Pollution Events at the High-Altitude Mountain Site Zugspitze-Schneefernerhaus (2670 m a.s.l.), Germany

Abstract

Within the CO₂ time series measured at the Environmental Research Station Schneefernerhaus (UFS), Germany, as part of the Global Atmospheric Watch (GAW) program, pollution episodes are traced back to local and regional emissions, identified by δ¹³C(CO₂) as well as ratios of CO and CH₄ to CO₂ mixing ratios. Seven episodes of sudden enhancements in the tropospheric CO₂ mixing ratio are identified in the measurements of mixing/isotopic ratios during five winter months from October 2012 to February 2013. The short-term CO₂ variations are closely correlated with changes in CO and CH₄ mixing ratios, achieving mean values of 6.0 ± 0.2 ppb/ppm for CO/CO₂ and 6.0 ± 0.1 ppb/ppm for CH₄/CO₂. The estimated isotopic signature of CO₂ sources (δ_s) ranges between −35‰ and −24‰, with higher values indicating contributions from coal combustion or wood burning, and lower values being the result of natural gas or gasoline. Moving Keeling plots with site-specific data selection criteria are applied to detect these pollution events. Furthermore, the HYSPLIT trajectory model is utilized to identify the trajectories during periods with CO₂ peak events. Short trajectories are found covering Western and Central Europe, while clean air masses flow from the Atlantic Ocean and the Arctic Ocean.

1. Introduction

Since the beginning of the industrial era, atmospheric greenhouse gas (GHG) concentrations have been rapidly increasing due to fossil fuel burning and deforestation [1]. Carbon dioxide (CO₂) is the most important anthropogenic GHG, and its concentration has increased by more than 45% from the pre-industrial level of 278 ppm (parts per million) to 405 ppm in the year 2017 [2]. Continuous atmospheric CO₂ measurements significantly contribute to a better understanding of the global carbon cycle and its effect on the earth’s climate. The Environmental Research Station Schneefernerhaus (UFS) has been part of the Global Atmospheric Watch (GAW) program of the World Meteorological Organization (WMO). Due to its remote and elevated location, the UFS is relatively less influenced by direct anthropogenic emissions and thus measurements of background concentrations there can be regarded as representative of a larger region [3,4]. Nevertheless, the station is occasionally influenced by local-to-regional emissions due to the air mass transport of polluted boundary layer air and/or local effects. These polluted air masses contain signals from continental sources, as indicated by the high variability in trace gas concentrations on time scales of hours and days [5]. In order to improve the retrieval of atmospheric background CO₂ concentrations from measurements at these remote sites, it is crucial to investigate these pollution-affected periods systematically, as well as to identify their source regions.

In order to study sudden increases and short-term variability in atmospheric trace gases at mountain stations, in situ measurements of trace gases with anthropogenic sources as well as statistical trajectory methods have been frequently applied to define the source, identify its origin, and elucidate and characterize the causes of the variation. For instance, Apadula et al. [6] analyzed atmospheric CO₂ concentrations at three Alpine sites (Plateau Rosa and Monte Cimone in Italy, and Zugspitze in Germany) and identified a short-lived episode of large fluctuations in the CO₂ mixing ratios. This was similarly observed at Plateau Rosa and Zugspitze, but with a delay of 12–18 h, as shown by backward trajectory and cluster analysis. Kaiser et al. [7] detected the potential source of nitrogen oxides (NO_x), carbon monoxide (CO), and ozone (O₃) using trajectory residence time statistics at five Alpine GAW stations (Zugspitze/Hohenpeissenberg (Germany), Sonnblick (Austria), Jungfraujoch (Switzerland), and Mt. Krvavec (Slovenia)). Furthermore, Uglietti et al. [8] investigated the transport of air masses with high CO₂ levels to the Jungfraujoch station (JFJ) using backward Lagrangian particle dispersion model simulations, and employed CO₂ and oxygen (O₂) signatures to classify these air masses. Tuzson et al. [9] focused on the pronounced enhancement in the tropospheric CO₂ at Jungfraujoch in February 2009. During this period, four events were captured in which CO and CO₂ mixing ratios were highly correlated as a proxy for fossil fuel CO₂, as well as the CO₂ mixing ratio and isotopic composition of atmospheric CO₂ δ¹³C, using the Keeling method. Ferrarese et al. [10] identified two significantly high CO₂ concentration events at four high-altitude European stations in February 2004 using the backward trajectories of the polluted air masses from the European boundary layer as indicated by CO and O₃. Ghasemifard et al. [4] showed that the high CO₂ mixing ratios in winter (from the year 2012 to 2014) at the UFS originated in the German Ruhr area, the Netherlands, and Lusatia (an area bounded by Germany, Poland, and the Czech Republic) based on CO₂ and δ¹³C measurements, as well as potential source contribution functions (PSCFs).

δ¹³C has been used in numerous studies to gain insight into different aspects of the carbon cycle (e.g., [9,11]). Nevertheless, the isotopic composition of atmospheric CO₂ is related to the atmosphere–biosphere exchange, because the terrestrial ecosystem (photosynthesis/respiration) and air–sea gas exchanges differently discriminate against ¹³C [12,13]. Therefore, the δ¹³C signature is often used to distinguish CO₂ contributions from oceanic, biospheric, and anthropogenic CO₂ emissions [14]. Furthermore, since different CO₂ emitters have different δ¹³C signatures, the carbon isotope ratio enables the identification and quantification of CO₂ emitters and their underlying mechanisms at the local to regional levels [15]. Various ranges of δ¹³C have been obtained for combustion sources: −44‰ to −37‰ for natural gas combustion, −32‰ to −26‰ for gasoline combustion, −27‰ to −24‰ for coal combustion, and −27‰ to −22‰ for wood combustion [14,16,17,18].

It is well known that a large amount of global CO, which is produced by the incomplete oxidation of carbon, originates from anthropogenic sources. Moreover, approximately 70% of the total sources of CO in Europe are anthropogenic [19]. Thus, with CO as a tracer, anthropogenic CO₂ emissions from fossil fuel combustion, and therefore CO₂ emissions from regional sources, can be identified (e.g., [9,20,21,22]). The ratio of atmospheric CO to CO₂ (CO/CO₂) has been utilized to quantify fossil fuel contribution in relation to CO₂ variations, and to determine the type of combustion and burning efficiency; high ratios indicate poor burning and trace traffic emissions [9,23,24].

Methane (CH₄), the second most essential anthropogenic GHG gas after CO₂, has increased from the pre-industrial level of 722 ppb to the present level of 1859 ppb (2017) due to increased emissions from anthropogenic sources [2]. CH₄ is emitted from a variety of natural and anthropogenic sources and also from ground level sources (e.g., ruminants, natural gas leakages, coal mining, or waste deposits). The source of CH₄ in Western Europe is mainly anthropogenic [25]. Though vehicle emissions are an insignificant source on a global scale [26], they can count for up to 30% of total emissions in areas with high traffic density [27]. Several studies have shown that there exist positive correlations between atmospheric CO₂ and CH₄ mixing ratios mostly in the winter (e.g., [28,29]). Based on the assumption that CO₂ and CH₄ originate from similar sources and sinks, the CH₄ to CO₂ emission ratio (CH₄/CO₂) is utilized to filter observed atmospheric CO₂ data influenced by local sources/sinks [30].

In this study, we performed the analysis from 1 October 2012 to 1 March 2013 at the UFS/Zugspitze. During this period, we captured several consecutive short-term pollution events with a strong increase (>~15 ppm) in the measured CO₂ concentration. The period was restricted to these five months of autumn/winter in order to minimize the effect of biological activities (such as photosynthesis and respiration) in the interpretation of pollution signals. It was found that employing the in situ measurements of isotopic signatures of CO₂ using δ¹³C(CO₂), the emission ratios of CO/CO₂ and CH₄/CO₂, and the backward trajectories from the HYSPLIT (Hybrid single-particle Lagrangian integrated trajectory model) provided essential information for the identification, characterization, and source estimation of pollution events, as shown in the subsequent sections.

2. Material and Methods

2.1. Measurement Site

The high Alpine Environmental Research Station Schneefernerhaus (UFS) is located in the northern Alps at an altitude of 2650 m above see level (a.s.l.) (47° 25′ N, 10° 59′ E), nearly 300 m below the summit of Mount Zugspitze, the highest mountain in Germany, and about 90 km southwest of Munich. The UFS is often above the planetary boundary layer (PBL) and receives well-mixed and/or free tropospheric air masses [31]. Nevertheless, UFS is occasionally influenced by regional emissions such as the transport of PBL air masses due to thermally induced flow systems [5,32,33,34], although this issue can be addressed by characterizing the measurements based on their air mass histories and origins. Further detailed site information is given in the GAW station information system [35].

2.2. Analyzers and Sampling Systems

In this study, all instruments were operated at the German Environment Agency (Umweltbundesamt, UBA) laboratory and connected to the same air inlet (2670 m a.s.l.); thus, they measured the same air samples.

We measured CO₂ and δ¹³C(CO₂) with a Picarro G1101-i analyzer (manufactured in 2010 by Picarro Inc., USA, and measuring at a rate of 0.1 Hz) and calibrated them using two standard gases with high and low concentrations (standard 1:350.1 ± 0.5 ppm CO₂ mixing ratio and −3.3‰ ± 0.2‰ δ¹³C; standard 2:503.4 ± 0.5 ppm CO₂ mixing ratio and −20.0‰ ± 1.0‰ δ¹³C) in synthetic air. Details on our air sampling inlet and calibration systems are presented in Ghasemifard et al. [4]. Notably, this analyzer was not upgraded to account for the spectral interferences caused by methane (bias δ¹³C by 0.4‰ CH₄ ppm⁻¹) and water vapor (including water vapor dilution, water vapor pressure broadening, and HDO spectral interference effects). Several studies have evaluated and validated the G1101-i analyzer measurements before and after upgrading in comparison with other measurement devices [4,36,37,38,39,40]. As described by Ghasemifard et al. [4], the measured CO₂ and δ¹³C were post-corrected, in which parallel measured CO₂, H₂O, and CH₄ were employed.

The UBA performed CH₄ measurements with a Picarro EnviroSense 3000i analyzer (using cavity ring-down spectroscopy) and recorded CO data by an Aero-Laser AL5002 (ultraviolet-fluorescence instrument). Further, the UBA carried out calibration and quality assurance for these atmospheric compounds following GAW quality standards. The UBA station standards for interconnection with the international standard reference were reported on the WMO/NOAA-2004 scale for CO and WMO X2004a scale for CH₄ by National Oceanic and Atmospheric Administration (NOAA), Boulder, Colorado.

For this study, we aggregated the data of CO, CH₄, CO₂, and δ¹³C to hourly averages by the statistical program R [41] and employed the R packages Openair [42] and ggplot2 [43] for further analyses.

2.3. Keeling Plot Method

In order to calculate the mean isotopic signature of a source, the Keeling plot method [44,45] links variations in CO₂ and δ¹³C by adding CO₂ from that source to the background atmospheric CO₂ mixing ratio and background isotopic ratio [46]. The mass balance of the total CO₂ is

$$C_{tot}=C_{bg}+C_s$$

(1)

where C_tot, C_bg, and C_s respectively refer to the total measured CO₂ mixing ratio, background CO₂ mixing ratio, and source CO₂ mixing ratio. Given the mass balance,

$${\delta }_{tot}·C_{tot}={\delta }_{bg}·C_{bg}+{\delta }_s·C_s$$

(2)

where δ represents δ¹³C of each CO₂ component. Rearranging and combining Equations (1) and (2) yields

$${\delta }_{tot}=C_{bg}/C_{tot}\left({\delta }_{bg}-{\delta }_s\right)+{\delta }_s$$

(3)

In the case of a single-source pollution event, a plot of the measured (total) δ¹³C versus 1/CO₂ yields a straight line with a y-intercept of ${\delta }_s$. It is crucial to observe that Equation (3) of the Keeling method involves two basic prerequisites. Firstly, the background and isotopic signature of the source need to be constant. Secondly, sources and sinks of different isotopic signatures may not co-occur during the observation [15]. The effect of disregarding these criteria is also graphically depicted in Vardag et al. [46]. Note that the background value does not have to be known in order to calculate the mean source signature [15].

As an example, Figure 1a shows one of the events in this study (E7), while Figure 1b displays the corresponding Keeling plot.

In order to calculate the source signature as an intercept, a linear regression needs to be fitted to the data (e.g., Figure 1b). Miller and Tans [47] compared two models for the regression, Model I (ordinary least-squares regression in which only y-errors are taken into account), and Model II (regression fit; e.g., weighted total least squares (WTLS), which accounts for errors on both x and y [48]). Miller and Tans [47] found that Model I is biased for the intercept, even when the values of R² are high. Therefore, in this paper, we employed WTLS to fit the linear regression in the Keeling method.

Based on the classical Keeling plot method, Vardag et al. [46] presented a method to accurately determine the source signature for a continuous data set (moving Keeling plot approach) following Equation (3). The advantage of this approach is that the prerequisites of the Keeling approach do not have to be checked individually for the entire data set, which would be very tedious for a continuous long-term data set. This method provides correct results only when a number of criteria are fulfilled. These criteria then automatically discard situations in which sources and sinks occur simultaneously and in which the mean source signature or the background value change. Note that the optimal set of parameters was found by Vardag et al. [46] and validated using a simulated data set in Heidelberg. The individual filter criteria and the rationale behind the filter criteria are adopted from Vardag et al. [46] and listed in

Table 1. Filter criteria and rationale based on Vardag et al. [46].

Filter Criteria	Rationale
Monotonous CO2 increase over 5 h	Reasons for a decrease could be a simultaneously occurring sink or a boundary layer break-up, both hurting the prerequisites of a Keeling plot.
Error of intercept < 2‰	Changes of mean source mix due to change of emission characteristics, background or footprint area show a non-linear behavior in a Keeling plot. Thus, choosing a maximum error of the intercept, δS of <2‰, can filter out these situations.

.

Moreover, there is another criterion in Vardag et al. [46] that is not applicable in this study, namely an increase in the CO₂ mixing ratio by more than 5 ppm. Simultaneously occurring sources and sinks bias the retrieved source signature more strongly the smaller the net CO₂ signal is. Therefore, the criterion is chosen to exclude periods where the photosynthetic sink is similarly strong to the total CO₂ sources. Vardag et al. [46] report that a threshold of 5 ppm is sufficient to eliminate strongly biased source signatures in Heidelberg, a medium-sized city that is strongly influenced by local sinks and sources and located about 20 km southwest of an industrial area (Mannheim/Ludwigshafen) [22]. In contrast, the UFS is a remote mountain site where there is no sudden or direct influence from the photosynthetic sink [4]. Therefore, provided that there is an increase in CO₂, the second criterion (>5 ppm) is redundant. Regarding the applied criteria in Table 1, note that we also tested shorter periods (i.e., 4 and 3 h/data point); however, the number of events detected was the same. Therefore, the initial 5 h period was preferred since a higher number of data points will supposedly yield more reliable results.

Seven events within the study period satisfied the criteria for the moving Keeling approach (see

Table 2. Date and duration of investigated episodes.

Episode	Day	Month-Year	Total Duration (h)	Duration until the Maximum of CO2 (h)
E1	09	10-2012	9	5
E2	10	10-2012	6	3
E3	27–28	10-2012	25	9
E4	28–30	11-2012	62	9
E5	15–18	01-2013	91	43
E6	25	01-2013	20	10
E7	19–21	02-2013	42	10

).

In Section 3.1, the moving Keeling approach is presented in parallel to the classical Keeling plot in order to detect the pollution source signature at the UFS. The moving Keeling plot method works so that, for example, for the determination of the mean source signature at 14:00 h, we use the hourly CO₂ and δ¹³C measurements from 12:00 h to 16:00 h and fit a regression line. To select episodes for the classical Keeling plots, periods with a CO₂ increase of more than ~15 ppm within timescales of hours and days were selected (see Table 2). In order to provide comparability with the moving Keeling plot analysis we removed the non-monotonously increasing parts of those episodes and considered only the front shoulder of the CO₂ mixing ratios. If we considered the latter selection in the example of Figure 1, the source signature would be −33.8 ± 1.7‰ instead of −30.5 ± 1.5‰, due to the bias caused by the decreasing part of the event.

2.4. Emission Ratio

In order to determine the emission ratios (slope of the regression) and accuracy (coefficient of determination R²), we examined the relationship between the CO and CO₂, as well as the CH₄ and CO₂ mixing ratios using scatter plots and fitted linear regressions, respectively. However, there are different approaches to determining the emission ratios (CO/CO₂ and CH₄/CO₂). Occasionally the emission ratios are calculated as ΔCO/ΔCO₂ or ΔCH₄/ΔCO₂, where Δ denotes the difference of the total concentration to the background concentration (which is individually defined using particular techniques (e.g., [49,50])). In some studies, the emission ratio is calculated as CO/ΔCO₂, whereas sometimes no background is removed (e.g., [51,52]). In this paper, in order to be consistent with the Keeling plot data selection, we calculated emission ratios on the total emission ratios for periods from the minimum (when the CO₂ mixing ratio started to increase) to the maximum (when the highest value of the CO₂ mixing ratio was reached). Table 2 presents the corresponding lengths of these periods for each event.

2.5. HYSPLIT Trajectory Model

We employed the HYSPLIT (hybrid single-particle Lagrangian integrated trajectory) model in order to study air transport during the episodes and characterize the CO₂ source regions [53,54]. The underlying meteorological model was obtained from the Global Forecast System and driven by 3-h forecast weather data at a 0.5° resolution. We calculated hourly backward trajectories for the 96 h before the air masses reached the UFS. The terrain altitude for the trajectory starting point at the UFS is 1312.4 m a.s.l. in the model. Thus, back trajectory calculation was started at an altitude of 1500 m above ground level (a.g.l.), resulting in 2812 m a.s.l., which roughly matches the sample inlet height (2670 m a.s.l.).

2.6. PSCF

Additionally, we utilized the potential source contribution function (PSCF) method [55,56] to plot the spatial distribution of potential geographic source locations. The PSCF calculates the probability (0 ≤ $P_{ij}$ ≤ 1) that a potential source is located at latitude $i$ and longitude $j$. Therefore, an air parcel passing through that location along the trajectory to the receptor site collects the material and characteristics from the source. The probability for a given cell is defined as $P_{ij}= m_{ij}/n_{ij}$, where $n_{ij}$ is the total number of times that the trajectories pass through the cell $\left(i,j\right)$ and $m_{ij}$ is the number of trajectories associated with concentration values at the receptor site greater than a specific threshold. In this study, the threshold was set to be the 90th percentile of the CO₂ mixing ratio and the 24 h measurement before and after each event is included in the 90th percentile calculation.

3. Results and Discussion

Figure 2 shows a high-resolution time series of the CO₂ mixing ratio, corresponding the δ¹³C values, CO mixing ratio, and CH₄ mixing ratio from October 2012 to February 2013. According to Figure 2, episodes of high CO₂ concentrations are observed during similar periods of CO and CH₄ mixing ratios. These episodes are labeled E1–E7. The mixing ratios of all three trace gases are highly correlated, while δ¹³C was anti-correlated to the CO₂ mixing ratio. In subsequent sections, we present a detailed investigation of the seven episodes exhibiting high CO₂. Table 2 contains the date, duration of each episode, and duration of their increasing part (i.e., from the start of the increase to the moment the maximum CO₂ concentration is reached).

3.1. δ¹³C(CO₂) and Keeling Plot

Figure 3a shows the resulting isotope signatures for each event using the classical Keeling plot method. It has to be noted again that only the increasing parts of each event are considered for this calculation. The intercept values range between −35‰ and −24‰, and δ¹³C_s is most enriched in E1 (−24.2‰ ± 1.2‰) and most depleted in E4 (−34.9‰ ± 0.6‰). For all the events, the δ¹³C to the 1/CO₂ relationship is well described by the WTLS regression with a high coefficient of determination (>0.90); except for one case (E5, R² = 0.82). The error bars in Figure 3a are the intercept error, which is listed in Table S1 and calculated by WTLS regression.

The averaged δ¹³C_s signatures (Figure 3a) (i.e., −29‰) are in agreement with the signatures (−20‰ to −30‰) determined by Tuzson et al. [9] at the high Alpine research station Jungfraujoch (JFJ) at an altitude of 3580 m a.s.l. This ¹³C signature represents a typical isotopic combination of anthropogenic and biological sources. The higher values of E1 and E2 (−24.2‰ ± 1.2‰ and −26.3‰ ± 0.9‰) suggest a significant contribution from coal combustion or wood burning, while the low values of E3–E7 (−34.9‰ to −29.6‰) are consistent with contributions from natural gas or gasoline [14,16].

The moving Keeling plot method with the specified data selection criteria (see Section 2.4) was employed to detect pollution events at the UFS automatically. The moving Keeling plot calculated the source signature for each of the five data points (i.e., 5 h), and depending on the length of the event, it generated several values per event; thus, the detected isotopic source signatures are presented as box plots. The whiskers of the boxplots indicate the minimum and maximum values, the lower and upper boundaries of the boxes are 25th- and 75th-percentiles, and the horizontal lines inside the boxes represent medians. The mean difference between the mean value of the classical and moving Keeling plots is 0.0‰ ± 0.5‰. This result confirms the reliability of the applied criteria.

3.2. CO/CO₂ Emission Ratios

Figure 4 shows a plot of the CO/CO₂ ratios for all individual events except for E1 and E2, for which no CO data were available. The CO/CO₂ ratios range between 3.5 and 8.0 ppb/ppm. High coefficients of determination (see Figure 4) support our hypothesis that a considerable part of the CO₂ contribution must have been due to combustion of fossil fuels (i.e., gasoline or diesel) and that CO and CO₂ were emitted from the same source. Notwithstanding, biogenic CO₂ from below and above ground respiration can be transported to the UFS along with anthropogenic CO₂, which includes a certain bias to CO and fossil fuel CO₂ relations that cannot be excluded.

While comparing our results with those obtained at other sites in Europe, a wide range of emission ratios from fossil fuel emissions was apparent, providing some idea of what to expect for the large-scale European source mix.

Table 3. CO/CO₂ emission ratios for traffic and anthropogenic emissions in Europe.

Reference	CO/CO2 (ppb/ppm)	Location	Environment	Year
Vollmer at al. (2007) [57]	9.19 ± 3.74	Switzerland	tunnel	2004
Vogel et al. (2010) [22]	13.5 ± 2.5	Germany	city	2002–2009
Tuzson et al. (2011) [9]	9.35 ± 2.66	Switzerland	remote site	2009
Popa et al. (2014) [23]	4.15 ± 0.34	Switzerland	tunnel	2011
Ammoura et al. (2014) [49]	5.68 ± 2.43	France	tunnel	2012
This study	6.02 ± 0.12	Germany	remote site	2012–2013

provides a list of some recent studies on CO/CO₂ emission ratios for local traffic emissions and fossil fuel combustion. The measured CO/CO₂ ratios at the JFJ in February 2009 by Tuzson et al. [9] are of particular interest. The authors studied four distinct pollution events within a month, and the emission ratios ranged from 7.3 to 13.1 ppb/ppm, which are generally higher than ours, except for E5 and E6. The higher values could be due to contributions from biogenic sinks, reducing CO₂ while leaving CO unaffected. However, the technological improvement of vehicles with lower CO emissions could explain the lower ratios when comparing the ratio values with studies from previous years (e.g., [23]). Popa et al. [23] measured CO and CO₂ close to each end of a highway tunnel in Islisberg, Switzerland, in the year 2011. Vollmer et al. [57] measured emissions at another highway tunnel (Gubrist) in the same region in the year 2004. The values of the year 2011 measurement are almost half of those obtained in the year 2004 [23].

3.3. CH₄/CO₂ Emission Ratios

Figure 5 shows the observed CH₄/CO₂ ratios for four events from October 2012 to March 2013 (no CH₄ data are available for E1, E2, and E6). The ratios (regression slope) vary from 4.7 to 7.4 ppb/ppm. It was found that emission ratios at the UFS are lower than at other mountain sites that reported 10 to 20 years ago (e.g., the Schauinsland (1205 m a.s.l.), Germany, with a mean ratio of 7.8 ± 1.0 ppb/ppm [52], and the Kasprowy Wierch (1987 m a.s.l.), Poland, with 10.7 ± 0.3 ppm/ppb [51]). Schmidt et al. [52] list a single diurnal cycle of CH₄ and CO₂ at the JFJ in December 1988, where the emission ratio was 5.3 ± 0.9 ppb/ppm. Worthy et al. [29] observed a decrease in anthropogenic methane emissions in Europe and Siberia from the year 1988 to 2005, which led to low ratios of CH₄/CO₂. The CH₄/CO₂ emission ratios obtained in this study compared well with the results of the highway tunnel measurements (4.6 ± 0.2 ppb/ppm) by Popa et al. [23]; indicating local traffic and transport characteristics of the western European vehicle fleet.

3.4. Backward Trajectories

In order to identify the origin and path followed by the air masses arriving at the UFS during each of the episodes, we merged the hourly output of the HYSPLIT model with the hourly data on atmospheric composition. Figure 6 shows the flow patterns and height of trajectories to the receptor site for E1 and E2. Trajectories are shown in three plots, which contain hourly trajectories of high CO₂ mixing ratios during the entire events (middle panel), 24 h before the events (left panel), and 24 h after the events (right panel). Trajectories E1 and E2 are combined due to the short time (4 h) between the two events.

CO₂ mixing ratio rapidly increased by 13 ppm in E1 from 397 to 410 ppm, and by 24 ppm in E2 from 399 to 423 ppm, within 5 h and 3 h, respectively. Also, these events were followed by rapid decreases in which the CO₂ mixing ratio dropped to 397 ppm within 4 h in E1, and to 397 ppm within 3 h in E2. Moreover, trajectory analysis for E1 and E2, which occurred only 4 h apart, shows no changes in air mass trajectories for 24 h before, during, or 24 h after the events (see Figure 6). Since almost all trajectories flow from the Atlantic Ocean and are assumed to carry clean air, this suggests that the concentration enhancements were caused by regions nearby. Regarding emissions, residential wood burning (mainly in the villages) in the Alpine area is a conventional heating system and a major contributor to CO₂ enhancement that fits well with the results of the Keeling plot intercept for E1 and E2 (−24.2‰ and −26.3‰, respectively).

In contrast to E1 and E2, the CO₂ mixing ratio increased within 20–91 h and decreased within 9–43 h during the other five events. E3–E7 exhibited short trajectories during the period of the events (CO₂ enhancement) from different directions over Europe in which air masses were transported up to the UFS (see Figure S1 in the Supplementary Materials). Trajectories from the 24 h before E6 showed westerly flow patterns in which clean air from the North Atlantic reached the site. Also, trajectories from 24 h after E3 and E7 showed that air masses reached the site via northerly flows from the Arctic Ocean. The case study in [33] is identical to E6 in this study with high formaldehyde, CO, nitrogen oxide, and nitrogen dioxide mixing ratios co-occurring during the event period. Based on the backward trajectory analysis (calculated with HYSPLIT), the influence of polluted PBL was also observed in Leuchner et al. [33], although the trajectories consider neither local wind fields nor complex topography due to the coarse grid resolution of 0.5° of the input model data.

In order to detect the area that most influenced the CO₂ mixing ratio, we applied the PSCF to the individual events (except E1 and E2). Figure 7 separately shows the spatial distribution of probable source locations of E3–E7 for 24 h before the events (left panel), during individual events (middle panel), and 24 h after the events (right panel). The coordinates of each map are from 38° to 58° for the latitude with a horizontal resolution of 0.6°. All detected potential sources are located in Germany or at the borders of Germany. This is not unexpected as, according to the European Environment Agency [58], total greenhouse gas emissions per capita in Europe are highest in Germany, where they are mostly associated with energy supply, energy use, and transportation. For more insight, a map of emissions regarding anthropogenic activities is depicted in Figure S2, which was extracted from the EDGAR Ver.432 database relevant to the year 2012. As shown above, E3 and E6 have very similar δ¹³C_s and, interestingly for these two events, the PSCF maps depict almost the same source area. Though δ¹³C_s in E4, E5, and E7 are comparable, the detected regions are not quite the same. The detected regions of E4, E5, and E7 partially overlap. However, the detected regions of E7 in the west and east of Germany are very similar to the region detected by Ghasemifard et al. [4]. They showed a PSCF map of wintertime over two and a half years (2012–2014), capturing the CO₂ mixing ratio for two specific directions—northwest and northeast of the UFS—where the highest CO₂ mixing ratios were measured and most likely linked to coal mining districts in East Germany and industrial regions in northwestern Europe, including the Netherlands and the German Ruhr area. Sturm et al. [59] simulated the isotope signatures of CO₂ at the JFJ using a backward Lagrangian model. They showed that short-term variabilities may depend on the origin of the air mass. This suggests that there is a close relationship between atmospheric composition and dominant transported air masses.

4. Conclusions

In this paper, seven events of high CO₂ concentration that occurred within five months at the Environmental Research Station Schneefernerhaus (UFS) were investigated. In order to identify and detect pollution events via continuous measurements of CO₂ and its stable carbon isotope, selected criteria were adopted for a moving Keeling plot method based on the UFS specifications. Regarding the characterization of the detected events, a combination of four types of information—namely, the δ¹³C_s signature determined by Keeling plots, CO/CO₂ correlation, CH₄/CO₂ correlation, and back trajectory analysis using the HYSPLIT mode—was employed. We were able to explain the observed short-term variation in atmospheric CO₂ mixing ratios by anthropogenic emissions. A summary of the results of the Keeling approaches and emission ratios is provided in Table S1 (Supplementary Materials). The δ¹³C_s signatures for the most enriched events (E1 and E2) indicated a significant contribution from coal combustion or wood burning, as well as for the other events that exhibited more depleted contributions from natural gas or gasoline. Based on the ratios between CO, CH₄, and CO₂, it was found that the variation in the atmospheric CO₂ mixing ratios was due to the anthropogenic emissions related to road-transport and traffic, which agrees well with δ¹³C source signature values. The CO/CO₂ emission ratio obtained in this research is less than those obtained in previous studies, which is likely due to the technological improvement of vehicles. Though the study period in the cold season minimized the bias of biological activities, it cannot be entirely excluded that biogenic CO₂ sources influenced the emission ratios. Back trajectory analysis established source-receptor relationships and revealed that the observed variations in the atmospheric CO₂ at the UFS were caused by transport of air masses from Central Europe and Germany (E3–E7), as well as by local influences for two events (E1 and E2).

Even high mountain stations can occasionally measure high CO₂ mixing ratios when air masses convey air pollution to these mountain sites. Our analysis revealed that atmospheric trace gas measurements are a vital tool to monitor and quantify air masses that are contaminated by anthropogenic pollution episodes. Measurement of the atmospheric stable isotope ratio ¹³C/¹²C and CO₂ mixing ratio at high-altitude mountain sites, as well as information about the characteristics of the measurement site, are well suited to the moving Keeling plot for the automatic detection of pollution events. Additional detailed analysis of Foehn episodes, summerly convective upwind systems, and atmospheric circulation using high-resolution meteorological modeling would provide further insights into the investigations of these kinds of events. Besides the potential to clearly explain the differences from free tropospheric conditions, this pollution analysis is an indispensable tool for source contribution, and thus for detection of pollution emissions in the areas surrounding measuring stations.