The measurements of atmospheric halocarbons were performed at the Blosenberg tall tower, close to Beromünster, Switzerland, at a sample inlet height of 212 ma.g.l. (above ground level). The tower is located in a rural area in the middle of the Swiss Plateau at 47.2∘ N, 8.2∘ E, at an altitude of 797 ma.s.l. The site was originally equipped with GHG and meteorological measurements at five different heights up to 212 ma.g.l. as part of the CarboCount-CH project (Oney et al., 2015; Berhanu et al., 2016). Later, it was integrated into the Swiss National Air Pollution Monitoring Network (NABEL), the European Monitoring and Evaluation Programme (EMEP), and the Global Atmosphere Watch (GAW) regional station network of the World Meteorological Organization (WMO). The area referred to as the Swiss Plateau is a landscape north (N) of the Swiss Alps, south (S) of the Jura Mountains, and confined by Lake Geneva in the southwest (SW) and Lake Constance in the northeast (NE). Covering one-third of the Swiss surface area, the Swiss Plateau represents the most densely populated (more than two-thirds of the Swiss population) and the industrially most active region of Switzerland. Medium- and large-sized industrial enterprises are located in this area. Cities with the highest population on the Swiss Plateau are Zurich (population of 415 000), about 40 km NE, Lucerne (population of 82 000) about 20 km south-southeast (SSE), and Bern (population of 144 000), about 70 km SW of the tower.

Wind observations collected at 212 ma.g.l. during the measurement campaign are summarized in Fig. 1. During the campaign, the major wind direction at Beromünster was from west-southwest (WSW), followed by wind from the direction of NE, with maximum wind velocities of 24 ms-1 at the top of the tower (212 m). This bimodal wind direction distribution is typical for the Swiss Plateau, with the near-surface flow being channelled by the mountain ridges to the south (Alps) and north (Jura).

Beromünster was determined to be sensitive to the major part of the Swiss Plateau (Fig. 2 and Oney et al., 2015) and is, thus, well suited to capturing halocarbon emissions from this area on a small to medium regional scale. Additional halocarbon measurements used for the inverse modelling in this study (Sect. 2.3 and 2.5) came from the AGAGE measurement stations at Jungfraujoch (46.5∘ N, 8.0∘ E; 3580 ma.s.l.; inlet height -15 ma.g.l., i.e. inlet height is below instrument height), Mace Head (53.3∘ N, -9.9∘ E; 8 ma.s.l.; inlet height 10 ma.g.l.), and Tacolneston (52.5∘ N, 1.1∘ E; 56 ma.s.l.; inlet height 185 ma.g.l.; Prinn et al., 2018).

In situ halocarbon measurements were conducted with a Medusa pre-concentration unit, coupled to GC and MS (Miller et al., 2008; Arnold et al., 2012; Prinn et al., 2018). Ambient air was sampled at 212 ma.g.l. through a continuously flushed tube, consisting of an innermost ethylene copolymer coating on a slit overlapping aluminium layer and an outer high-density polyethylene jacket (SERTOflex; 12 mm outer diameter (OD); 8 mm inner diameter (ID)). From this sample stream, 2 L of air were diverted to the Medusa every 70 min at a flow rate of 100 mLmin-1. To prevent condensed water from entering the sampling module, the main air stream was drawn through a custom-built in-line water trap consisting of a stainless-steel dip tube, which was regularly checked for liquid water. The integrity of the sampling line was confirmed by comparing the in situ measurements with those of simultaneously drawn flask samples from a second sampling line at the same inlet height. Additionally, all joints which connected the sampling line to the measurement devices were sprayed with high-concentration gaseous tracers and found to be leak free.

In the AGAGE measurement set-up, the air sample is pre-concentrated at low temperatures in a two-trap system (Medusa). Our Medusa used a Stirling cooler (AMETEK, Inc./Sunpower Inc., CryoTel GT), leading to sample trapping temperatures of -165 ∘C for the first trap and -180 ∘C for the second trap. After pre-concentration, the analytes were flushed into a GC (Agilent 6890N), using helium as the carrier gas (He 6.0; Messer Schweiz AG), and detected by quadrupole electron ionization MS (Agilent 5975) in selected ion mode (SIM). Chromatographic separation was achieved with a CP-PoraBOND Q column (0.32 mm internal diameter × 25 m; 5 µm film thickness; Agilent). GCWerks was used as instrument control and data processing software.

To correct for short-term instrumental drift, the measurements of two consecutive ambient air samples were bracketed by the measurements of a working standard. The working standard consisted of ambient air compressed into 34 L internally electropolished and humidified stainless-steel tanks (Essex Industries, Inc., USA), using an oil-free air compressor (SA-6; RIX Industries, USA). These real-air working standards are collected at Rigi–Seebodenalp, Switzerland (47.1∘ N, 8.5∘ E; 1030 ma.s.l.), during relatively clean-air conditions. To better track the MS sensitivities for low-abundance substances like HFO-1234yf, HFO-1234ze(E), and HCFO-1233zd(E), we spiked these substances into the standards. For HFO-1234yf, HFO-1234ze(E), and HCFO-1233zd(E), the respective final working standard concentrations were, on average, 1.5, 5.0, and 0.6 ppt (parts per trillion; pmol mol-1). The working standards used at Beromünster were cross-calibrated within the AGAGE relative calibration scale by a predefined inter-calibration routine, which consists of a hierarchy of calibration standards provided by the Scripps Institution of Oceanography (SIO; Miller et al., 2008). In addition, some substances were calibrated on primary scales produced by the Swiss Federal Laboratories for Materials Science and Technologies (Empa), the Federal Institute of Metrology in Switzerland (METAS), or the University of Bristol (UB).

In total, three types of uncertainties were considered, i.e. the accuracy of the primary calibration scale originating from the production of the primary standard, the uncertainty from the propagation of the calibration standards within the AGAGE calibration hierarchy (Prinn et al., 2018), and the measurement precisions. The uncertainties are listed in Sect. S2. If the accuracy of the primary calibration scale was unknown for a substance, it was set to 2 %. If the uncertainty of the standard propagation was unknown, the measurement precision was propagated instead. Measurement precisions were derived from the difference of the pair of working standard measurements that were bracketing the actual air measurements. They were below 3 % for 26 of the 28 reported halocarbons and below 5 % for HFC-4310mee and HFC-236fa.

The analytical set-up was operated in a temperature-controlled (± 2 ∘C) trailer adjacent to the tower. R410a (a mixture of HFC-32 and HFC-125; 50 % by weight each) was used as refrigerant for the air conditioner. Weekly trailer indoor air measurements suggested the absence of any major refrigerant leaks.

For the use as tracer in the tracer ratio method, CO was measured as part of the NABEL network with a Picarro G5310 analyser (Picarro Inc., USA) using cavity ring-down spectroscopy (CRDS). Daily instrument calibration was performed with three working standards of different concentrations. These working standards were produced at Empa and calibrated against reference standards of the GAW Programme.

Receptor-oriented backward simulations with the Lagrangian particle dispersion model (LPDM) FLEXPART (Stohl et al., 2005; Pisso et al., 2019) were carried out to estimate the sensitivity of the observed atmospheric concentrations to regional emissions (source sensitivities; Seibert et al., 2004). Depending on the location of the sites, different versions of the model, driven by different meteorological input fields, were utilized. For the Swiss sites in complex terrain (Jungfraujoch and Beromünster), the FLEXPART version adjusted for input from the regional numerical weather prediction (NWP) model COSMO was used with meteorological analysis fields operationally provided hourly by the Swiss national weather service (MeteoSwiss) at a horizontal resolution of approximately 7 km by 7 km (COSMO-7). For the two additional sites located on the British Isles (Mace Head and Tacolneston), which is towards the northwestern domain boundary of the COSMO-7 domain, the main version of FLEXPART (version 9.2_Empa) for use with input from the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS) was applied. The employed input fields had a horizontal resolution of 0.2∘ by 0.2∘ in the Alpine area (4∘ W to 16∘ E and 39∘ N to 51∘ N) and 1∘ by 1∘ elsewhere. For both model versions, a similar backward simulation strategy was followed, releasing 50 000 model particles during 3 h intervals at each measurement location and tracing these particles backward in time for 8 and 10 d for FLEXPART–COSMO and FLEXPART–IFS, respectively.

The resulting source sensitivities, mi,j, connect the spatial distribution of the emissions (Ei,j) with the concentration of a tracer in the receptor (χ) via a linear relationship, as follows: 1χ=∑i,jmi,jEi,j+χb, where χb is the average concentration of the particles at the endpoints of the backward simulation and is comparable to boundary conditions in Eulerian model simulations. Here, χb was not explicitly simulated or taken from a larger-scale model but is replaced by an observation-based baseline concentration (Sect. 2.5).

Besides total receptor concentrations, spatially resolved FLEXPART source sensitivities were used to identify situations in which air masses sampled at Beromünster were dominated by surface contact over the Swiss domain. For this purpose, the sum of mi,j for different land regions was calculated for each model interval, and the fraction of Swiss residence time to total residence time over land was determined.

The TRM relates emissions of a target analyte to those of a tracer with known emissions during pollution events. The underlying assumption of the TRM is that both the target analyte and the tracer have similar spatial and temporal emission sources and are transported to the receptor, i.e. the measurement site, without any significant additional production or loss (Yao et al., 2012). This also implies that the analyte and the tracer behave similarly in the atmosphere or that the transport distance to the measurement site is either short enough for the analyte and tracer ratio to be preserved or long enough so that analyte and tracer emissions from multiple sources are well mixed. In this case, a sufficiently large catchment area is needed for substances with distinct emission areas to result in improved mixing with the tracer. Overall, we consider Beromünster as an adequate site for the TRM because the Swiss pollution sources are at a significant distance (i.e. Lucerne is the nearest large town 20 km away), and thus well mixed, and because the observed substances are very stable in the atmosphere and thus not modified during their transport.

In this study, CO was used as the reference tracer. It was assumed that, in Switzerland, CO is only emitted anthropogenically during the combustion of biofuels and fossil fuels in road transport and non-industrial combustion processes, leading to relatively constant CO emissions for all seasons (Guevara et al., 2021). Other sources of CO, i.e. emissions from wildfires or oxidation of methane and non-methane volatile organic carbons (VOCs), were assumed negligible or constant for the Swiss domain (Oney et al., 2017). The main sink of CO is oxidation with hydroxyl radicals, resulting in an atmospheric lifetime of 22 d in summer in the Northern Hemisphere (Miller et al., 2012), which is much longer than the transport time of CO from Swiss emission sources to Beromünster. Hence, it was possible to neglect CO degradation in the current approach.

The emissions (Ex) of a target analyte in units of megagrams per year (hereafter Mgyr-1) were calculated by applying Eq. () as follows: 2Ex=ECOΔXΔCOMxMCO.

ΔX (in units of ppt) and ΔCO (in units of ppb; parts per billion; nmol mol-1) are the respective enhancements of atmospheric concentrations above background of the target analyte and CO, respectively. To determine enhancement above background, the CO data were first averaged within the sampling time interval of each halocarbon measurement. Pollution events were then distinguished from background concentrations by applying the statistical robust extraction of baseline signal (REBS) method (Ruckstuhl et al., 2012). Since this method yields the atmospheric background level as a baseline, the two expressions are used synonymously in the following. To determine a method uncertainty, different baseline variations were calculated and incorporated into the emission estimation. REBS settings included the use of a tuning factor (b = 3.5) and three different settings of the temporal window width, i.e. a bandwidth parameter of 15, 30, and 60 d. Next to a smooth baseline, the REBS method provides a global estimate of the uncertainty of the baseline. To add to the determination of the method uncertainty, different subsets of data above the baseline were created by scaling the REBS uncertainty to different levels by multiplying with specific factors, i.e. 1, 1.5, and 2. The fractions of measured baseline concentrations relative to the total number of data points and the average background concentrations for the considered halocarbons for a specific example of REBS settings are given in Sect. S4. Multiplication with the ratio of molecular weights (Mx and MCO; both in units of gmol-1) is needed to convert from mass to volume mixing ratios. ECO is the a priori emission inventory value of CO. To cover the time frame of the measurement campaign, the CO inventory values for the years 2019 and 2020 were weighted accordingly, to result in a Swiss CO emission value of 152.9 Ggyr-1. Correspondent to Reimann et al. (2021), the yet unreported Swiss CO inventory value for 2020 was calculated from the latest available CO inventory values reported to Convention on Long-range Transboundary Air Pollution (CLRTAP)/EMEP and the average trend over the preceding 3 years.

For the emission calculation, the data were filtered based on the simulated near-surface residence time of the sampled air masses (Sect. 2.3). Near-surface residence times were evaluated by country/region. Residence times over Switzerland were divided by the total residence time over land areas in the model domain. In total, two different minimal ratios of Swiss residence time were used for observation filtering, i.e. for events with 50 % and 75 % relative country residence times, respectively. A map, depicting the distribution of the relative country residence times of the air masses measured at Beromünster for different regions, is given in Sect. S4. The air measured at Beromünster resided over Switzerland, France, and Germany most of the time.

To calculate the halocarbon–CO emission ratio, all remaining pollution events above baseline were summed up to give a term for ΔX and ΔCO, respectively, and the two terms were then divided. The values arising from the different baseline and country residence time settings were averaged to give a final emission value.

Several sources of error were taken into account throughout the tracer ratio calculations. For the halocarbon and CO measurements, the corresponding measurement precisions (Sect. 2.2) at 1σ (68 %) confidence level and the uncertainty of the modelled baseline fit were propagated by standard Gaussian error propagation. Then the two types of calibration accuracies (Sect. 2.2) for the halocarbon measurements were added to the uncertainty of the term ΔX before calculation of the halocarbon–CO emission ratio. Final uncertainties for emission estimations were calculated at the 2σ (95 %) confidence level. As described before, method uncertainties arising from the choice of parameters for baseline fitting and relative country residence time were incorporated by taking into account the different settings for the baseline fitting parameters and the two different levels of relative country residence times.

For CFC-13, CFC-115, H-2402, HCFC-22, HCFC-124, HFC-23 (CHF3), HFC-236fa, PFC-318, PFC-14, and NF3, only a very limited number of pollution events with significant magnitude were observed. Thus, depending on the set REBS and country residence time parameters, the number of data points included in the emission estimation with the TRM was greatly reduced (Sect. S4). Emissions calculated with fewer than 10 data points were deemed unreliable.

The second top-down approach for estimating Swiss emissions utilizes the FLEXPART simulated source sensitivities coupled with a Bayesian inversion (BI) framework. The latter is used to obtain an optimized state of the emissions combining a priori knowledge of the emissions, model-simulated source sensitivities, and observations at the receptor sites. The method applied in this study was extensively described by Henne et al. (2016) for methane and frequently applied to halocarbon emissions (e.g. Park et al., 2021; Simmonds et al., 2020; Rigby et al., 2019; Schönenberger, 2018).

The inversions in this study cover the period of the field campaign in Beromünster. Daily mean values of the observations at Beromünster, Jungfraujoch, Tacolneston, and Mace Head were utilized. Daily mean observations were preferred over the use of short aggregation intervals (e.g. 3 h) because few changes in total and spatially resolved emissions were seen when using the latter. The use of the longer aggregates reduces the inverse problem size and, hence, allows for a faster and, in our experience, more robust estimation of covariance parameters. Moreover, sensitivity inversions for HFO-1234yf were performed in order to quantify the sensitivity of the a posteriori emissions in Switzerland to the selection of measurement sites. When adding additional observations from the Taunus Observatory in central Germany or when removing the observations from the British Isles, the changes in total Swiss emissions were smaller than 5 %.

Modelled source sensitivities for the four receptor sites were used together with an a priori estimate for the emissions xb and the observations χo in a BI framework in order to obtain an optimized state for the emissions. The inversion domain extends from 12.0∘ W to 21.1∘ E and 36.0 to 57.5∘ N. The a posteriori emissions were calculated by minimizing the following cost function: 3J=12(x-xb)TB-1(x-xb)+12(Mx-χo)TR-1(Mx-χo), where x denotes the a posteriori state vector comprised of gridded emissions and baseline concentrations, xb gives the a priori state, M corresponds to the source sensitivities, χo to the observations, and B and R are the uncertainty covariance matrices of the a priori emissions and the combined model–observation uncertainty, respectively. Both B and R are symmetric block matrices. Matrix B contains two blocks, BE and BB, which describe the uncertainty covariance of the gridded emissions and the baseline, respectively. The diagonal elements of matrix BE are proportional to the a priori emissions, whereas the off-diagonal elements are assumed to be spatially correlated with an exponential decay with the distance, as follows: 4Bi,iE=(fExb,i)25Bi,jE=e-di,jLBi,iEBj,jE,i≠j.

Factor fE is optimized during a maximum likelihood step (Henne et al., 2016). Factor di,j is the distance between grid cells, and L is the correlation length optimized again in the maximum likelihood step. The diagonal elements of block BB are set to a fixed value proportional to an estimate of the baseline uncertainty, and the non-diagonal elements are assumed to be correlated with an exponentially decreasing correlation of the baseline uncertainty between baseline nodes at a given site, as follows: 6Bi,iB=fBσb27Bi,jB=e-Ti,jτbBi,iBBj,jB,i≠j, where Ti,j is the time difference between the nodes and τb is the temporal correlation length. The factors fB and τb are optimized during the maximum likelihood step.

Block matrix R contains as many block matrices as the number of the receptor sites (in our case four), and each block matrix contains both model and observations uncertainty covariances. Diagonal elements of R contain both observations and model uncertainties, while off-diagonal elements are again assumed to be correlated with an exponentially decreasing structure with time, as follows: 8Ri,i=σ02+σmin⁡2+σsrr2χp,i29Ri,j=e-Ti,jτ0Ri,iRj,j,i≠j, where σ0 is the uncertainty of the observations, while σmin⁡ and σsrr are uncertainties related to the transport model optimized again during the maximum likelihood step. Factor Ti,j is the time difference between measurements, and τ0 is the temporal correlation length set to 0.01 d, reflecting very low autocorrelation when using daily average observations. No covariance between different observing sites was considered.

The minimization of the cost function Eq. () reduces the difference between observed and simulated values (χo,Mx), which are, additionally, constrained by the difference between the a priori and a posteriori emission estimates. A posteriori emissions are given by the following: 10x=xb+BMT(MBMT+R)-1(χo-Mxb).

For this study, the a priori baseline of each halocarbon was determined by the REBS method (Sect. 2.4) and was optimized according to the corresponding site. REBS settings included the use of the tuning factor b equal to 3.5, a temporal window width of 30 d, and a maximum of 10 iterations. For Beromünster, the baselines calculated for Jungfraujoch were applied because synoptic-scale baselines are needed to follow particles to their endpoints of the simulation outside the inversion domain. The a priori emissions used for the HFCs (except HFC-152a), PFCs, SF6, and NF3 were taken from the 2018 national reports to the UNFCCC (UNFCCC, 2021) for the reporting countries in the model domain. Since, for HFC-134a, a major difference between the 2018 and 2019 inventories is reported, an additional test inversion was run using the 2019 inventory values. However, this had a negligible effect on the result, and a priori values for Switzerland were applied, as listed in Table 1. For HFC-152a, the Swiss inventory value was not used as a priori since, for this substance, emissions are attributed to the manufacturing and not the emitting countries. For HFC-152a, CFCs (except CFC-13), halons, HCFCs, and HFOs, the 2019 Swiss emissions estimated based on the Jungfraujoch data (Reimann et al., 2021), and as annually submitted to the Swiss Office for the Environment (FOEN), were used as a priori. For CFC-13, the 2020 value was used as a priori, as the 2019 value indicated depletion. The a priori emissions of these substances for the remaining countries in the domain were distributed according to the Swiss emissions on a per capita basis. A priori uncertainties were obtained through the above-mentioned maximum likelihood approach. Because the Jungfraujoch-based estimates rely on the TRM that is based on a very limited number of data points (Reimann et al., 2021), sensitivity inversions were implemented to assess the variability in the emissions if the confidence interval of the Jungfraujoch-based a priori emission values was large. For the sensitivity testing, half and double of the a priori estimates for all countries in the inversion domain were used. The sensitivity simulations showed only little response to the used a priori value; therefore, only the results with the mean a priori simulations are reported here. Except for HFO-1234yf, the HFOs were treated as inert for the inversions, assuming that the transport times from emission sources to BRM are sufficiently small to avoid larger chemical losses. Monthly average atmospheric lifetimes of HFO-1234yf, based on Henne et al. (2012), were used to update the source sensitivities specifically for this compound. Subsequently, these updated source sensitivities were used in the inversion. Resulting Swiss emissions were about 10 % higher than when assuming inert HFO-1234yf. The other HFOs treated here have longer atmospheric lifetimes (Sect. S1), and hence, their lifetime impact on Swiss emissions is smaller and was deemed negligible in the light of other uncertainties.

For all the halocarbons in this study, population-based a priori emissions were utilized. Flat a priori distributions (homogeneous within each country and zero over the ocean) were tested for the subset of HFC-23, SF6, and PFC-14. However, the simulation results were inferior in terms of performance, most likely due to the specific topographic situation in Switzerland, where population distribution and industrial activity are closely linked and separated from the high-altitude regions of the Alps. Thus, the population-based distribution was pursued. The inversion output provided gridded emissions for the major part of the European domain. To calculate the Swiss emissions, the emissions from the grids lying inside the Swiss borders were extracted.

The inversions performance and reliability of results for the different substances was assessed through several statistical measures. In Sect. S5 the most important statistical measures and the inversions reliability are summarized. The χ2 index (defined as χ2=J(x)2/d, with d being the number of observations) assesses the probability density distribution of the a posteriori model residuals and a posteriori emission differences, which should follow a χ2 distribution with the mean equal to d/2 (e.g. Berchet et al., 2013). Ideally, a χ2 index with a value close to one can be achieved with well-chosen covariance matrices. The degree of freedom (DOF) describes the reduction in the normalized error of the a priori values due to the number of available observations and, thus, is a measure of improvement of the a posteriori when compared to the a priori. Additional comparison statistics for the a posteriori simulated versus observed regional (above baseline) concentrations at Beromünster were the correlation coefficient, r2, and the normalized standard deviation, nSD, which is the standard deviation of the simulations divided by the standard deviation of the observations. For both statistics, values close to one indicate favourable model performance. Based on the above-mentioned parameters, the reliability of the inversion calculations for the different substances was evaluated. Inversion results with r2 greater than 0.1 were deemed reliable, which was the case for most of the substances. Exceptions were CFC-13, CFC-115, H-2402, HCFC-124, HFC-23, HFC-236fa, PFC-318, and PFC-14, for which reported results should not be over-interpreted.

The 1-year Beromünster records are shown in Fig. 3 and Sect. S3 for the halocarbons with the highest atmospheric abundance and/or for which emissions were calculated. This includes four CFCs, two halons, four HCFCs, 10 HFCs, three PFCs, SF6, NF3, three HFOs, and CO. Concentrations are reported as dry air mole fractions in units of ppt for the halocarbons and in units of ppb for CO.

The majority of the records show pollution events during the whole measurement campaign. For HCFC-142b, HFC-134a, HFC-125, HFC-32, HFC-152a, HFO-1234yf, HFO-1234ze(E), and HCFO-1233zd(E), the pollution events quickly succeed each other within hourly or daily time intervals. Therefore, for these substances, a background concentration level is scarcely reached during the whole year or only seasonally during the winter months. This can be explained by the extensive use of these halocarbons in various applications and their continuous emissions from regional sources captured by the site. In the following, the individual substances are discussed in more detail.

For CFC-11, CFC-12, and H-1211, we were able to define solid baselines with some overlying structure arising from ongoing emissions of existing installations (banks), underlining the extensive historical use. For CFC-13, CFC-115, and H-2402, pollution events were virtually absent, indicating minor historic use.

Furthermore, HCFC-22, HCFC-141b, and the minor HCFC-124 showed few distinctive pollution events, whereas for HCFC-142b, pollution events were more frequent with enhancements above baseline reaching a maximum larger than 6 ppt. Our observations underline the finding that these HCFCs are still emitted from outgassing of existing foams and refrigeration units.

For the HFCs, our measurements show the highest abundances and most frequent pollution events for HFC-134a, HFC-125, and HFC-32, with average background concentrations of around 120, 40, and 30 ppt, respectively. HFC-134a reached the highest pollution events of more than 80 ppt over baseline, whereas pollution events of HFC-125 and HFC-32 were smaller with maximum values of approximately 20 and 30 ppt over baseline. For HFC-152a, we observed an average baseline concentration of about 10 ppt with only minor enhancements. This can be explained by small HFC-152a emissions in Switzerland arising only from outgassing of existing foams, as there is no production of foams, and the use as a propellant is forbidden. HFC-245fa, HFC-365mfc, HFC-227ea, HFC-236fa, and HFC-4310mee all showed clear baselines with a few distinctive pollution events, which, however, only lasted for a few hours. For these five HFCs, the majority of the emissions were localized outside of Switzerland – only 19 % of the pollution events had a Swiss residence time above 50 % and 5 % were above 75 % residence time (REBS bandwidth of 30 d and multiplication factor of 1.5). For HFC-227ea, we observed a few distinctive pollution events with a Swiss residence time larger than 50 %. Although the specified events were exceeding the baseline by about 1 ppt only, they may point to non-continuous, recurrent emissions of HFC-227ea. As one of the more abundant HFCs in the atmosphere, HFC-23 showed an average baseline level of approximately 34 ppt. However, there were only very few pollution events. This can be ascribed to the absence of HCFC-22 synthesis in Switzerland, from which HFC-23, besides minor emissions from specific other applications (Oram et al., 1998; Miller et al., 2010; Montzka et al., 2010; Stanley et al., 2020), is mainly emitted as an unwanted byproduct (Keller et al., 2011; WMO, 2018).

For SF6, sporadic pollution episodes were observed, with only 17 % of the pollution events greater than 1 ppt over baseline, however, showing a high contribution from Switzerland. For NF3 and PFC-14, the time series show little variability beyond measurement precision. PFC-14 is emitted during the production of aluminium, while NF3 is produced during manufacturing processes in the electronic industry, both absent from within the footprint of the station. The other investigated PFCs are used in the semiconductor industry. PFC-116 shows sporadic pollution events, especially during the summer months, and PFC-318 only shows two discernible events in March and July 2020. For both substances, only about 20 % of the pollution events are attributed to Switzerland with more than 50 % country residence time.

For the newly produced HFO-1234yf, HFO-1234ze(E), and HCFO-1233zd(E), the average baseline levels were as low as 0.4, 0.5, and 0.2 ppt, respectively. More than 50 % of the measurements were assigned to pollution events. The highest pollution events were seen for HFO-1234ze(E), exceeding 40 ppt over baseline, followed by lower events of about 4 ppt over baseline for HFO-1234yf and 2 ppt over baseline for HCFO-1233zd(E). Our observations confirm the increasing use of HFO-1234ze(E) and HFO-1234yf as refrigerants in Switzerland and the fact that HCFO-1233zd(E) is only allowed for limited application as cooling agent and as foam blowing agent.

The time series of CO (used as the tracer in the TRM) shows a pronounced variability in the baseline, with a maximum at the end of March and a minimum in July. Many events of high concentration occur during the winter months, followed by fewer observed events during spring and summer; however, this is not implying smaller emissions in summer but rather reflecting the faster mixing of surface emissions throughout a larger fraction of the troposphere (Sect. 2.4). The seasonality in the baseline and in the amplitude and frequency of pollution events is typical for the Northern Hemisphere. The seasonal variation in CO at Beromünster was already studied in detail by Satar et al. (2016), who found smaller seasonal amplitudes compared to other European tall tower stations.

Based on the atmospheric measurements at Beromünster, Jungfraujoch, Mace Head, and Tacolneston, Swiss emissions were determined with the TRM (Sect. 2.4) and BI models (Sect. 2.5). The results are listed in Table 2 and are compared to the Swiss top-down emission estimates based on the Jungfraujoch data (Reimann et al., 2021), and, for HFCs, PFCs, SF6, and NF3, to the latest (2019) bottom-up inventory reported to the UNFCCC (UNFCCC, 2021). Because the Jungfraujoch-based Swiss emissions are calculated as 3-year averages, the 2019 value is used for comparison instead of the 2020 value. The time series of the Swiss emission values since 2015 are presented in Fig. 4, and supplementary numbers are listed in Sect. S6. For the Jungfraujoch and the inventory emissions, error bars are only depicted for the 2019 values. For the inventory values, uncertainties were assessed based on the information and uncertainties (at 1σ (68 %) confidence level) given in the NIR (FOEN, 2021a) for the individual emission source categories and on the detailed emission numbers submitted to the UNFCCC. Figure 5 summarizes the 2019/2020 emission values, grouped for different halocarbon classes, and shows the corresponding values converted to CO2 equivalents.

In the following, the estimated emissions are discussed in more detail. Emissions including Beromünster measurements are given as the averaged values of the TRM and BI methods. In the text, emissions are given with two significant digits.

A significant asset of BI is its ability to geographically locate a posteriori emission distributions. The corresponding maps for selected halocarbons are shown in Fig. 6 (Switzerland) and in Sect. S7 (European domain). The subset of substances includes CFC-11 and HCFC-142b, as representatives for the CFCs and HCFCs, the three most highly emitted HFCs, which are used as refrigerants, the foam blowing agent HFC-365mfc, SF6, and three HFOs.

All a posteriori distributions continue to reflect the population-based distribution, which was used in the a priori emissions. However, there are differences between the compounds, visible as differing pronounced hotspots. CFC-11, HCFC-142b, and HFO-1234ze(E) emissions were especially pronounced for the area around Zurich, which has the highest population density. HFC-134a, HFC-125, HFC-32, and HFO-1234yf, all used as cooling agents, were emitted more diffusely across the Swiss Plateau, showing, in addition to the more populated areas, hotspots in other regions (e.g. between Lausanne and Bern). The foam blowing agent HFC-365mfc showed somewhat fewer pronounced emissions from the populated areas, with more emissions from outside Switzerland dominating the distribution at a wider scale. SF6 showed generally diffuse emissions from the Swiss Plateau; however, there were some more pronounced grids at Basel and Bern. HCFO-1233zd(E) emissions were especially emphasized in the areas of Basel, Bern, and Lausanne.

In the European frame, the Swiss emissions of CFC-11, HCFC-142b, and HFO-1234ze(E) were of comparable magnitude to the emissions from the border areas of France and Germany, except for the significantly enhanced area around Zurich. For HFC-134a, HFC-125, HFC-32, SF6, HFO-1234yf, and HCFO-1233zd(E), the Swiss emissions were of comparable magnitude to the emissions north of the border in Germany and France, with a diverse distribution of locations with increased emissions. For HFC-365mfc, compared to Switzerland, enhanced emissions were attributed to France and Germany.

The maps of the absolute differences between the a priori and the a posteriori emissions (Sect. S7) show somewhat the same population-driven distribution as the relative source maps. However, for all substances, except CFC-11 and HFO-1234ze(E), the model decreased the a posteriori emissions at the region around Zurich as compared to the a priori distribution. For HFO-1234ze(E), a posteriori emissions were especially increased around Zurich. For the HFCs and SF6, the model increased the a posteriori emissions from the southwestern region towards Lausanne.