., 2003;Pandey Deolal et al., 2013) and back trajectory analysis using footprint clustering (more details in Sect.2.4). Figure S2 shows an example time series of both 222 Rn and NO y / CO.Both methods assume different signatures in the FT and the IL. 222Rn is naturally emitted from ice-free land surfaces (Yamazawa et al., 2005;Griffiths et al., 2014) and has a halflife of less than 4 days.Hence, elevated 222 Rn concentrations at the JFJ are a good indicator for influence of air masses with boundary layer origin.A threshold of 1.5 Bq m −3 was chosen with periods with 222 Rn concentrations below the threshold categorised as FT conditions.This choice of threshold is in accordance with the recommendations from Griffiths et al. (2014) (0.75-2 Bq m −3 ).The NO y / CO ratio is a suitable proxy for the "age" of an air mass with ratios of 0.1 to 0.16 close to anthropogenic sources and ratios of < 0.01 after a few days of transport (Jaeglé et al., 1998;Stohl et al., 2002).Zellweger et al. (2003) reported NO y / CO ratios of 0.002 to 0.005 for free tropospheric conditions at the JFJ, depending on the season.In this study, a threshold ratio of 0.004 was selected.Both thresholds are in accordance with the thresholds found by Herrmann et al. (2015) at the same site.A comparison of the 222 Rn and NO y / CO time traces in Fig. S2 shows similar behaviour with radon being the slightly stricter criterion -222 Rn: free tropospheric conditions during 41.1 % of the overlapping measurement period, NO y / CO: free tropospheric conditions during 47.7 % of the overlapping measurement period.These fractions are higher than the annual fractions reported by Herrmann et al. (2015) (25.7 and 39.1 %, respectively), which is due to the inclusion of two summers (injection of PBL air is more frequent in summer) and a longer period without measurement in winter 2012/2013 (see Fig. 1a).In contrast, cluster no. 1 of the backward dispersion (BD) analysis identified as FT was only active during 11.2 % of the overlapping measurement period due to selection of BD only with maximal time (10 days) passed since last PBL contact (t PBL ).A more careful exploration of this parameter as it is performed by Herrmann et al. (2015) may allow for a lower t PBL threshold.

Saharan dust events
Transport of dust plumes from the Saharan desert (SDE for Saharan dust event) up to the latitudes of the JFJ happens several times a year (Schwikowski et al., 1995;Collaud Coen et al., 2004;Bukowiecki et al., 2015).These plumes sometimes carry sufficient material with them that the dust can be seen by the naked eye as a brown-reddish layer on the Great Aletsch glacier and all other snow-covered surfaces.According to Collaud Coen et al. ( 2004), the distinctively different size, index of refraction and chemical composition of the SDE particles compared to the usual aerosol at the JFJ make SDEs easily detectable by their optical properties: while typically the exponent of the single scattering albedo (SSA) increases with wavelength, it decreases if mineral dust is present.The SSA is routinely measured by a nephelometer and an aethalometer at the JFJ.During the study period, eight extended SDEs (longer than 12 h) were detected.These periods are highlighted in light brown in Fig. 1b.

Time series and concentrations
The top and middle panels of Fig. 1 show the full time series of organics (green), sulfate (red), nitrate (blue), and ammonium (orange) in the aerosol measured by the ToF-ACSM and the equivalent black carbon (black) concentration time series measured by the aethalometer from 27 July 2012 to 2 October 2013 in relative (top) and absolute (middle) terms.On top of Fig. 1b the periods used for the separation of seasons in the seasonal analyses of the following sections are identified.Saharan dust events are highlighted in light brown in the middle panel.In the summer periods average concentrations of all species were significantly enhanced compared to winter.The two bottom panels of Fig. 1 show expanded views of typical periods during summer (left) and winter (right).In summer the concentrations exhibit a strong variability between night/morning and afternoon/evening.While total PM 1 concentrations in the night and morning usually were below 1 µg m −3 , they increased to several µg m −3 in the evening hours (maximum 9.6 µg m −3 on 1 August 2012, 20:50 UTC).This pattern is caused mainly by thermally induced vertical transport and injection of PBL air up to the altitudes of the JFJ as detailed in Sect.3.1.3.Usually concentrations decrease again during the night.Night-time elevated levels, e.g. 12 to 13 August 2012 (Fig. 1c, left), were caused by larger-scale meteorology.In the example above, a likely reason is a thunderstorm (several instances of remote lightening detected in the evening of the 12th) causing large-scale turbulences, i.e. vertical mixing of air masses which then were subsequently transported to the JFJ site.Additionally, several short-term spike-like increases in the organic fraction can be observed during daytime (also before the vertical mixing sets in).These spikes are attributed to local pollution by tourism or construction-related activities (cf.Sect.3.2.1).It is noted that the resolution of Fig. 1b which shows the whole 14-month measurement period does not allow discrimination of spikes, and therefore average concentrations given in Fig. 2 cannot be estimated by eye from Fig. 1b.There are no afternoon increases in winter and concentrations are typically below 0.3 µg m −3 (less than during night and morning in summer).The daytime spikes are likewise visible, however smaller than in summer due to decreased tourism activity in the cold season.

Seasonal variations
In Fig. 2 the statistics for each season of the species mass concentrations, measured by the ACSM are shown in absolute and relative terms (including eBC).Corresponding concentration values for each species and season are given in Table 1.It is evident that the mean concentrations in every case lie well above the median concentrations due to the strong increases during injection of air masses from the PBL, which results in a skewed distribution of the observations (note the logarithmic scale of the y axes).Generally SO 2− 4 and organics were the dominant species, together comprising 73 to 81 % of the total PM 1 mass over all seasons.NO − 3 , NH + 4 and eBC contributed lower amounts, with none of these species accounting for more than 10 % of the mass in any season.Average contributions as well as absolute mass concentrations agree well with previous short-term AMS studies which reported average NR-PM 1 concentrations of 1.7 µg m −3 in March 2004, 2.5 µg m −3 in July/August 2005 (Cozic et al., 2008) and 1.6 µg m −3 in May 2008 (Lanz et al., 2010).A winter period from February/March 2005 reported in Cozic et al. (2008), which had atypical, spring-like meteorological conditions, featured much higher absolute concentrations (2.5 µg m −3 ) than recorded during winter in the present work (0.4 µg m −3 ).
Organic species dominated the total PM 1 mass, especially during the warm months (summer 2012 and 2013) when transport from the PBL was most abundant.In summer, the The arithmetic mean of each species is indicated by a star and the geometric mean by a hash in the respective boxplot.Mean and median concentration values for each species and season can be found in Table 1.Relative concentrations are given in percent in the pie charts.
contribution of organics to total PM 1 was between 50 and 57 %, while during the rest of the year it remained between 37 and 42 %.Average absolute concentrations of organics were around 1 µg m −3 during summer.In autumn, winter and spring they only amounted to 0.18, 0.16 and 0.20 µg m −3 , respectively.This seasonal pattern of organic mass concentrations indicates that the organics in the FT mainly originate from organics that are emitted or formed in the PBL and mixed into the FT, while SOA production within the FT gives a minor contribution.Sulfate was the dominant PM 1 species in autumn (44 %), while in winter sulfate and organics were equally abundant (39 %).Absolute concentrations of sulfate did not vary as much between winter and summer as other species.This can be explained as follows: the atmospheric lifetime of SO 2 is, due to the low oxidation rate, similar to or longer than the timescale for mixing of PBL air into the FT air (Lee et al., 2011).This results in moderate vertical gradients of SO 2 and thus formation of sulfate within the FT.Consequently, the difference between the sulfate concentration in the PBL and the FT is moderate (cf.Sect.3.1.3),and the seasonal pattern of the sulfate concentration in the FT is weaker than that of particulate species, which are exclusively emitted/formed in the PBL.
The nitrate-to-sulfate ratio observed in this study at the JFJ is low, in accordance with previous results from the JFJ (Henning et al., 2003;Herich et al., 2008) and other high-elevation sites around the world (e.g.Hidden Valley, Nepal, 5050 m a.s.l., Shrestha et al., 1997;Vallot Observatory, France, 4361 m a.s.l., Preunkert et al., 2002; Mauna Loa, USA, 3397 m a.s.l., Galasyn et al., 1987).This can be explained with the faster production of nitrate from NO x compared to sulfate from SO 2 .Therefore, NO x is rapidly depleted with increasing age of air masses such that most nitrate formation occurs within the PBL, whereas nitrate formation within the FT is of minor importance.Accordingly, the ratio between nitrate concentration in the PBL and nitrate concentration in the FT is larger than the ratio for sulfate (cf.Sect.3.1.3),and nitrate at the JFJ shows a distinct seasonal pattern linked to vertical transport.
Indeed, nitrate was generally low (highest average concentrations of 0.12 µg m −3 in the summers of 2012 and 2013), with exceptionally low median values in autumn and winter when transport from the PBL was less frequent.However, the large 90th percentile and a closer look at the full time trace of nitrate in Fig. 1b in winter reveal occasional strong increases of NO − 3 possibly attributed to special synoptic conditions that favour orographically driven vertical transport that can also occur in the colder seasons.
The ammonium mass fraction approximately corresponds to the amount required to neutralise the inorganic cations (sulfate and nitrate).Accordingly, the ammonium shows a small seasonal variability (7-10 % mass fraction) as the inorganic mass fraction varies between 28 and 48 % mass fraction.It is important to note that neutralised aerosol does not imply excess ammonia, as nitric acid only partitions into the particle phase in significant amounts together with ammonia.Therefore, the amount of particulate phase ammonium nitrate may be limited by the availability of either gas-phase ammonia or gas-phase nitric acid.Furthermore, it is noted that recent studies have shown that the molecular ratio, i.e. degree of neutralisation, is not a good proxy for the determination of aerosol pH aside from giving very rough information on whether an aerosol is alkaline or acidic (Hennigan et al., 2015;Guo et al., 2015).
Equivalent black carbon concentrations were between 4 and 6 % year-round except in winter (8 %).During winter the concentrations were lower in general, leading to an increased influence on the total mass of occasional PBL contributions.The correlation of the eBC concentrations with the overall yearly cycle (i.e. higher total PM 1 concentrations with higher outside temperatures; see Fig. S3) presumably is mostly related to increased vertical pollution transport, which is implied by good covariance with nitrate mass.The higher eBC fraction in summer 2013 compared to summer 2012 may in parts be related to increased local construction work activity in 2013 including the operation of diesel engine generators, but also to higher organic concentrations in 2012.
In summary, the seasonal variations of concentration and chemical contribution showed the importance of the injection layer and hence aerosol transport from the PBL in the vicinity of the JFJ for the total measured concentrations at the JFJ.

Diurnal variations
Median and mean diurnal cycles (in µg m −3 ) of organic, ammonium, nitrate, and sulfate for each season are shown in Fig. 3. Thermally induced transport caused increased nitrate and to a lesser extent also organic mass concentrations in the afternoon starting around 11:00 to 12:00 UTC and peaking between 16:00 and 18:00 UTC.This behaviour was strongest during the summer periods (orange: 2012; red: 2013) and also evident in spring (green), but mostly absent in autumn (brown) and winter (blue).
The diurnal pattern shows that injections from the PBL result in substantially increased nitrate and organic particulate mass concentrations.This can be attributed to the fact that they are mainly emitted and formed within the PBL, as already argued in Sect.3.1.1.In contrast, sulfate only exhibits a very weak diurnal pattern.This indicates that the sulfate concentrations in the upper PBL are similar to those in the FT, as it is also produced in the FT (cf.Sect.3.1.1).The relative magnitude of the diurnal cycle of ammonium lies in between those of nitrate and sulfate.
Generally, the concentrations observed during summer 2012 and 2013 were comparable for all species except SO 2− 4 .A partitioning of both summers by air mass origin showed an increased frequency of southerly origins (51 % in summer 2012 vs. 39 % in summer 2013; cf.Fig. S4) at the expense of air mass transport from the north.In Sect.3.1.4it is shown that air masses arriving from the south on average were carrying more sulfate, which leads to higher SO 2− 4 concentrations in summer 2012.Mean and median show the same trends except in autumn 2012 where the mean diurnal cycles of organics and nitrate showed a strong increase during daytime, while no such increase is seen in the median.This increase stems only from 1 day (14 October 2012) when especially high nitrate concentrations (the highest values of the whole campaign, up to 4 µg m −3 ) and increased concentrations of all other species were measured.This was attributed to non-thermally induced transport of air masses from the valley floor (frontal lifting

Free tropospheric concentrations
Three methods to distinguish periods with free tropospheric air masses at the JFJ from periods with recent influence of the PBL were compared, using the ToF-ACSM data set to explore the chemical composition of FT air. Figure 4a shows the median concentrations during periods independently classified as free tropospheric conditions by the three criteria introduced in Sect.2.5: 222 Rn, NO y / CO and BD clustering.All three methods yielded similar median concentrations comparable to the mean concentrations in winter: 0.10-0.12µg m −3 of organics, 0.02-0.03µg m −3 of NH + 4 , 0.16-0.18µg m −3 of SO 2− 4 , 0.01 µg m −3 of NO − 3 and 0.02 µg m −3 of eBC.
The boxplots of Fig. 4b-d show both, concentrations during FT conditions (left boxes, highlighted in light blue) and times when PBL aerosol reached the JFJ (non-FT, right boxes, highlighted in light yellow) with mean (star) and median (horizontal line) indicated in the plots and given as numbers above each boxplot.The boxes indicate the interquartile range (IQR) and the whiskers the 10th and 90th percentiles.Again, a comparison of (b), (c) and (d) shows good agreement of all three methods with slightly lower FT concentrations for the BD method.This difference may arise from a more strict FT definition used by the BD method: only BD without significant PBL contact during the last 10 days was assigned to the FT cluster.Each method uses specific cut-off parameters, i.e.FT thresholds which are all subject to uncertainties.A more strict cut-off will lead to lower mean concentrations during both FT and non-FT conditions.
A comparison of the median concentration values ( 222 Rn or NO y / CO criterion) during non-FT and FT conditions yielded ratios of 1.8 for organics and 1.7 for NH + 4 , a higher ratio of 2 to 3 for NO − 3 and a lower ratio close to 1 : 1 for SO 2− 4 .This is consistent with the interpretation of the diurnal patterns (Sect.3.1.2)suggesting that the aerosol from PBL injections has higher mass concentrations and mass fractions of organics and nitrate compared to the FT aerosol.A more detailed look at the mass spectra (local primary or-  The boxes represent the IQR and the whiskers the 10th to 90th percentile range.The pie charts show the relative species concentration of SO 2− 4 (red), NH + 4 (orange), NO − 3 (blue) and organics (green) for FT conditions (e) and for non-FT conditions (f) separated by seasons using NO y / CO for FT discrimination.No NO y / CO ratios were available in spring 2013. 222Rn data were not available after February 2013.ganic aerosol contributions were subtracted from the spectra in advance; cf.Sect.3.2.1)showed a non-FT / FT ratio of the signal at m/z 44 of 1.9, which is comparable to the non-FT / FT ratio for total organic mass.However, m/z 55, 57 and 60, which are mainly related to primary sources, showed larger ratios of around 3 (m/z 55: 2.9; m/z 57: 3.6; m/z 60: 3.3) similar to NO − 3 .These values are similar to m/z 43 (3.0), which is present in primary OA but also in less oxidised SOA.These results indicate an increased importance of primary OA (POA) sources during periods when PBL aerosol was injected into the JFJ region.
Similar conclusions could be drawn from the differences of the (seasonal) relative species fractions between FT (Fig. 4e) and non-FT conditions (Fig. 4f).During autumn and winter the fraction of SO 2− 4 was increased by about 60 % during FT conditions, while the organic fraction was decreased by about 30 %.The NO − 3 fraction showed by far the largest difference with a decrease of about a factor 3 during FT conditions (autumn and winter).The NH + 4 contributions, however, stayed more or less the same.The difference between FT and non-FT conditions was much smaller in both summers, also showing decreased NO − 3 during FT conditions but largely similar contributions of the other species.These observations lead to several conclusions: (1) most nitrate measured at the JFJ was injected from the PBL, (2) significant amounts of organics were injected from the PBL, (3) sulfate loadings in the FT and upper PBL are similar such that PBL influence does not cause distinct diurnal or seasonal patterns, (4) the constant NH + 4 fraction through all seasons is due to the compensating effects of NO − 3 and SO 2− 4 with which NH + 4 is associated, and (5) in summer aerosol levels remained elevated and retained a similar chemical composition also during FT conditions, indicating an overall larger PBL influence on the regional-scale lower free troposphere.S5).Similar findings of decreased nitrate and enhanced sulfate and organic concentrations during FT conditions were reported from the French elevated site Puy-de-Dôme (1465 m a.s.l.) by Freney et al. (2011).

Geographical aerosol origins
Its central European location makes the JFJ a good receptor site for pollutants from different source regions within Europe (and beyond, e.g.SDE).A clustering of air mass origin into seven footprint regions is shown in Fig. 5 together with the corresponding NR-PM 1 aerosol composition and equivalent black carbon.Mean and median concentrations of each cluster are given below the pie charts.For all clusters good statistics with total measuring times between 765 h and 1527 h were achieved.The clusters are numbered from no. 1 to no. 7 and can be characterised by the prevailing cardinal directions/origins (cf.Table 2, including frequencies of each cluster.)It should be noted again that footprints were calculated using meteorological input data with a horizontal resolution of 0.2 • × 0.2 • .While these simulations are able to capture the large-scale synoptic flow, local meso-scale flow patterns in the complex mountain topography are not represented well in the model and are only estimated by the inclusion of enhanced turbulent transport over terrain with large sub-gridscale topography variations (Stohl et al., 2005).Hence, some locally driven transport events may be missed by the model and could be misclassified by the BD clustering.
Highest mean and median total concentrations were recorded from clusters no. 3 to no. 5 (south-west to southeast directions) with average concentrations between 1.59 and 1.63 µg m −3 (median: 0.63-1.03µg m −3 ).Clusters no.6 and no.7 (north-west to north-east directions) showed about 20 % lower mean concentrations between 1.25 and 1.37 µg m −3 (median: 0.55-0.69µg m −3 ) and cluster no. 2 (west) showed ∼ 50 % lower mean concentrations of 0.85 µg m −3 (median: 0.43 µg m −3 ).The free tropospheric cluster no. 1 showed an equally low median (0.40 µg m −3 ) but higher mean, indicating a stronger influence of regional transport phenomena that was not captured by the transport model.A possible explanation is a frequent connection between subsidence (i.e.FT air masses) and anticyclonic conditions leading to fair weather with higher temperatures (i.e.increased thermally induced regional vertical transport).
A comparison of the different aerosol chemical species showed dominant organic aerosol from all regions, with highest OA concentrations from the north-west (63 %) and for the FT cluster (66 %) and lowest OA concentrations from the south (43 %).Sulfate was most abundant in air masses from the southern (no.3: 27 %; no.4: 33 %; no.5: 27 %) and western regions (no.2: 28 %).The lowest sulfate frac-tion was detected from the north-west (17 %).Nitrate fractions were relatively constant (8-10 %) from all regions with slightly lower contributions in clusters no. 1 and no. 2 (6 %).The same was true for eBC which ranged between 3 and 5 %.Large fractions of the total NO − 3 and eBC are expected to be imported from the regional boundary layer (cf.Sect.3.1.3).The level of ammonium depended on the sum of NO − 3 and SO 2− 4 , suggesting mainly ammonium nitrate and ammonium sulfate aerosol in all clusters.
Summarised, PM 1 concentrations in air masses advected from south of the Alps (where regions with known high aerosol loadings like the Po Valley are located; cf.van Donkelaar et al., 2010) were about 20 % higher (BeNeLux, UK, Germany) to 50 % higher (France, Atlantic) than the PM 1 in air masses from north of the Alps (Germany, Scandinavia).These air masses also carried higher fractions of sulfate aerosol, while in the northern air masses larger fractions of organic aerosol were measured.This higher sulfate mass fraction was also observed if only FT air masses in clusters no. 2 to no. 7 using the NO y / CO criterion for additional air mass separation were considered.45 % of the mass was sulfate in FT air masses from clusters no. 3, no. 4 and no. 5, while the mass fraction of sulfate from the northern clusters no.6 and no.7 is at 29 % (cf.pie charts in Fig. 6).The apparent discrepancy between organics dominance in cluster no. 1 and the above analysis where, for FT conditions, sulfate was most abundant, is due to the fact that the footprint analysis is mostly restricted to northern air masses in cluster no. 1, while sulfate is higher with southern air masses.Air masses advected from the west over France and the Atlantic Ocean carried the lowest mean aerosol loading but a sulfate fraction comparable to aerosols from the south.This could be a hint towards influence of shipping emissions off the Atlantic coast.

Organic mass spectra
A large fraction of all m/z channels in the aerosol mass spectra is attributed to organic compounds.However, fragmentation, thermal decomposition and partly also insufficient mass resolution (although the possibility of limited high-resolution peak fitting of ToF-ACSM data has been demonstrated by Fröhlich et al., 2013) cause the loss of information about the parent compounds.Nevertheless, techniques like ME-2 or PMF allow for a further separation according to emission/production sources.Furthermore, a closer look at m/z channels known to be dominated by well-understood ions can shed more light on the evolution and sources of the organic fraction.
Figures 7a-e show the fractions of total OA (the contributions of a minor local primary OA factor -see Sect.3.2.1weresubtracted for total OA beforehand) of the UMR m/z 43, 44, 55, 57 and 60 (f 43 , f 44 , f 55 , f 57 and f 60 ).A clear maximum during the summer periods and the minimum in winter of f 43 can be recognised, while f 44 stayed more or less constant throughout the year.Both of those m/z are typically found in secondary organic aerosols (SOA), with f 43 being more abundant in more recently formed SOA (Chhabra et al., 2010;Pfaffenberger et al., 2013), while f 44 increases with age and oxidation state (Ng et al., 2010(Ng et al., , 2011a)).Addi-tionally, f 43 is an important fragment in many POAs (e.g.cooking, traffic).A constant f 44 points towards similarly aged background aerosols in summer and winter and the increase of f 43 during the summer is attributed to vertical transport of: (1) less oxidised SOA (possibly associated with

FT only
Figure 6.Relative species contributions only during FT conditions separated by air mass origin.Left: southern air masses (clusters no. 3, no. 4 and no. 5 of Fig. 5); right: northern air masses (clusters no.6 and no.7 of Fig. 5).Note that contributions of cluster no. 1 were not considered.The major part of f 57 is usually attributed to hydrocarbonlike emissions e.g. by combustion engines (Zhang et al., 2005).No clear trend is visible for f 57 .It contributed around 0.5 % year-round which suggests a generally low, but constant influence of HOA at the JFJ.Part of the OA fraction at   The contribution of local primary organic aerosols was subtracted from the mass spectra before calculation of the fractions.It is noted, that the increased width of the distribution in winter and autumn is due to smaller signal-to-noise ratio during these seasons.
m/z 55 is also attributed to HOA emissions, but f 55 also has other important sources like cooking (Mohr et al., 2012) or SOA at lower oxidation stages.The f 55 followed the same trend as f 43 and was elevated during the summer period, which was likewise due to increased up-slope transport.The effect of varying local pollution was accounted for by subtracting the local POA identified by ME-2 from the organic mass spectra (for details see Sect.3.2.1).The winter/summer difference was a little more pronounced for f 43 but overall the seasonal behaviour of f 55 was comparable.The mean and median fraction of m/z 60 exceeded the 0.3 % threshold reported to be characteristic of biomass burning only in winter (Cubison et al., 2011).
The IQR of all fractions was increased during the colder months, which is mainly due to the increased influence of noise resulting from low concentrations.In general the anal-ysis of these markers supported the conclusion of increased vertical transport from the PBL during summer and indicated biomass burning influence in winter, which could be either from longer-range transport or local (or regional) vertical mixing.

ME-2 results
ME-2 organic source apportionment was performed separately for all seasons to account for seasonal changes in factor profiles, especially OOA.Seasonal PMF for long-term data sets was also performed before by Minguillón et al. (2015) and Tiitta et al. (2014).Parworth et al. (2015) went one step further and were performing separate PMF analyses on shifting 2-week windows (which they call the "rolling window technique") over a full period of 19 months to capture changes in the OOA and BBOA factors.ME-2 input matrices with an m/z range of 12-120 were prepared.Higher m/z channels were omitted due to low signal to noise at the low mean concentrations prevalent at the JFJ.Errors were calculated accounting for Poisson counting statistics and electronic noise in accordance with Allan et al. (2003).Peaks calculated from m/z 44 were considered in the ME-2 analysis but their uncertainties were adjusted according to the recommendations of Ulbrich et al. (2009) to avoid overweighting.
The daily vertical mixing (during appropriate seasons) poses a challenge to the ME-2 analysis since PMF relies on unique source variabilities for a good separation of sources.Simultaneous advection of aerosols from different sources hence cannot be separated autonomously.The approach applied to deal with this issue can be described as an extended chemical mass balance (CMB, Watson et al., 1997;Ng et al., 2011b).A priori information of organic aerosol source spectra is introduced into the model using the a value technique available in ME-2.In the absence of distinct criteria for the determination of ideal a values due to contemporaneous advection of typical tracers a large range of a values was taken into account for the final result which ultimately constitutes the arithmetic mean of all reasonable solutions using the range of a values indicated in Table S1 in the Supplement.This table also contains reasons for the respective choice of a ranges.Variability (1σ ) of the result as percent of the total source contribution is given in Table S2.There, in addition to the variability due to different a values, also the variability due to different random starting points of the ME-2 (seeds) is taken into account by averaging five seeds for each set of a values.It is noted that although large a value ranges were considered the variability usually is below ±10 %.
Three to five factors could be identified depending on season.During all seasons, a primary hydrocarbon-like OA (HOA), a locally emitted primary OA (POA loc ) and secondary oxygenated OA (OOA), which in summer was split up into LV-OOA I and LV-OOA II, could be identified.Due to the absence of direct volatility measurements the attribution of the OOAs to the LV-OOA family was made indirectly on the basis of spectral similarities to the range shown in Ng et al. (2010).A biomass-burning-related OA (BBOA) was identified when the signal of its most specific fragment peak at m/z 60 was above the background level.This occurred in winter as well as in summer 2013, when three distinct biomass-burning events were identified (presumably due to forestry-related fires; see Fig. S6).Mean HOA and BBOA reference spectra taken from Ng et al. (2011b) were used in the ME-2.The anchor spectrum of POA loc was extracted directly from the data set by a separate unconstrained PMF only on all short-term peaks visible in the organic time series (cf.Fig. 1c).This separate PMF resolved (next to the always present background OOA) only one local primary OA factor, suggesting that the local emissions mainly came from a single local source.Also an additional k-means clustering analysis of all mass spectra of the short-term peaks (with sub-traction of the background by interpolation of the spectrum before and after each peak) only resulted in a single cluster.
In the following analysis POA loc always remained tightly constrained (a = 0.05), while for HOA (and if applicable, BBOA) many solutions with different a values (varied in steps of a = 0.1) were averaged.The OOAs were always completely unconstrained (no a priori mass spectrum).
As an example, the solution (time series and profiles) from summer 2012 is shown in Fig. 8a-c.All other solutions discussed here behave similarly and their time series and profiles can be found in the Supplement (Figs.S6 and S7).The profiles of Fig. 8a show a similarity between HOA and POA loc , while the time series of these two factors are completely independent.HOA follows the general diurnal pattern influenced by vertical transport, similar to eBC (cf.Fig. S10) and POA loc mainly consists of sharp short-term peaks only appearing during daytime (see zoom in Fig. 8c).It is noted that due to the large similarity of HOA and cooking-related (COA) profiles it cannot be excluded that a fraction of the presented HOA may be influenced by COA.
The two OOAs generally followed the same diurnal pattern as HOA with LV-OOA I showing less temporal variability and a higher relative signal at m/z 44 than LV-OOA II.This points towards a more aged oxidised OA.Although LV-OOA I presumably does have the lower volatility since it is more oxidised, the separation in this case was rather driven by origin or formation processes than by volatility since temperatures are always low at the JFJ.The vertical transport pattern was pronounced in the less oxidised LV-OOA II and it showed the same diurnal trends as nitrate (cf.Fig. S8), while the other (LV-OOA I) represents the background OOA and correlated better with sulfate (cf.Fig. S8).This indicates that further oxidation of the OA takes place in the FT.
The seasonal diurnal cycles of HOA, POA loc and OOA are shown in the bottom panels of Fig. 8. HOA in (d) and LV-OOA II in (f) showed afternoon increases which were less pronounced in winter and autumn.LV-OOA I however exhibited a much less pronounced diurnal trend then LV-OOA II confirming its background nature.POA loc (Fig. 8e) was mainly observed between 09:00 and 15:00 LT coinciding with the opening hours of the tourist facilities and the train to the JFJ.In autumn to spring, the concentrations of POA loc outside the opening hours went down almost to zero, while in the peak tourist season in summer low concentrations of POA loc were also found in the non-spiky background during the night, which indicates an imperfect separation of POA loc , i.e. a small overestimation during the summers.The autumn diurnals were dominated by two large non-thermally induced vertical transport events on 14 and 26 October causing the peak between 06:00 and 14:00 UTC in HOA and OOA (see Figs. S6 and S7), similar to Sect.3.1.2.
Relative source contributions for all five seasons between summer 2012 and summer 2013 are depicted in Fig. 9   constituting between 71 % (winter 2012/2013) and 88 % (summer 2013).POA loc contributed 7-8 % in both summers and spring and 12 % in autumn and winter.This means that relative contributions of POA loc were higher in winter when non-local OA was generally low, while the absolute concentrations of POA loc were higher in summer when there were more visitors at the station, i.e. a higher tourist activity.The time series of POA loc overlayed with a shaded background in either blue (days when global radiation is below average) or red (days when global radiation is above average) is shown in the top panel of Fig. S9 for a period of 2 months in summer 2013.The bottom panel shows number concentrations measured with a CPC (time resolution 10 s) for the same period.A strong evidence for a connection of POA loc to tourist activities is the generally lower occurrence of spikes during the (less sunny) "blue" days than the "red" days.This was confirmed by the CPC number concentrations, which also showed fewer spikes on the same days.
Possible sources of local pollution by tourism-related activities are cigarette smoke, fuel combustion by helicopters or snowcats, and emissions of the three on-site restaurants.Previous studies have shown that significant differences be-tween source profiles of these three sources exist (see e.g.Mohr et al., 2009;Faber et al., 2013).The results of PMF and the cluster analysis on the spikes described above suggested a single source for the majority of the peaks.Figure S10 shows a one-week time series (7 to 13 August 2012) of HOA, POA loc , eBC and CPC with high time resolution (for values cf.legend).The fact that eBC did not show any shortterm spikes at all while POA loc and CPC number showed many spikes during daytime allowed for the exclusion of combustion engines as likely POA loc source, since they are expected to emit significant amounts of black carbon as opposed to cigarettes or restaurants (Hildemann et al., 1991;Kleeman et al., 1999).Furthermore, the spikes are too frequent (> 100), too regular and distributed over the whole day with time scales of one to a few minutes to have been emitted by occasional helicopter flybys (less frequent), irregular construction work (more constant emissions) or snowcat activities (typically only in the morning and evening).A correlation of the POA loc profile (see Fig. S11) with known reference profiles of traffic and cooking (COA) resulted in R 2 of 0.74 for HOA from Ng et al. (2011b), R 2 = 0.77 for COA from Crippa et al. (2013) and R 2 = 0.94 for cigarette smoke OA from Faber et al. (2013).Only the m/z range be-tween 45 and 120 was considered for the correlations to avoid uncertainties related to f 44 (for details see Fröhlich et al., 2015).The cigarette smoke OA from Faber et al. (2013) is also shown in Fig. S11.The striking similarity of POA loc and cigarette smoke OA above m/z 44 points towards smoking of cigarettes as the most likely source for the local pollution spikes.This assumption is supported by a concurrent increase in the near-ultraviolet (370 nm) absorption signal of the aethalometer (see Fig. S18).Tobacco smoke is known to have increased optical absorption at the near-ultraviolet wavelength compared to longer wavelengths (Hansen, 2005).Some of the f 44 difference between POA loc and cigarette smoke OA is expected due to the mentioned f 44 uncertainties in the ACSM systems but the POA loc exhibits a particularly low f 44 close to zero.This may indicate that a small fraction of the POA loc is misallocated to one of the other factors, most likely the OOA, partly offsetting the potential overestimation during night-time.In the literature, several cigarette smoke OA spectra were reported showing different amounts of f 44 .The cigarette smoke OA of Northway et al. (2007) shows low f 44 (f 44 /f 43 ≈ 0.08), while Tang et al. (2012) show a f 44 /f 43 ratio of about 0.25 in sidestream environmental tobacco smoke and f 44 /f 43 ≈ 0.5 in mainstream smoke, similar to Faber et al. (2013).However, ME-2 including a cigarette smoke OA anchor profile (or COA anchor profile) did not result in a similarly good separation of the local spikes from the HOA.It is noted that, also under the assumption that the "real" f 44 was to Faber et al. (2013), the difference in the total POA loc concentration was smaller than 10 %, which lies within the estimated uncertainties of the method reported by Fröhlich et al. (2015).Morrical and Zenobi (2002) have previously identified environmental tobacco smoke markers in the particulate phase at the JFJ in a non-quantitative way.
HOA contributions varied between 3 % (summer 2013) and 9 % (winter 2012/2013).The higher relative HOA fractions in the colder months can be explained by the large contribution to the total OA mass of a few vertical transport events containing higher amounts of HOA as opposed to the FT air masses (cf.HOA time series in Fig. S7).In addition, the restriction of the PMF output to positive values can cause a small positive bias at low concentrations.Absolute concentrations, however, were higher in summer due to increased overall influence of vertical transport.Previous PMF studies of a short-term AMS data set from May 2008 found HOA contributions of 6 % (Crippa et al., 2014) and 7 % (Lanz et al., 2010), similar to the 8 % found in this study for spring 2013.
Table 3 shows the HOA-to-BC ratios as a function of season.Values around 0.5 were observed, with the exception of summer 2013 when a significantly lower ratio of 0.27 was measured.A likely reason is a faster oxidation of the semivolatile fraction of HOA reducing the remaining HOA/BC (Hildebrandt et al., 2010).Another possible influence on the changed ratio is increased construction work activity in  2011) and smog chamber studies of a EURO 3 diesel passenger car (0.43 ± 0.17, Chirico et al., 2010).BBOA was only identified in winter 2012/2013 when it comprised 7 % of OA and summer 2013 with about 1 %.The flat diurnal of BBOA (see Fig. S16) and its correlation with the overall trend of OOA in winter suggest long-range transport of BBOA rather than regional vertical transport.A possible source region for BBOA could be Alpine valleys or the surrounding regions of the Alps where wood burning is very important in winter (Lanz et al., 2010;Herich et al., 2014;Zotter et al., 2014).During all other seasons BBOA could not be identified reliably.Even forcing a BBOA factor profile with strict constraints (a ≤ 0.05) into the ME-2 solution always resulted in BBOA mass fractions of 1 % or less.

f 44 vs. f 43 triangle plots
Evolution of OA particles is often represented in the f 44 vs. f 43 space introduced by Ng et al. (2010).There, the relative contribution to total OA by the two major fragments at m/z 44 (mainly CO + 2 ) and m/z 43 (mainly C 2 H 3 O + and C 3 H + 7 ) of ambient AMS data typically fall into a triangular space indicated with red lines in Fig. 10 and temporal changes of the position up and down along the edges of the triangle can be interpreted as ageing, while changes in the horizontal direction indicate anthropogenic (biomass burning: left; traffic: left to middle) or biogenic (right) influence.The frequent injection of OA from the PBL which often experienced a different amount of ageing than the OA in the FT poses an additional important factor governing the f 43 /f 44 differences at the JFJ.Absolute values of f 44 in the ACSM have to be treated with caution due to large instrument dependent variations of f 44 found in an intercomparison of ACSM instruments by Fröhlich et al. (2015).However, the interpretations of the f 44 vs. f 43 triangle still hold for variations within an ACSM instrument.
In panel (a) of Fig. 10  sons.The subtraction of the fresh POA loc which contains no f 44 and about 6 % f 43 (see Fig. S11) leads to an increase of f 44 and a small decrease of f 43 during all seasons, i.e. a movement in the f 44 vs. f 43 space towards the upper left.
It is noted that a potential underestimation of f 44 in POA loc as discussed in Sect.3.2.1 would change the shown f 44 by less than 1 %.The overall picture, however, remains similar: in both summers a larger f 43 was measured compared to the rest of the year indicating a mix including less aged OA with a possible predominant biogenic origin (Canonaco et al., 2015).Of the two summers, the point of summer 2012 lies more to the bottom right in the diagram than summer 2013.This is caused by a stronger contribution of vertical transport in summer 2012 compared to summer 2013 as already shown by the SA (higher LV-OOA II fraction; cf.Fig. 9) and the higher f43 (cf.Fig. 7).The points of autumn 2012 and spring 2013 almost lie on top of each other in the diagram, indicating similar ageing and sources.Before subtraction of POA loc the autumn point showed a lower f 44 caused by the larger amount of POA loc detected compared to spring (12 vs. 7 %; cf.Fig. 9).The winter point is the only one lying on the left side of the triangle, meaning larger influence of the remaining primary sources (anthropogenic wood burning OA Heringa et al., 2011, 2012 or HOA) also after subtraction of POA loc .This is justified by the increased fraction of HOA detected at the JFJ in winter compared to other seasons, which mainly originated from rare vertical transport events in winter dominating the mean values.Additionally, different importance of sources for OOA in winter (more anthropogenic) and in summer (more biogenic) may play a role (see e.g.Andreani-Aksoyoglu et al., 2008;Hoyle et al., 2011).
Panel (b) of Fig. 10 shows the mean f 44 and f 43 in the presence of mainly free tropospheric air masses (white square) and during times with influence from the PBL (black square) after the subtraction of POA loc .The separation was made with the method using NO y / CO which was described in Sect.2.5.FT OA shows smaller f 43 and larger f 44 .This difference is mainly due to a higher degree of oxidation of the OOA in free tropospheric air masses consistent with the absence of sources emitting fresh OA and the longer life time of particles in the FT. Figure S13 shows a 30 % higher mass fraction of LV-OOA II to total OOA when PBL injection is happening compared to FT conditions in summer 2013.
A separation according to air mass origin (using the source area clusters described in Sect.2.4 and displayed in Fig. 5) is shown in Fig. 10c (POA loc subtracted).The very similar position of all points indicated a similar mix of OA sources from all cardinal directions.If the vertical position in Fig. 10c is interpreted as indicator for ageing, the highest degree of ageing was found in air masses from the north and southeast and the lowest degree of ageing in air masses from the west and south-west.However, the differences were small and may lie within the range of uncertainties of the methods applied.The same plot is shown in Fig. S14 after additional subtraction of the non-local POA sources identified in the ME-2 source apportionment (HOA and BBOA).The points moved up even more along the right edge of the triangle and form two clusters: a lower one containing only air masses from the west and south-west and a higher one containing the rest indicating a small difference (less oxidation) in SOA arriving from the Atlantic Ocean.
The position of only the OOA factors determined in the ME-2 analysis are shown in Fig. 10d (full OOA spectra are shown in Fig. S12 of the Supplement).For both summers LV-OOA I, LV-OOA II and the combination of both are plotted separately.Systematic variability of the degree of oxidation of OOA was observed over the course of the year.The combined OOA in summer 2012 showed the lowest f 44 and the A ToF-ACSM was operated at the Jungfraujoch Swiss site at 3580 m a.s.l. for a 14-month period spanning all meteorological seasons.A general trend of elevated PM 1 concentrations in summer, connected with increased PBL air influence due to thermally induced vertical transport and lower concentrations in winter was observed.The largest fraction (37-57 %) was comprised of organic compounds, especially in summer, while in winter and during the transient months comparably high fractions of SO 2− 4 were measured.NO − 3 and NH + 4 contributed lower amounts.All over the year the inorganic anions were neutralised by NH + 4 within the measurement accuracy.The major fraction of NO − 3 all over the year and significant fractions of organics mainly in summer at the JFJ were imported from the PBL by regional vertical transport.In contrast, SO 2− 4 concentrations in the FT were higher and comparable to the concentrations of the PBL injections.Similar transport effects are expected to take place all over the Alpine region, contributing to the exchange of air masses between PBL and FT.For the separation of PBL air from FT air three methods ( 222 Rn, NO y / CO and back trajectory clustering) were applied and compared, yielding similar results.This way seasonal averages of chemically speciated free tropospheric aerosol concentrations could be determined, representing the FT aerosol background over central Europe.The highly time resolved chemical composition data provide the opportunity to explore important aerosol parameters like CCN (cloud condensation nuclei) activity, which represents an important variable for the estimation of aerosol climate effects.
In addition, air masses transported to the JFJ were divided into seven origin regions by backward dispersion cluster analysis.Highest total concentrations were measured in southerly air masses (south-west to south-east) and lowest concentrations in westerly air masses from the Atlantic Ocean.Organics were dominant from all cardinal directions, but the fraction of SO 2− 4 was significantly elevated from the south.
A closer investigation of spectral markers revealed summer maxima of f 43 and f 55 , confirming the influence of vertical transport, while constant f 44 during all seasons indicated highly aged background aerosol.f 60 was increased only in winter indicating the presence of wood burning-related OA at the elevation of the JFJ.
A more detailed seasonal source apportionment using ME-2 (SoFi interface) allowed for a separation of the locally emitted OA fraction due to tourist activities (7-12 %).Comparison with highly time resolved aethalometer and CPC measurements together with a comparison of source profiles suggested cigarette smoke as main source of the local pollution.This observation indicates that increasing numbers of visitors may pose an issue for the validity of atmospheric observations at the JFJ during daytime (no local pollution was detected during the night hours when visitors are ab-sent).Generally, the observed local pollution represents only a minor fraction of the total aerosol mass ( 5 % of NR-PM 1 ) and only the organic fraction is affected, i.e. it does not play a dominant role for most long-term observations performed at the JFJ.Nevertheless, if a separation is desired, the high frequency of the local pollution spikes makes it difficult to exclude all events manually.Since cigarette smoke was identified as likely source of the local pollution a smoking ban on the visitors terrace of the Sphinx research station (see sketch of Fig. S15) should be considered.
Furthermore, a small HOA fraction (3-9 %) advected to the JFJ by vertical transport with HOA / BC ratios around 1 : 2 was found.The prevailing OA source year-round however was the aged, oxygenated OOA (71-88 %) splitting up into a background LV-OOA I and a vertically transported LV-OOA II during summer.f 44 in OOA spectra decreased in the warmer seasons while f 43 increased at the same time.OOA profiles of autumn and spring showed a close similarity.BBOA was only detected in a few rare regional vertical transport events in summer 2013 and continuously during winter 2012/2013 pointing to longer-range regional transport.
The Supplement related to this article is available online at doi:10.5194/acp-15-11373-2015-supplement.

11374R.
Fröhlich et al.: ToF-ACSM measurements of Alpine NR-PM 1 aerosol Figure 1.(a) Time series of relative organic (green), sulfate (red), nitrate (blue), and ammonium (orange) contributions to the total PM 1 aerosol measured by the ToF-ACSM and the eBC (black) concentration time series measured by the aethalometer from 27 July 2012 to 02 October 2013.(b) Same as (a) in absolute concentrations (µg m −3 ).Saharan dust events (SDE) are highlighted in light brown.On top of (b) the periods used in the following seasonal analyses are identified.Gaps between seasons are caused by gaps in the data.(c) Left: mass concentrations in µg m −3 during a typical summer period (7 August 2012 00:00 to 15 August 2012 00:00); right: mass concentrations in µg m −3 during a typical winter period (01 January 2013, 00:00 to 08 January 2013, 00:00).Note the different y scales of the two plots in (c).Times are given in UTC.

Figure 2 .
Figure 2. Pie charts of the relative average concentration and boxplots (in µg m −3 ; line: median; box: interquartile range; whiskers: 10-90th percentile range) separated by seasons(summer 2012, autumn 2012, winter 2012/2013, spring 2013 and summer 2013)  for species measured with the ACSM (organic: green; sulfate: red; nitrate: blue; and ammonium: orange) and eBC derived from optical absorption measurements.The arithmetic mean of each species is indicated by a star and the geometric mean by a hash in the respective boxplot.Mean and median concentration values for each species and season can be found in Table1.Relative concentrations are given in percent in the pie charts.

Figure 3 .
Figure 3. Seasonal diurnal plots for organics, ammonium, nitrate and sulfate (from top to bottom) in µg m −3 .On the left: mean diurnal concentrations; on the right: median diurnal concentrations.The different traces show the seasons: summer 2012 in orange, autumn 2012 in brown, winter 2012/2013 in blue, spring 2013 in green and summer 2013 in red.Times are shown in UTC.

Figure 4 .
Figure 4. (a) Median NR-PM 1 species concentrations during free tropospheric conditions for three different criteria of FT separation: 222 Rn, NO y / CO and backward dispersion clustering.(b, d) Show mean (star) and median (horizontal bar) concentrations for each species measured by the ToF-ACSM during the periods with data for all three different criteria -(b): NO y / CO, (c) 222 Rn, (d) BD clustering.Blue background: FT condition; yellow background: non-FT condition.Absolute mean (top) and median (bottom) values in µg m −3 are given above each plot.The boxes represent the IQR and the whiskers the 10th to 90th percentile range.The pie charts show the relative species concentration of SO 2− 4 (red), NH + 4 (orange), NO − 3 (blue) and organics (green) for FT conditions (e) and for non-FT conditions (f) separated by seasons using NO y / CO for FT discrimination.No NO y / CO ratios were available in spring 2013.222Rn data were not available after February 2013.

Figure 5 .
Figure 5. Aerosol footprint clusters with corresponding relative contributions of PM 1 organics (green), sulfate (red), nitrate (blue), ammonium (orange) and equivalent black carbon (black).Arithmetic mean as well as median total concentrations are given below each pie chart.The number N in each pie chart indicates the number of 10 min averages included in the calculation of the mean and median.The unitless RCTN (residence times by cluster normalised) shows the normalised difference between cluster average and total average residence times.Positive (negative) values indicate increased (decreased) surface sensitivity compared to the overall footprint.The scale extends from −2 to 2 with values larger than 1 (smaller than −1), indicating regions with at least 3 times longer (shorter) residence times; for details, see Sturm et al. (2013).

Figure 7 .
Figure 7. Seasonal variation of fractions of total OA for different mass-to-charge ratios: (a) f 43 , (b) f 44 , (c) f 55 , (d) f 60 .Mean (star) and median (horizontal bar) values are given above each boxplot.The IQR is shown as a box and the 10th to 90th percentile range as whiskers.The contribution of local primary organic aerosols was subtracted from the mass spectra before calculation of the fractions.It is noted, that the increased width of the distribution in winter and autumn is due to smaller signal-to-noise ratio during these seasons.

Figure 8 .
Figure 8. ME-2 example solution for summer 2012 and diurnal plots.(a) Factor profiles of HOA (grey), POA loc (orange), LV-OOA I (dark green) and LV-OOA II (light green).(b) Factor time series of HOA, POA loc , LV-OOA II and LV-OOA I. (c) Zoom of the region indicated in red (about 5 days) in the time series plot.(d) HOA diurnal cycles for all seasons.Orange: summer 2012; brown: autumn 2012; blue: winter 2012/2013; green: spring 2013; red: summer 2013.(e) POA loc diurnal cycles for all seasons.(f) OOA diurnal cycles for all seasons.Dashed line: LV-OOA I; solid line: LV-OOA II.All times are in UTC.

Table 1 .
Mean and median concentrations in µg m −3 during each season.
* Annual geometric/arithmetic mean weighted by seasonal data cover for the period 1 October 2012 to 1 October 2013.

Table 2 .
Transport footprint clusters and main origins.
as pie charts.Secondary OOA dominated the OA all over the year,

Table 3 .
Ratios of HOA to eBC for all seasons.
mean f 44 and f 43 are shown for the full OA mass (circles) and after a subtraction of the locally emitted OA (POA loc , squares), again separated by sea-OA source contributions for all seasons.Orange: POA loc ; grey: HOA; brown: BBOA; dark green: LV-OOA I; light green: LV-OOA II.