Establishing Long-term Measurements of Halocarbons at Taunus Observatory

In late 2013, a whole air flask collection program started at the Taunus Observatory (TO) in central Germany. Being a rural site in close vicinity to the Rhein-Main area, Taunus Observatory allows to assess emissions from a densely populated region. Owed to its altitude of 825 m, the site also regularly experiences background conditions, especially when air masses approach from north-westerly directions. With a large footprint area mainly covering central Europe north of the Alps, halocarbon mea5 surements at the site have the potential to improve the data base for estimation of regional and total European halogenated greenhouse gas emissions. Flask samples are collected weekly for offline analysis using a GC/MS system simultaneously employing a quadrupole as well as a time-of-flight mass spectrometer. As background reference, additional samples are collected approximately once every two weeks at the Mace Head Atmospheric Research Station (MHD) when air masses approach from the site’s clean air sector. Thus the time series at TO can be linked to the in-situ AGAGE measurements and the NOAA 10 flask sampling program at MHD. An iterative baseline identification procedure separates polluted samples from baseline data. While there is good agreement of baseline mixing ratios between TO and MHD, with a larger variability of mixing ratios at the continental site, measurements at TO are regularly influenced by elevated halocarbon mixing ratios. Here, first time series are presented for CFC-11, CFC-12, HCFC-22, HFC-134a, HFC-227ea, HFC-245fa, and dichloromethane. While atmospheric mixing ratios of the chlorofluorocarbons (CFCs) decrease, they increase for the hydrochlorofluorocarbons (HCFCs) 15 and the hydrofluorocarbons (HFCs). Small unexpected differences between CFC-11 and CFC-12 are found with regard to frequency and relative enhancement of high mixing ratio events and seasonality, although production and use of both compounds are strictly regulated by the Montreal Protocol, and therefore a similar decrease of atmospheric mixing ratios should occur. Dichloromethane, a solvent about which recently concerns have risen regarding its growing influence on stratospheric ozone depletion, does not show a significant trend with regard to both, baseline mixing ratios and the occurrence of pollution events 20 at Taunus Observatory for the time period covered, indicating stable emissions in the regions that influence the site. An analysis of trajectories from the Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model reveals differences in halocarbon mixing ranges depending on air mass origin.

Extending the setup described by Hoker et al. (2015), the system has been automated for unattended operation of up to ten individual sample canisters in one sequence. This has been achieved with pressure-operated on/off valves (Vici Valco AS-FVO2HT3) for stream selection (helium, sample, standard) and a 10-port multi-position valve (Vici Valco EMT2SD10MWE) for sample selection. All valves are heated and kept at temperatures around 80 • C.
Each air sample is analysed twice, each double measurement being bracketed by a single measurement of a whole air 5 standard which was cryogenically filled in December 2007 at Jungfraujoch, Switzerland. Mixing ratios of this working standard have been calibrated against an AGAGE gas standard. All data are reported on AGAGE scales as listed in table 1. A full measurement series also includes a blank measurement of the purified helium used as carrier gas, a vacuum blank and a measurement of a target standard.
Chromatographic peaks are integrated with a custom designed software written in the programming language IDL. The peak 10 fitting algorithm applies Gaussian fits with a constant or linear baseline. Noise calculation is performed on baseline sections close to peak retention times by determining the threefold standard deviation of the residuals between baseline data points and a second order polynomial fit. Peaks with a signal-to-noise ratio below 1.5 are rejected. The integrated detector signal is normalised to the exact enriched sample volume, determined by a pressure measurement. To account for detector drift during measurement series, the calibration measurements bracketing the sample pairs are interpolated linearly. The relative response 15 for each sample is calculated as the ratio between sample and corresponding interpolated calibration point.

Data Quality and Long-term Stability of Measurements
To ensure a high-quality dataset, automated procedures filter the data based on instrumental precision. The precision was determined for each substance and individually for the two mass spectrometers based on two sequences of 20 measurements of the Jungfraujoch working standard as described in Hoker et al. (2015). After changes were made to the enrichment unit in 2016, 20 the reproducibility experiment was repeated with no significant difference from the previous results. In a second assessment of reproducibility the instrument precision was determined using another working standard pressurized at Taunus Observatory in 2015. In this experiment, the TO working standard was analysed 13 times in a measurement sequence following the same procedure as regular air samples, and this was repeated on three different days. Instrument precision was calculated as the standard deviation of these measurements after application of the drift correction. Instrumental precisions derived from the 25 second working standard in this way agreed with values from Hoker et al. (2015), thus for consistency the latter are used in the following. Table 1 lists precisions of the substances presented here.
Precision values range for the quadrupole MS from 0.14 % (CFC-11) to 9.2 % (HFC-245fa). For the time-of-flight-MS relative precisions range from 0.20 %(CFC-11) to 9.4 % (HCFC-225cb). For most substances the quadrupole-MS yields a slightly better precision, which may partly be due to the split ratio of the gas flow. Therefore, and because TOF data do not 30 cover the time after September 2017, quadrupole data are shown if not mentioned otherwise.
Based on the instrumental precision, two types of filter routines are applied after integration of the chromatograms and calculation of drift corrected relative responses: i) Precision criterion: for the two analyses performed for each sample canister, the standard deviation of the two resulting values of the relative response are calculated and compared with the instrumental precision for each substance. If the standard 35 deviation of the double analysis exceeds three times the system precision, the sample analysis is rejected.
ii) Overlap criterion: in addition to double analysis of each sample, canisters are collected in pairs. For each pair it is checked, whether the results agree within 2σ, σ being the standard deviation of the double analysis of each canister. Table 1. System precision (prc) in % for selected substances for detection with the quadrupole (QP) and the time-of-flight mass spectrometer (TOF). The value of the system precision represents the best repeatability for the system as deduced from dedicated measurements. Precision for a particular measurement day can be different. Mixing ratios are reported as dry mixing ratios on AGAGE scales as listed in the first column. Columns labelled standard-1 and standard-2 contain long-term stability deduced from measurements of two primary standards. Only if both criteria are fulfilled, the data are included in the final time series. The mixing ratio is then calculated from the mean relative response of the detector for the sample pair. If for a pair sufficient overlap was diagnosed but only one canister meets the precision criterion, the precision rejected sample is excluded and the mixing ratio is calculated for the remaining canister only. Data which do not meet both, the precision criteria of threefold precision and the 2σ overlap criterion for double samples, are excluded from further analysis.

5
For each sample measurement day, an average daily value of the system precision is calculated from the standard deviation of the double analyses that have met the precision limit. The relative error of each final mixing ratio is reported as either the daily precision of the day when the canisters were analysed or the instrumental precision derived from the dedicated experiment, whichever is larger.
To monitor long-term stability of the GC-MS system, a primary standard is measured as target at least once per month.
10 This is usually done as part of a regular sample measurement routine, measuring the target standard relative to the working standard. Since the working standard has been calibrated versus the two primary standards, this procedure checks for relative drifts of the standards. The target standard is treated as an air sample in this procedure, and data are filtered for data precision as described above for air samples. Two different target standards were used for this, standard-1 being measured regularly from October 2013 through October 2014, standard-2 since then. Individual measurements of standard-1 were also performed 15 in 2017. Slopes of the obtained target time series agree with 0 confirming no relative drift of the primary and the working standards. Table 1 lists the corresponding standard deviations for both mass spectrometers and both standards.
While the system precisions in Table 1 reflect the repeatability of measurements on short time scales (i. e. hours), these target measurements assess long-term stability of the GC-MS system and the calibration standard. Ideally, both standard deviations should be comparable, but system precision represents a lower limit to the variability of standard measurements on the time 20 scales of years. Standard deviations of the target measurements in Table 1 are comparable with system precisions for most substances but deviate for the three HFCs and for dichloromethane. In standard-2, mixing ratios of HFC-134a, HFC-245fa and dichloromethane are markedly below current atmospheric mixing ratios and below the mixing ratios of the working standard  Open symbols denote samples flagged as outliers.
which was used for calibration and to determine the system precisions given in Table 1. Thus, the signal-to-noise ratio of peaks gets smaller which can worsen repeatability (cf. e. g. Fig. 5 of (Obersteiner et al., 2016)). This could explain the higher variability in comparison to system precision. For standard-1 the number of high precision measurements of HFC-227ea and HFC-245fa was too small for statistical analysis.
The filter procedures outlined before yield a high quality data set. The filtering is only based on precision and data consistency 5 but does not interpret measured mixing ratios. In a further step this data set is evaluated to distinguish between background measurements, i. e. baseline data, and outlier data points potentially influenced by regional emissions.
The baseline data are identified by fitting the following function: Data outside a 2σ-band around the residual mean are flagged as outliers. The remaining data are fitted again and data points 10 which fall outside 2σ of the new residual are again flagged as outliers. This is iterated until the mean of the residual does not change by more than 10 % in the subsequent iteration. If in one step the standard deviation of the residual is smaller than the mean error of mixing ratios for a specific substance, the latter is used instead. This procedure was adopted similar to the AGAGE pollution identification algorithm (cf. (O'Doherty et al., 2009) and references therein). While it is expected that outliers are mostly caused by pollution with mixing ratios above the baseline, outliers below the baseline can for example be 15 due to a stratospheric influence when the aged stratospheric air contains lower mixing ratios or due to transport from lower latitudes for substances which exhibit a latitudinal gradient.
Application of the data quality filters and the outlier filter yield a quality assessed dataset separated into baseline data and outlier events. As an example Fig. 2  The Frankfurt GC-QP-MS system was characterised and used before for studies by e. g. Laube and Engel (2008) Sampling for the NOAA network and for the dataset presented here is done sequentially. Mechanical connection of the samples and canister flushing amount to a time lag of typically 30-60 min. Although sampling is from the clean air sector, both data sets still contain some outliers with elevated mixing ratios. However, because sampling is not parallel but sequential, 15 outliers in one data set that arise from atmospheric variability are in general not apparent in the other.
Correlation coefficients r 2 above 0.9 are obtained for all substances discussed here, except for CFC-11 (r 2 = 0.81). A special case is HFC-134a, for which good agreement with NOAA data is obtained with the quadrupole instrument, but data from the time-of-flight mass spectrometer deviate for mixing ratios above 90 ppt. A similar result is obtained when correlating data from both instruments at TO, pointing to a non-linearity of the time-of-flight mass spectrometer for HFC-134a. This is apparent from

Trends and Seasonality
Air sample collection at Mace Head is restricted to times when air masses approach from the clean air sector and the data therefore represent a baseline case for the time series of halogenated compounds. In contrast, weekly air sample collection at Taunus Observatory is performed irrespective of wind direction. Therefore mixing ratios at Taunus Observatory are supposed 30 to be higher than at the coastal site except for substances which are strongly influenced by marine emissions, such as for example carbonyl sulfide or iodomethane (not shown here). Because Taunus Observatory is located closer to sources, not only higher absolute mixing ratios of most halocarbons are expected to be measured but also a higher atmospheric variability for substances with ongoing emissions.   For HFC-134a a non-linearity is observed for the TOF-MS but not for HFC-245fa. Error bars of HFC-134a mixing ratios are smaller than symbol size. an episode in September 2016 with exceptionally low mixing ratios occurring at both, TO and MHD, which is apparent for CFC-11 and CFC-12, but is more pronounced for CFC-12. A CFC-12 mixing ratio of only 509.7 ppt was measured on 16.
September 2016. Flasks from the two sites for this period were analysed on different days, making a measurement artefact unlikely.
Comparing the means of the detrended time series, TO baseline data agree with measurements at Mace Head for CFC-11 5 and CFC-12 within their respective standard deviation, although mixing ratios at TO are on average higher by 1 ppt (0.5 %) for CFC-11 but only 0.4 ppt (< 0.1 %) higher for CFC-12. Applying a linear fit function to the time series, baseline mixing ratios of CFC-11 at TO decrease at a rate of -1.2 ± 0.1 ppt/year, at MHD with a rate of -1.0 ± 0.1 ppt/year. This result agrees with the global decrease rate of -1.0 ± 0.2 ppt/year determined for the period 2015-2017 at NOAA background measurement sites (Montzka et al., 2018). Production and use of both CFCs are regulated, and their emissions should slowly approach zero. Their seasonality should 20 therefore be driven mainly by transport patterns, in particular the intrusion of aged air with lower mixing ratios from the lower stratosphere (Rosenlof, 1995;Škerlak et al., 2014). To assess the seasonality of mixing ratios, baseline time series were detrended relative to January 2013 by subtracting the linear and quadratic term of equation 1. Fig. 5 shows the resulting seasonal cycles for the two CFCs as differences to their respective annual mean for TO and MHD. Shown are monthly means and error bars indicate the error of the mean. 25 In Fig. 5, both CFCs show elevated mixing ratios in winter with regard to the annual mean and reach minimum mixing ratios in spring/summer. Commonly, such behaviour occurs for gases which are predominantly removed from the atmosphere by reaction with OH or/and have increased winter time emissions which in Europe is typically the case for combustion products.
CFC-11 and CFC-12 do not have an OH sink and their emissions are not related to combustion processes. The seasonal cycle of CFCs is driven by the seasonality of stratosphere-troposphere exchange which in the northern hemisphere maximizes in late 30 winter and spring (Škerlak et al., 2014). If remaining emission sources were regionally co-located, seasonality of large-scale transport patterns should affect CFC-11 and CFC-12 similarly. In Fig. 5, the cycle of CFC-11 is shifted forward by about two months in comparison to that of CFC-12. Another difference between the two compounds with regard to seasonality is that, as mentioned above, the fit curve of CFC-11 has a positive curvature in summer pointing to a small periodic increase in mixing ratios whereas for CFC-12 curvature is negative at all times, which means that mixing ratios continuously decrease. This holds  In this Bayesian inversion study using data from Monte Cimone and from Mace Head, European emissions were estimated on the national level with emissions occurring all over Europe but predominantly in Western Europe.
The main removing process of HCFC-22 from the atmosphere is via reaction with OH with a lifetime of 10.8 years (SPARC, 2013). Thus, a seasonality of atmospheric mixing ratios with a summer minimum and a winter maximum is expected. However, 15 for both observational sites, Taunus Observatory and Mace Head, a semi-annual cycle, adding higher harmonic terms to the fit equation yields a better fit to the seasonal cycle derived from the detrended data as shown in 6 (b) (cf. equation 2).
This has been taken into account for flagging individual samples as outliers. Of 168 valid data points, 27 are identified as outliers above the baseline which, keeping in mind the limited statistics, occur most frequently and with highest enhancements 20 during the summer months. The outlier frequency of the still used compound HCFC-22 is thus comparable to that of CFC-11 which should have been phased out globally and therefore should exhibit fewer outliers. HCFC-22 enhancements of up to approx. 20 ppt were measured (average 5.9 ppt (2.3 %)), CFC-11 enhancements even reached 22 ppt (average 3.9 ppt (1.6 %)).
11 outliers occurred simultaneously in HCFC-22 and in CFC-11, while only one outlier sample was found to be enhanced in CFC-11 and CFC-12 and only four in CFC-12 and HCFC-22.  Long-lived hydrofluorocarbons: HFC-134a, HFC-245fa, and HFC-227ea As replacement for the ozone-depleting CFCs and HCFCs, hydrofluorocarbons are now commonly used. Not containing chlorine or bromine atoms, they are not ozone depleting substances. However, long-lived HFCs are strong greenhouse gases con-15 tributing to global warming (cf. as well as the TOF mass spectrometer, the correlation of the two datasets yields a slope of 0.97 ± 0.04 ppt/ppt (axis offset of 0.14 ± 0.14 ppt) taking into account the precision of both instruments in an orthogonal data fitting routine (cf. Fig 4).

Dichloromethane
As an exemplary substance with strong seasonality, Fig. 9 Dichloromethane originates mainly from anthropogenic sources, it is used as a solvent and as a chemical feedstock, but also has a minor contribution from natural sources such as oceanic emissions and biomass burning. At current, short-lived 10 chlorinated compounds such as dichloromethane provide a small source of chlorine to the stratosphere, thus they represent a minor contribution to the stratospheric halogen load Hossaini et al., 2017;Oram et al., 2017). It was recently suggested that the importance of short-lived chlorinated compounds, among them dichloromethane, as a source to the stratosphere increases as emissions in particular from Asia could rise while other contributions such as from CFCs and HCFCs are decreasing (Hossaini et al., 2017;Oram et al., 2017).

15
Globally, after a period of decreasing surface mixing ratios dichloromethane levelled off around the year 2000 but started to increase again soon after. In 2013, a steep increase of surface mixing ratios occurred followed again by several years of little change (Simmonds et al., 2006;Hossaini et al., 2017). Upper tropospheric measurements over Southeast Asia revealed high spatial variability pointing to high regional emissions and rapid vertical transport (Oram et al., 2017). At Taunus Observatory, outliers in the time series occur mainly during spring but highest enhancements in individual samples are observed in summer 20 and autumn reaching values more than twice the respective baseline averages. On average, enhancement of outliers at TO is approx. 30 % above the baseline mixing ratio. Outliers with high mixing ratios of dichloromethane were also outliers with regard to their CFC-11 mixing ratio in 16 cases, thus in more than half of all observed CFC-11 outliers also elevated mixing ratios of dichloromethane were found.  In the time series shown in Fig. 9, mixing ratios were relatively stable at both sites over the period 2014-2016. From 2016 to 2017 an increase of approximately 3.5 ppt is registered at Mace Head, at Taunus Observatory, the baseline mixing ratios (annual means) increased by 2 ppt from 2016 to 2017 which is less than the annual variability of the baseline data (ca. 8.0-8.5 ppt). A more pronounced increase seems to occur at both sites in 2018.

5
For a first assessment of air mass origin, HYSPLIT back trajectories were calculated over 120 h for each individual sample collected at Taunus Observatory using the 1 • x 1 • GDAS meteorological dataset (Stein et al., 2015). The trajectories were attributed geometrically to one of four sectors depending on their angle of approach to the site, as illustrated in Fig. 10, not taking into account altitude. A specific trajectory is counted in one sector if more than 50 % of the 120 h period is spent in it. Trajectories crossing several sectors with no sector containing more than 50 % of the trajectory points remain undefined.  Outliers in all substances except CFC-12 and HFC-245fa most frequently occur when air masses approach TO from the south-west. Air masses approaching from south westerly directions often indicate slow moving air masses which are more likely to experience surface influence. Evaluating the most distant point of each 120h-trajectory, trajectories from the southwest travel the shortest distances. Air masses of this type are likely related to the regional influence of the nearby Rhein-Main region but might also carry emission signals from regions further south west as trajectories can reach as far as 3000 km from 5 TO within the calculation period of 5 days.
Irrespective of trajectory sector, outliers occur most frequently with trajectories which spent the 120h period closer to TO, again for all substances presented here except CFC-12 and HFC-245fa. For these two compounds no dependency of outlier frequency with trajectory extension is apparent. Outliers associated with easterly and westerly wind directions occur at comparable rates with slightly more outliers being associated with westerly winds. Trajectories from the west reach out furthest with 10 maximum distances above 5000 km, whereas easterly trajectories are slow moving and do not extend beyond 4000 km. Outlier occurrence did not clearly correlate with any other trajectory parameters such as altitude or absolute length.
Dichloromethane has, like most of the presented substances, outliers occurring most frequently when air masses approach from the south-west, but relative enhancements can also be very high in outliers with easterly trajectories. CFC-11 exhibits highest enhancements when air mass origin is in the westerly sector as does HFC-245fa. For the other substances discussed 15 here, highest enhancements are associated with trajectories from the south-west with a large spread of the measured enhancement ratios.
HCFC-22, while most of its outliers are measured when air masses approach TO from the south west, also has a significant number of outliers when trajectories originate in the clean north-west sector. These samples have very high enhancements of HCFC-22 relative to its baseline mixing ratios with the median and the 25-and 75-percentile being above those of the other 20 sectors. The north-west sector comprises two types of trajectories, namely slowly moving ones which approach predominantly at lower altitudes and fast moving ones at higher altitudes with a higher probability of both, stratospheric and maritime impact, depending on altitude. Of these, air masses associated with a slow approach at lower altitudes might bear characteristics similar to those approaching at low pace from the west sector with a higher probability of polluted air being transported from industrial regions in Western Germany and the Benelux countries. 25

Conclusions
After now more than four years of regular sample collection, we presented the first results of halocarbon measurements at Taunus Observatory for CFC-11, CFC-12, HCFC-22, HFC-134a, HFC-227ea, HFC-245fa, and for dichloromethane. Measurements are performed off-line using an automated GC/MS-system employing two mass spectrometers. Data are shown predominantly from the quadrupole mass spectrometer as it yields higher data precision and has better data coverage. However, owing 30 to the full mass scan of the time-of flight mass spectrometer operated in parallel, the number of compounds detected with this instrument is larger than for the QP instrument which is operated in SIM mode, currently detecting a pre-defined suite of 47 All data are quality filtered based on instrument precisions, and the final datasets for both sites are divided into baseline data 5 and outliers, using an iterative outlier identification algorithm. While outliers related to pollution events with mixing ratios above the baseline variability dominate the outlier statistics, occasionally also very low mixing ratios occur.
CFC-11 and CFC-12, for which production and use has been regulated longest, mixing ratios decrease overall, but more episodic high mixing ratio events are observed for CFC-11 than for CFC-12. Exceptionally high mixing ratios of CFC-11 most often correlate with enhancements of HCFC-22, HFC-134a and of dichloromethane but not with enhancements of CFC-12. In 10 addition, during summer CFC-11 mixing ratios behave different from CFC-12 mixing ratios. While the latter monotonically decrease, CFC-11 mixing ratios show a very small increase in summer following a springtime minimum.
As an example of first-generation replacement compounds HCFC-22 is shown. The substance does not show the typical seasonal cycle expected for a compound which is predominantly removed from the atmosphere via the reaction with OH, but exhibits a second maximum in summer. This is consistent with inversion-based model results predicting emissions of this 15 compound widely used for cooling applications to maximise in summer. While this is also predicted for HFC-134a, almost no seasonality of this compound is observed at TO. A possible explanation is that emissions in summer dampen the seasonality imposed by reaction with OH and in addition high variability of mixing ratios masks seasonal variation. Mixing ratios of both compounds increase. This is also the case for the two other HFCs presented here, HFC-245fa and HFC-227ea, mixing ratios of which still increase continuously at TO. However, the mixing ratio increase of the shorter-lived HFC-245fa has recently slowed 20 down, while this is not observed for the longer-lived HFC-227ea.
Based on a HYSPLIT trajectory analysis, most outliers are detected in air masses approaching TO from south-westerly direction. An exception to this represents CFC-12, for which the otherwise dominated by low mixing ratios north-west sector, normally associated with clean air containing background mixing ratios, has the highest occurrence of outliers above the baseline. Also HCFC-22 outlier occurrence in this sector is very high. Maximum mixing ratio enhancements of outliers are 25 observed when air masses arrive at the site from westerly or south-westerly directions with exception of HFC-227ea. Mixing ratio enhancements of dichloromethane can also be very high when air masses approach from the east sector.
Halocarbon mixing ratios at TO are found to be variable with polluted outliers occurring regularly. This confirms the site's sensitivity to European emissions. Measurements of halocarbons at Taunus Observatory therefore provide an extension of current surface data with the potential to further constrain regional European emissions, in particular as the site regularly expe-30 riences polluted conditions with air masses approaching over densely populated regions with industrial activity. Measurements will be continued and potentially extended, thus increasing the current database.

Data availability
Trace gas mixing ratio data are available from the corresponding author upon individual request.