The UFS site (47∘25′00′′ N, 10∘58′47′′ E, 2650 m a.s.l.) is located on a steep south-facing slope, approximately 300 m below the summit of Zugspitze (2962 m a.s.l.), which is the highest mountain in the German Alps and represents the first real barrier for northwesterly advection from the Alpine foreland. At UFS, westerly and easterly wind directions dominate due to the local topography . Owing to the trace gas and aerosol measurements that take place at the site, the UFS has reduced its emissions of these substances to a minimum. Nevertheless, the measurements can be influenced by local emissions from the surrounding area, which is a highly frequented tourist area all year round. Local emission sources include nearby cable car stations at the Zugspitze Summit (ZSG) and Zugspitzplatt (ZPLT), which is a gently sloping plateau below the UFS site (Fig. ). Additionally, a skiing area is situated at ZPLT. The large metropolitan area of Munich is also approximately 90 km north of the study site.

In addition to UFS, seven weather stations at different altitudes at a maximum horizontal distance of 11 km from UFS were available and were included in this study (Fig. ). One of these stations is located at ZSG. Another weather station is located on the ZPLT and is surrounded by mountain ridges except to the east where the plateau leads to the narrow, deep Reintal Valley. The Schachen (1830 m a.s.l.), Kreuzalm (1600 m a.s.l.), Felsenkanzel (1250 m a.s.l.), and Brandwiese (900 m a.s.l.) weather stations are part of the KLIMAGRAD project . The Schachen site is situated on a plateau above the middle Reintal in close vicinity of some trees. Kreuzalm is situated in a meadow on a mountain saddle. Brandwiese is located in a meadow surrounded by forest, where a tributary valley reaches the Reintal from the west. At Felsenkanzel, the measurements are made on a steep south-facing slope northwest of the town of Garmisch-Partenkirchen (GAP). The GAP site (720 m a.s.l.) is located on the western periphery of the town where a broad west–east oriented valley turns northeast towards the alpine foreland (Fig. ).

The data set that was analyzed in this study spans a period of 1 year from 22 August 2013 to 21 August 2014, which was selected due to high data availability. Table  gives an overview of the chemical measurements, associated instruments, measurement principles, and research institutions that provided the data. Most of the atmospheric constituents are measured at UFS, including carbon mono- (CO) and dioxide (CO2), methane (CH4), ozone (O3), the sum of oxidized nitrogen species (NOy), nitrogen oxides (NOx=NO+NO2 with NO and NO2 referring to nitrogen mono- and dioxide, respectively), formaldehyde (HCHO), the ambient particle number size distribution (dN(dlog⁡dp)-1) for particle diameters (dp) between 10 and 600 nm, the mass concentration of particulate matter with dp<10 µm (PM10), and equivalent black carbon (eBC). The radioisotopes beryllium-7 (7Be) and radon-222 (222Rn) are sampled at ZSG and on the mountain ridge (2825 m a.s.l.) directly above UFS, respectively. At the remaining sites, only meteorological parameters are recorded.

While most of the chemical data were available as 1 min averages, HCHO and the particle size distribution were provided at 10 min intervals and 222Rn and 7Be were available as 2 h and 12 h averages, respectively (Table ). The HCHO data contained a large data gap between 15 December 2013 and 15 July 2014 and the 222Rn data were only available from 1 January 2014. The 222Rn data were downloaded from the World Data Centre for Greenhouse Gases . Because 7Be attaches to aerosol particles, it is measured by collecting the carrier aerosol using a glass-fiber filter. While NO can be measured directly, NOy and NO2 have to be converted to NO before detection. This conversion of NOy and NO2 is performed using photolysis and gold/CO converters, respectively.

The standard meteorological measurements included air temperature (T), relative humidity (RH), global radiation (Rg), and horizontal wind velocity and direction at all sites. Air pressure (p) was available at the UFS, ZSG, and GAP sites. Additionally, year-round precipitation measurements from an electronic weighing system and a windbreak ring (Sommer Messtechnik, Koblach, Austria) were used at the ZPLT site. At the ZPLT and the four KLIMAGRAD sites, the wind data are measured using propeller anemometers (wind monitor model 05103, Young, Traverse City, USA), whereas at GAP and ZSG, cup anemometers and wind vanes (SK-565 and SK-566, respectively, Thies Clima, Göttingen, Germany) are used. At UFS, an ultrasonic anemometer (model 2-D, Thies Clima, Göttingen, Germany) is used. The standard meteorological data were available as 10 min averages except at UFS (1 min averages).

At GAP, a ceilometer (CHM 15k, Lufft, Fellbach, Germany) provides 15 s averages of up to three aerosol layer heights (ALHs) with a vertical resolution of 15 m using a wavelet algorithm from the manufacturer, which detects strong gradients in the range-corrected attenuated backscatter profile. In the absence of clouds, the signal-to-noise ratio of the ceilometer is typically > 1 up to a height of 4 to 5 km a.g.l. during daytime and up to greater heights at night .

All time stamps were converted to local standard time. The data were aggregated on a 30 min basis while requiring a data availability of ≥ 66 % in each interval unless stated otherwise. The 222Rn and 7Be concentrations were interpolated using nearest neighbor interpolation to match them with the 30 min intervals used for the analysis.

In search of criteria for thermally induced anabatic and katabatic winds that influence the UFS site, the wind directions of all sites were investigated. The wind direction at UFS was not useful because upslope and downslope flows were not clearly visible but appeared to be superimposed by synoptically driven winds due to the relatively exposed location of the UFS site (not shown). The wind directions only exhibited pronounced diurnal patterns that indicated thermally induced mountain winds at GAP, Felsenkanzel, and ZPLT, especially in summer (Fig. ). However, the wind data from Felsenkanzel were not used because of problems with shifts in the sensor orientation. Up-valley and down-valley winds were characterized by northeasterly and southwesterly wind direction sectors at GAP, respectively, and by easterly and west-southwesterly wind direction sectors at ZPLT, respectively (Fig. ). The wind patterns at ZPLT were consistent with the study of , which reported a thermal circulation above the same plateau with an easterly inflow of a few hundred meters depth at daytime and a westerly outflow at night under fair weather conditions in summer. Modeling suggested that the inflow ascended the narrow, deep Reintal Valley before reaching ZPLT.

The wind direction sectors at ZPLT and GAP were used as a criterion for anabatic and katabatic winds (Table ). Additionally, anabatic and katabatic winds were restricted to cases with a weak static stability (Δθv<8 K) and a strong static stability (Δθv≥8 K), respectively, which was justified by the intersect between the frequency distributions of Δθv for the wind direction sectors described above (see Fig. S1b in the Supplement). For katabatic winds, a low wind velocity of <3 m s-1 was required at ZPLT and GAP to exclude cases with a strong synoptic forcing. Additionally, both anabatic and katabatic winds were required to persist for at least 1 h, i.e., two time intervals (Table ).

The detection of these winds and the ceilometer-based MLH at GAP were combined using a mechanistic approach that primarily accounted for thermally induced vertical transport. For this purpose, the MLHGAP was discarded if clouds were detected at a height of < 4 km a.s.l. because low-level clouds increase the uncertainty of the MLH retrieval. The mechanistic approach distinguished between the following three conditions: (a) anabatic winds occur or the UFS site is below MLHGAP, which suggests BL air masses; (b) katabatic winds occur and the UFS site is above MLHGAP, which suggests UFT/SIN or HYBRID air masses; (c) the UFS site is above MLHGAP and the winds are not thermally induced, which also suggests UFT/SIN or HYBRID air masses under the assumption of identical MLHs at UFS and GAP. This approach was limited by the low availability of the MLH (22 % after discarding periods with low-level clouds) but still allowed for a partial verification of the statistical approach.

The PCA converts several input variables into the same number of PCs while the variance and, thus, the importance of the PCs is highest for the first few PCs. In the winter half year, the percentage of explained variance of the PCs decreased strongly from PC1 to PC2 and more slightly towards higher-order PCs. In the summer half year, the explained variance decreased approximately linearly with increasing PC number (Fig. ). The first two PCs explained a total of 60 % to 72 % and 53 % to 58 % of the variance in the winter half year and summer half year, respectively. Hence, most of the variance was maintained when reducing the number of considered PCs from six or seven to two, especially in the winter half year.

The loadings of the first two PCs were similar among the 2-month periods of the winter half year but different and more variable in the summer half year (Fig. ). Although the loadings differed between the 2-month periods, PC1 was always a meaningful indicator for air mass classification, which will be explained in the following. PC2, as well as higher-order PCs, did not always allow for an unambiguous interpretation.

According to Eq. (), variables with a high absolute loading determine the PC scores, i.e., the linear combinations, to a large extent. If, for example, an original variable is much higher than its 2-month mean value and its loading on PC1 is strongly negative then the variable will strongly contribute to a large negative score of PC1. In the winter half year, large negative scores of PC1 primarily reflected high CO, CO2, and CH4 mixing ratios and rather low values of Δθv (i.e., rather low static stability) and the air pressure at GAP (pGAP), which suggested an uplift of BL air masses, while large positive scores of PC1 reflected the opposite characteristics suggesting a subsidence of UFT/SIN air masses (Fig. ). PC2 primarily separated air masses with low q, high O3, and rather high CH4 mixing ratios from air masses with the opposite characteristics in the winter half year. On its own, PC2 was a less reliable indicator for vertical transport because low q and high O3 mixing ratios suggested a subsidence of UFT/SIN air masses while high CH4 mixing ratios suggested an uplift of BL air masses. Possibly, PC2 reflects not only vertical but also horizontal concentration gradients.

In the summer half year, the loading of CO2 on PC1 had the opposite sign compared with the loadings of CH4 and CO on PC1. This observation suggests that low CO2 mixing ratios were generally indicative of BL air masses during the vegetation period due to CO2 removal by photosynthesis. Additionally, pGAP contributed more strongly to the first two PCs than in the winter half year and was anticorrelated with q. Δθv exhibited very small absolute loadings on the first two PCs in the April–May and June–July periods, indicating a poor correlation with the other variables.

In April and May, an uplift of BL air masses was indicated by large negative scores of PC1, which primarily reflected high q, low CO2 mixing ratios, and low pGAP, and by large negative scores of PC2, which primarily reflected high CO and CH4 mixing ratios.

In June and July, PC1 separated cases with high CH4, low CO2, and low pGAP, which indicated an uplift of BL air masses, from cases with opposite characteristics, which indicated a subsidence of UFT/SIN air masses. In the same period, PC2 could not be interpreted unambiguously with respect to vertical transport because high CO mixing ratios, which suggested BL air masses, but also low q and high pGAP, which suggested UFT/SIN air masses, contributed to large negative scores of PC2.

In August and September, PC1 separated cases with high CO and CH4 and low CO2 and Δθv (i.e., low static stability), which was typical for BL air masses, from cases with the opposite characteristics, which was typical for UFT/SIN air masses. Again, PC2 was a less reliable air mass indicator because low pGAP and high q, which suggested an uplift of BL air masses, but also high CO2 mixing ratios, which were typical for UFT/SIN air masses, increased the scores of PC2 in August and September (Fig. ).

The interpretation of the air mass regimes, which were confined by the 33.3 % and 66.6 % percentiles of the first two PCs, differed among the 2-month periods except for the December–January and February–March periods. As an example, the period February–March is discussed in detail using summary statistics. A case study illustrating the classification results in the measured time series is shown in Sect. S3 in the Supplement. The joint data availability of the PCA input variables was 87 % of the time in February–March.

For the regimes I, II, and III, the summary statistics of all seven PCA input variables were consistent with a recent uplift of BL air masses (Fig. a–g): compared with the other regimes, CO, CH4, and CO2 generally exhibited high mixing ratios and q was also relatively high while O3 and pGAP were low or intermediate. Δθv indicated a weakly stable stratification with a median of approximately 8 K for regimes I to III.

The other chemical and meteorological measurements supported the interpretation of regimes I to III (Fig. h–r). Relatively high RH (medians between 80 % and 90 %) and total precipitation values were consistent with a current uplift of air masses. For NOy and NOx, local pollution events (σNOx>0.4 ppb) were excluded, while for 7Be, N90, eBC, and PM10, cases with precipitation at ZPLT were excluded from the summary statistics because local emissions or wet deposition obscure the chemical signature resulting from vertical transport processes. Nevertheless, the data for these variables were available 75 % to 78 % of the time. Regimes I to III exhibited high NOy, NOx, and 222Rn and low 7Be concentrations compared with the other regimes, which emphasizes the strong influence of the BL. The high NOx mixing ratios suggest a strong influence of combustion processes. High 222Rn concentrations are typical for the continental BL because of natural emissions from the ice-free land surface . 7Be concentrations are generally low in the BL due to the formation by cosmic rays in the stratosphere and upper troposphere . The aerosol concentrations, N90, eBC, and PM10 were somewhat elevated for regimes I to III but not as strongly as the gas-phase measurements. This finding could be due to wet deposition during the uplift of the air masses. At ZSG, southeasterly to southerly winds with varying velocities were frequently observed for the three regimes under consideration (see Fig. S2a, d, g in the Supplement). The observations suggest that the influence of BL air masses was mostly caused by low-pressure systems or south foehn events in February and March.

Regimes IV, V, and VI exhibited intermediate CO, CH4, and CO2 mixing ratios and an intermediate pGAP (Fig. a–g). For these air masses, vertical transport processes were not inferable but the intermediate mixing ratios suggest an influence from both the BL and the free troposphere. Thus, the three regimes under consideration were attributed to the HYBRID air mass class, which was in line with intermediate NOy, NOx, and 222Rn concentrations (Fig. l–n).

Regime VII was generally characterized by the lowest CO, CH4, and CO2 mixing ratios and a strongly stable stratification (median Δθv of 15 K). However, low O3 mixing ratios, high q, and an intermediate and strongly variable pGAP indicated that the air masses did not originate from the upper troposphere or stratosphere but from the lower troposphere or the marine boundary layer (Fig. a–g). This interpretation was in line with intermediate 7Be and 222Rn concentrations and low NOy, NOx, N90, eBC, and PM10 concentrations (Fig. l–r). Intermediate RH with a median of 54 % corroborated some recent influence of the free troposphere. The air temperature was high compared with the other regimes (Fig. i), the wind direction at ZSG almost always had a southerly component, and the wind velocity at ZSG showed the highest mean wind speed of 10.22 m s-1 among the regimes (see Fig. S2c in the Supplement). The summary statistics did not allow for a final conclusion on the influence of the marine boundary layer to be drawn; thus, regime VII was attributed to the HYBRID air mass class and the MBL/UFT subclass unless the criteria for LRMD were fulfilled.

clustered back trajectories for Jungfraujoch, another mountain site in the Alps, and demonstrated that some air masses originated from the marine boundary layer above the tropical Atlantic Ocean and were transported to Jungfraujoch inside the free troposphere within 5 to 15 days, resulting in a rather high RH (average of 67 %), high temperatures, and low O3, CO, NOy, and NOx mixing ratios. Air masses with similar trajectories may also reach Zugspitze and represent, at least partly, regime VII.

For regimes VIII and IX, the CO, CH4, and CO2 mixing ratios were generally low but not as low as for regime VII. O3 was intermediate for regime VIII and high for regime IX. In combination with a strongly positive Δθv (median of approximately 15 K), i.e., a strongly stable stratification, and a generally high pGAP, these observations indicated a subsidence of UFT/SIN air masses for the two regimes under consideration unless the criteria for LRMD were fulfilled (Fig. a–g). This interpretation was supported by a generally low RH (medians of 45 % and 25 %), a total precipitation of almost zero, low NOy, NOx, N90, eBC, and PM10 concentrations, and high 7Be concentrations compared with the other regimes (Fig. h–r).

For all regimes except III, the box plots of global radiation (Rg) were similar with medians close to zero (Fig. k), indicating that the air mass characteristics were generally independent of day- and nighttime in February and March.

In the entire year, the data availability limited the percentage of classified cases to 78 % of the time. Data gaps predominantly occurred in the June–July and October–November periods (Fig. ). On average, BL, UFT/SIN, and HYBRID air masses accounted for 31 %, 14 %, and 55 % of the classified cases, respectively. The percentage of UFT/SIN air masses is in reasonable agreement with the study of , in which 13.6 % of the long-term CO2 data from UFS were selected as baseline concentrations using an univariate statistical approach called “adaptive diurnal minimum variation selection”. For earlier CO2 data from ZSG and a pedestrian tunnel, approximately 70 m above the UFS, the percentage of baseline concentrations was 19.5 % and 9.9 %, respectively, according to the same study.

The BL air mass class was attributable to three of the nine air mass regimes in all 2-month periods except June–July and October–November, when the BL class comprised two regimes (Fig. ). This finding implies that air masses are lifted up to UFS with similar frequencies in all seasons although the underlying processes may vary. However, the frequency of the air mass classes was constrained to some extent by using the 33.3 % and 66.6 % percentiles of the first two PCs as thresholds between the air mass regimes. These percentiles were somewhat arbitrary but allowed for a distinction between more unambiguous air masses at both edges of the distribution and more ambiguous air masses in the center of the distribution.

UFT/SIN air masses were attributable to two regimes in the winter half year but only to one regime in the August–September period and to none of the regimes in the April–May and June–July periods. UFT/SIN air masses were only evident if the summary statistics of an air mass regime indicated a dominating influence of subsidence. From April to July, subsidence appeared to be very rare at UFS, such that UFT/SIN air masses could not be isolated by splitting the data at the 33.3 % and 66.6 % percentiles of the first two PCs. This result can be explained by the fact that subsidence in high-pressure systems is counteracted by thermally induced uplift during daytime, especially in the summer half year. However, the data availability in June–July was only 56 %, which calls the long-term representativeness for this period into question. Other statistical classifications at the Jungfraujoch and Mount Bachelor Observatory high-alpine sites indicated that air masses of the undisturbed free troposphere occur year-round but less frequently in summer than in winter . In principle, these air masses do not have to descend but can also be advected horizontally to high-alpine sites. Such cases may be included in the HYBRID class and most likely in the MBL/UFT subclass.

HYBRID air masses including MBL/UFT comprised four to five regimes from August to March and even six and seven regimes in the April–May and June–July periods, respectively. In most of the 2-month periods, MBL/UFT air masses accounted for one regime (Fig. ). Seven LRMD events were identified during the whole year, predominantly from April to July, which is in line with the study of that found 5 to 15 Saharan dust events per year at the Hohenpeissenberg mountain site, approximately 40 km north of Zugspitze.

The ceilometer-based MLH at GAP exhibited a pronounced diurnal cycle with a maximum in the late afternoon or early evening on days with a high total global radiation (not shown). Little diurnal variation in the MLH was observed for days with a low total global radiation. Thermally induced anabatic winds were most frequently observed during summer and daytime, whereas thermally induced katabatic winds occurred most frequently at night and in the first few hours after sunrise (not shown). The patterns described above were expected and confirmed the plausibility of the mechanistic approach.

A total of 25 % of the statistically classified cases were evaluated using the mechanistic approach. In 3 % of the statistically classified cases, the mechanistic approach suggested an influence of the local BL on the UFS, mostly due to thermally induced anabatic winds. It should be noted that the mechanistic approach is not able to detect synoptic uplift processes because low-level clouds and precipitation result in data gaps of the MLHGAP. In 5 % of the statistically classified cases, katabatic winds and a low MLHGAP compared with the UFS level suggested non-BL air masses. In another 17 % of the statistically classified cases, the winds were not thermally induced and the MLHGAP suggested non-BL air masses.

When the mechanistic approach suggested BL air masses, the statistical approach agreed in 56 % of the cases and yielded HYBRID air masses in almost all other cases (Fig. ). This finding indicates poor performance of the statistical approach in the case of thermally induced uplift. However, thermally induced uplift only accounts for a small fraction of the time according to the mechanistic approach. Other uplift processes including non-local uplift were not identifiable with the applied mechanistic approach. LRMD was found in 3 % of the cases, which were mechanistically classified as being influenced by the BL, likely due to an overestimation of the ceilometer-based MLHGAP if the advected dust layer merged with the mixing layer.

In the presence of katabatic winds and a low MLHGAP compared with the UFS level, the UFT/SIN and HYBRID statistical classes including LRMD and MBL/UFT were both considered to agree with the mechanistic approach (illustrated by the thick black line in Fig. ). Thus, the level of agreement was 91 % for katabatic winds and a relatively low MLHGAP, whereby UFT/SIN and HYBRID air masses accounted for approximately half of these cases, respectively. When the MLHGAP was lower than the UFS level and the winds were not thermally induced, the level of agreement between the classifications was a bit lower (86 %) with UFT/SIN and HYBRID air masses accounting for 34 % and 52 %, respectively.

Uncertainties in the mechanistic approach mainly arise from the assumption of identical MLHs at the UFS and GAP sites and the lack of information on non-local uplift followed by horizontal advection of air masses. Uncertainties in the statistical approach primarily result from the lack of objective thresholds between the air mass classes and a varying significance of sources and sinks of the gases used. In view of these uncertainties, a reasonable agreement between the approaches was found when the mechanistic approach indicated non-BL air masses.

To classify air masses in real-time, the scores of the first two PCs can be approximated using the loadings that were calculated in this study. Thus, the same thresholds described here can be applied to determine the regime and the class of the current air mass. To standardize the original variables, the arithmetic mean and standard deviation of the current 2-month period are required but these statistics are only known at the end of the 2-month period. Due to the long-term trends of CO2, CH4, and CO and the interannual variability of all variables, it is advisable not to use the 2-month means and standard deviations from the year investigated in the present study and to estimate these statistics from recent multiyear records. To make use of the known part of the current 2-month period, the mean (μ) and standard deviation (σ) could be updated regularly using weighted averages, 7μ=(1-f)μl+fμc8σ=(1-f)σl+fσc, where μl and σl are the 2-month mean and standard deviation, respectively, estimated from the long-term record; μc and σc are the mean and standard deviation of the known part of the current 2-month period; and f is the time fraction that is known in the current 2-month period. In real-time operational mode, additional uncertainties can arise from uncorrected measurement artifacts.