A detailed description of the Beromünster tall tower measurement system as well as a characterization of the site with respect to local meteorological conditions, seasonal and diurnal variations of greenhouse gases, and regional representativeness can be obtained from previous publications (Oney et al., 2015; Berhanu et al., 2016; Satar et al., 2016). In brief, the tower is located near the southern border of the Swiss Plateau, the comparatively flat part of Switzerland between the Alps in the south and the Jura Mountains in the northwest (47∘11′23′′ N, 8∘10′32′′ E, 797 m a.s.l.), which is characterized by intense agriculture and rather high population density (Fig. 1). The tower is 217.5 m tall with access to five sampling heights (12.5, 44.6, 71.5, 131.6, 212.5 m) for measuring CO, CO2, CH4, and H2O using cavity ring-down spectroscopy (CRDS) (Picarro Inc., G-2401). By sequentially switching from the highest to the lowest level, mixing ratios of these trace gases were recorded continuously for 3 min per height, but only the last 60 s was retained for data analysis. The calibration procedure for ambient air includes measurements of reference gases with high and low mixing ratios traceable to international standards (WMO-X2007 for CO2 and WMO-X2004 for CO and CH4), as well as target gas and more frequent working gas determinations to ensure the quality of the measurement system. From 2 years of data a long-term reproducibility of 2.79 ppb, 0.05 ppm, and 0.29 ppb for CO, CO2, and CH4, respectively, was determined for this system (Berhanu et al., 2016).

Air samples for 14CO2 analysis were collected from the highest inlet usually between 09:00 to 13:00 UTC. At the beginning we collected one sample per month, which was eventually changed to sampling every second week from November 2013 onwards. During each sampling event, three samples were collected over a 15 min interval in 100 L PE-AL-PE bags (TESSERAUX, Germany) from the flush pump exhaust line of the 212.5 m sampling inlet, which has a flow rate of about 9 L min-1 at ambient conditions. The sampling interval was chosen to ensure radiocarbon sample collection in parallel with the continuous CO and CO2 measurements by the CRDS analyzer at the highest level. Each bag was filled at ambient air pressure for 6 to 8 min, and a total air volume of 50 to 70 L (at STP) was collected.

CO2 extraction was conducted cryogenically in the laboratory at the University of Bern usually the day after the sample collection. During the extraction step, the air sample was first pumped through a stainless-steel water trap (-75 ∘C), which was filled with glass beads (Raschig rings, 5 mm, Germany). A flow controller (Analyt-MTC, Aalborg, USA) with a flow totalizer tool was attached to this trap to maintain a constant flow of air (1.2 L min-1) towards the second trap (trap 2), a spiral-shaped stainless-steel tube (1/4 in.) filled with glass beads (∼ 2 mm) and immersed in liquid nitrogen to freeze out CO2. When the flow ceased, trap 2 was isolated from the line and evacuated to remove gases which are non-condensable at this temperature. Then, trap 2 was warmed to room temperature and eventually immersed in slush at -75 ∘C to freeze out any remaining water. Finally, the extracted CO2 was expanded and collected in a 50 mL glass flask immersed in liquid nitrogen.

Sample extraction efficiency was calculated by comparing the amount of the cryogenically extracted CO2 with the CO2 measured in situ by the CRDS analyzer during the time of sampling. The amount of CO2 extracted is determined first by transferring the extracted CO2 cryogenically to a vacuum line of predetermined volume. Then, based on the pressure reading of the expanded gas, and the total volume of air collected determined by the mass flow controller with a totalizer function attached to trap 1, CO2 mixing ratios were calculated.

At the end of 2014 we noticed that there was a leakage from the sampling line exhaust pumps, which resulted in unrealistically high CO2 mixing ratios (usually more than 500 ppm). Therefore, we replaced all the exhaust pumps. To further ensure that the leakage problem during sampling is solved, we regularly check for leaks before sampling by closing the needle valves leading to the pumps and monitoring in case there is any flow with the flow meter attached after the pump. Since the replacement we have not observed any indication of leakage. Seven samples which were suspected to be contaminated due to this issue were consequently excluded. The sample extraction efficiency since then has usually been better than 99 %. We also conducted a blank test to check the presence of any leaks or contamination during sample processing but did not observe any of these issues. Five more samples were excluded in 2014 due to a strong mismatch among triplicates in the measured CO2 after the sample extraction which indicated contamination.

Prior to radiocarbon measurement, the extracted CO2 was analyzed for the stable isotopes δ13C and δ18O using an isotope ratio mass spectrometer (IRMS, Finnigan MAT 250) at the Climate and Environmental Physics Division of the University of Bern, which has an accuracy and precision of better than 0.1 ‰ for both δ13C and δ18O (Leuenberger et al., 2003). 14C analysis of the extracted CO2 was performed with an accelerator mass spectrometer (AMS), MICADAS (MIni CArbon DAting System), at the Laboratory for the Analysis of Radiocarbon (LARA) at the Department of Chemistry and Biochemistry of the University of Bern (Szidat et al., 2014). The Automated Graphitization Equipment (AGE) was used to prepare solid target gas (Nemec et al., 2010) from the extracted CO2 stored in 50 mL glass flasks. A measurement series consisted of up to 15 air samples converted to 30 solid graphite targets (duplicates), together with four and three targets from CO2 produced by combustion of the NIST standard oxalic acid II (SRM 4990C) and fossil CO2 (Carbagas, Gümligen), respectively, which were used for the blank subtraction, standard normalization, and correction for isotopic fractionations. For the fractionation correction, δ13C values of the AMS were used, which show a long-term standard uncertainty of ±1.2 ‰ (Szidat et al., 2014). The AMS δ13C values agree well on average with the corresponding IRMS results, revealing a statistically insignificant difference of -0.2 ± 1.2 ‰ with slightly more depleted AMS results.

Data reduction was performed using the BATS program (Wacker et al., 2010). The uncertainty of an individual 14C measurement typically amounts to ∼ 2.1 ‰, including contributions from counting statistics (∼ 1.1 ‰), corrections of normalization (i.e., blank subtraction, standard normalization, and correction for isotopic fractionations) (∼ 1.1 ‰), and an unaccounted-for long-term variability of sampling and 14C analysis according to Szidat et al. (2014) (1.5 ‰). These contributions are comparable to previous observations (Graven et al., 2007).

During calculation of weighted averages of the duplicates, the uncertainty of the mean is determined with the contributions of the counting statistics and the normalization, whereas the uncertainty of the unaccounted-for long-term variability is considered fully afterwards, as this contribution cannot be reduced by averaging of two measurements performed on the same day. This uncertainty of the weighted average typically amounts to ∼ 1.9 ‰; it is compared with the standard deviation of the duplicates, and the larger of these values is used as the final uncertainty of the duplicates. The mean of the three individual samples from the same day, which is used below in Sect. 2.4.1 as Δ14Cmeas, is then determined and associated with the average uncertainty of the three duplicates, as the variability of the three samples is comparable to this average uncertainty for all cases.

As the 14C / 12C from Beromünster was measured at the LARA laboratory in Bern, whereas the corresponding background samples from Jungfraujoch were analyzed at the low-level counting (LLC) facility of the Institute of Environmental Physics, Heidelberg University, the data sets needed to be adjusted to each other. A recent interlaboratory compatibility test between the LARA lab (code no. 2) and Heidelberg (LLC) estimated a small bias (Hammer et al., 2016). The measurement bias (i.e., the mean difference of the measured Δ14C minus the consensus value of the participating laboratories for all investigated CO2 samples) is +1.8 ± 0.1 and -0.3 ± 0.5 ‰ for Bern and Heidelberg, respectively, from which the bias between both labs of 2.1 ± 0.5 ‰ is determined with a larger measured Δ14C for Bern. Consequently, 2.1 ± 0.5 ‰ was subtracted from the 14C measurements of the Beromünster samples.

Figure 2a shows the in situ measured hourly mean CO2 dry-air mole fractions at Beromünster (black) from the 212.5 m sample inlet matching at hours when air samples were collected for radiocarbon analysis and the corresponding background CO2 at Jungfraujoch (blue). During the measurement period, we recorded CO2 mixing ratios between 389 and 417 ppm. Spikes of CO2 were observed mainly during winter, associated with weak vertical mixing and enhanced anthropogenic emissions, while lower CO2 mixing ratios were recorded during summer due to strong vertical mixing and photosynthetic uptake.

Isotopic analysis of the air samples yielded Δ14Cmeas between -12.3 and +22.8 ‰, with no clear seasonal trend, after correction for the model-simulated contribution from NPPs (Fig. 2b). Based on the simulations described in Sect. 2.4.2, we have calculated a mean enhancement in Δ14C of +1.6 ‰ and a maximum of +8.4 ‰ due to NPPs. This agrees qualitatively with the coarse-resolution simulations of Graven and Gruber (2011), which suggest a mean enhancement of +1.4 to +2.8 ‰ over this region (Graven and Gruber, 2011). While about 70 % of this contribution is due to Swiss NPPs, the remaining contribution is of foreign origin. About 75 % of the contribution from the Swiss NPPs is due to Mühleberg, which is located west of Beromünster and hence frequently upstream of the site, due to the prevailing westerly winds (Oney et al., 2015). Note that each data point represents a mean value of the triplicate samples collected consecutively with a standard error of 2 ‰ among triplicates. During this period, the background Δ14C values measured at Jungfraujoch varied between 15 and 28 ‰. Regional depletions in Δ14C due to fossil fuel emissions, i.e., differences between Beromünster and the clean-air reference site Jungfraujoch, were in the range of -0.7 to -29.9 ‰ with a mean value of -9.9 ‰.

Figure 2c shows the corresponding CO2ff determined after correcting for radiocarbon emissions from NPPs. The typical uncertainty in CO2ff is 1.2 ppm calculated by quadratically combining a mean Δ14C measurement uncertainty of 2.0 ‰ in both the sample and the background values, 0.3 ‰ from biospheric correction, 0.5 ‰ from interlaboratory offset, and a mean uncertainty of 1.2 ‰ in the estimation of 14C contribution from NPPs. A mean fossil fuel CO2 contribution of 4.3 ppm was calculated from these samples. Few cases, notably the sample from 27 March 2014, showed a higher CO2ff and a strong depletion in Δ14Cmeas, consistent with the high CO2 mixing ratio shown in the top panel. This could be due to a strong local fossil fuel contribution or a polluted air mass transported from other regions of Europe coinciding with the grab samplings. As this event occurred during a period with moderate temperatures (mean temperature of 6.8 ∘C measured at the highest level of the Beromünster tower between March and May), strong fossil fuel CO2 emissions due to heating are not expected. The FLEXPART-COSMO transport simulations for this event suggest an air mass origin from southeastern Europe (see Supplement). Periods with winds from the east, colloquially known as Bise, are well known to be associated with very stable boundary layers and correspondingly strong accumulation of air pollutants during the cold months of the year between autumn and spring. Air masses reaching Beromünster from eastern Europe have recently been reported to contain unusually high levels of CO during late winter and early spring periods, coinciding with this sampling period (Oney et al., 2017).

By subtracting the background and fossil fuel CO2 contributions from the measured mixing ratios, CO2bio values were also determined, ranging between +11.2 and -12.4 ppm (Fig. 2d). Even if there is no clear seasonal trend, the lowest CO2bio values were recorded during summer, implying net photosynthetic CO2 uptake, while most of the values in winter are positive or close to zero due to respiration. During summer 2015, we observed strong variability in both CO2 and CO2bio (Fig. 2a and d). However, this period was one of the hottest and driest summers in central Europe (Orth et al., 2016). In Switzerland, it was the second-hottest summer since the beginning of measurements in 1864 with most of the extreme dates in July (MeteoSuisse, 2015). Such climate extremes can lead to enhanced respiration and reduced photosynthesis and, in turn, higher CO2 and CO2bio in the atmosphere. Looking specifically at the two data points in June and July 2015, the daily average temperatures recorded at Beromünster were 24.6 and 26 ∘C at the highest inlet of 212.5 m (Fig. 2e). Based on measurements at Beromünster and other cities of the CarboCount CH network in 2013, Oney et al. (2017) reported that for a daily mean temperature of greater than 20 ∘C the biosphere over the Swiss plateau tends to become a net CO2 source. The observed positive spikes in CO2 (Fig. 2a) and CO2bio (Fig. 2d) likely resulted from such extremes.

From the simultaneous CO and radiocarbon measurements, we calculated an RCO of 13.4 ± 1.3 mmol CO/mol CO2 with a correlation coefficient (r2) of 0.7 and a median value of 11.2 mmol CO/mol CO2 (note that change in RCO is insignificant when we use smoothed 14C background from Jungfraujoch). When we split the data seasonally, RCO values of 12.5 ± 3.3 and 14.1 ± 4.0 mmol CO/mol CO2 were obtained during winter and summer, respectively (Table 1). Even if the two values are not significantly different considering the uncertainties, the very low correlation coefficient during summer (r2=0.3) implies a larger uncertainty in the derived RCO. Our wintertime estimate is well within the range of values from previous studies (10–15 mmol mol-1) observed at other sites in Europe and North America (Gamnitzer et al., 2006; Vogel et al., 2010; Turnbull et al., 2011a). To test the sensitivity of this ratio to the selection of background site, we additionally calculated RCO using background values estimated with the REBS method from the in situ CO measurements at Beromünster instead of Jungfraujoch. The value obtained in this way (12.7 ± 1.2, r2=0.6) is not significantly different from the value obtained using Jungfraujoch as the background site. Considering the persistent decrease in CO emissions (Zellweger et al., 2009) in response to the European emission legislation, our estimated RCO is surprisingly high. A recent study investigating the CO / CO2 ratio from road traffic in Islisberg tunnel, Switzerland, also observed a significant decrease in this ratio compared to previous estimates, pointing to a substantial reduction in CO emissions from road traffic, with a CO / CO2 ratio of 4.15 ± 0.34 ppb ppm-1 (Popa et al., 2014). This may indicate a significant contribution from non-road traffic emissions, which account for more than 70 % of the total CO2 emissions leading to the high apparent RCO.

The RCO value derived in this study is significantly higher than the anthropogenic CO / CO2 emission ratio of 7.8 mmol mol-1 calculated from Switzerland's greenhouse gas inventory report for 2013 (FOEN, 2015b, a). However, this can be due to enhanced CO emissions transported from other European cities towards Beromünster. Oney et al. (2017) observed particularly large CO / CO2 ratios at Beromünster during several pollution events in late winter and early spring 2013 which were associated with air mass transport from eastern Europe, where poorly controlled combustion of biofuels and coal likely results in high ratios.

Figure 3 shows the seasonally resolved correlations of ΔCO with ΔCO2 derived from in situ measured CO and CO2 enhancements over the background observed at Jungfraujoch, for which we estimated a tracer ratio of 8.3 ± 0.1 mmol mol-1 (r2=0.5) for the entire measurement period. Considering the seasonally resolved ΔCO / ΔCO2 ratios, barely any correlation is observed in summer, and weak correlations (r2 < 0.4) are observed during spring and autumn. This can be due to the dominance of biogenic fluxes over fossil fuel fluxes during these periods of the year. From measurements during winter, when the two species are most strongly correlated, a ΔCO / ΔCO2 ratio of 7.3 ± 0.1 mmol mol-1 (r2=0.9) is obtained. Recently, Oney et al. (2017) reported a higher wintertime ratio of 8.3 mmol mol-1 for the same combination of measurements at Beromünster and Jungfraujoch but for a different time period. If we consider only winter 2013 as in their data, we obtain essentially the same value, while much lower ratios of 6.5 and 6.4 mmol mol-1 were calculated for 2014 and 2015, respectively. The higher ratios in winter 2013 are likely related to the unusually cold conditions and extended periods of air mass transport from eastern Europe. Note that, in contrast to RCO, these enhancement ratios also include emissions from non-fossil sources such as biofuels and biomass burning as well as the influence of biogenic fluxes. The Swiss national inventory attributes about 15 % of total CO2 emissions in 2014 to non-fossil-fuel sources (FOEN, 2015b). If we correct for these sources assuming a constant contribution throughout the year, the wintertime ΔCO / ΔCO2 ratio for the 3-year data becomes 8.7 mmol mol-1.

This ratio of 8.7 mmol mol-1 is still about 30 % lower than the RCO estimate for the same period of 12.5 mmol mol-1, shown as a black line in Fig. 3. This suggests that despite the strong correlation between ΔCO and ΔCO2 in winter the regional CO2 enhancements are not only caused by anthropogenic emissions but include a significant contribution from biospheric respiration. Miller et al. (2012) showed that such strong correlations between CO2 and CO during winter may arise from respiratory fluxes co-located with fossil fuel fluxes trapped under the wintertime shallow and stable boundary layer but with strongly biased ratios when compared to RCO. Turnbull et al. (2011b) also observed a substantial contribution of biospheric CO2 fluxes even during winter (20–30 % from non-fossil-fuel sources including photosynthesis and respiration) from samples collected at two sites in East Asia. The magnitude of these fluxes was roughly similar to the CO2ff flux when continental background was used (Turnbull et al., 2015). Hence, the observed correlation between ΔCO and ΔCO2 in this study is not only due to spatially and temporally correlated sources but is caused to a large extent by meteorological variability associated with more or less accumulation of trace gases in the boundary layer irrespective of their sources. This interpretation is also supported by the fact that a strong correlation (r2 > 0.7) was also observed between CO and CH4 during winter at the same tower site (Satar et al., 2016) despite their sources being vastly distinct. In Switzerland about 80 % of CH4 emissions are from agriculture (mainly from ruminants), while more than 85 % of CO emissions are from the transport sector and residential heating (FOEN, 2015a).

Figure 4 shows the hourly mean CO mixing ratios at Jungfraujoch and Beromünster between 2013 and 2015. CO mixing ratios as high as 480 ppb were recorded at Beromünster, while generally lower CO values were recorded at the more remote site Jungfraujoch. A pronounced seasonality in CO can be observed at Beromünster with higher values in winter and lower values during summer due to stronger vertical mixing and chemical depletion of CO by OH (Satar et al., 2016). The hourly mean CO2ff time series calculated using these continuous CO measurements and the seasonally resolved RCO values derived using the radiocarbon measurements are displayed in Fig. 4c. A seasonal trend in the calculated CO2ff is observed with frequent spikes of CO2ff during winter, while summer values show less variability. We calculated a monthly mean amplitude (peak to trough) of 6.3 ppm with a maximum in February and a minimum in July. During the measurement period, we observed CO2ff mixing ratios ranging up to 27 ppm coinciding with cold periods and likely from enhanced anthropogenic emissions due to heating. Instances of slightly negative CO2ff contributions, which occurred during less than 5 % of the time, were associated with negative enhancements in CO (i.e., ΔCO < 0). This could be simply due to an overestimation of background values by the REBS function during these periods.

Figure 5a shows the hourly averaged residual CO2bio values, which exhibit not only a clear seasonal cycle but also a considerable scatter in all seasons, ranging from -13 to +30 ppm. During winter, most values were close to zero or positive, implying a dominance of respiration fluxes. In summer, conversely, pronounced negative and positive excursions were observed mostly due to the diurnal cycle in net CO2 fluxes, which are dominated by photosynthetic uptake during daytime and respiration at night. Another factor contributing to such variations may be the application of a constant emission ratio neglecting any diurnal variability (Vogel et al., 2010).

It should also be noted that any non-fossil-fuel CO2 sources such as emissions from biofuels would be incorporated into the CO2bio term since CO2ff in Eq. (1) represents the fossil fuel sources only, adding more variability to the data set. In order to reduce the influence of these diurnal factors, we have looked into afternoon CO2bio values (12:00–15:00 UTC), when the CO2 mixing ratios along the tower are uniform (Satar et al., 2016) and RCO variability is minimal. Similar to the seasonal pattern in Fig. 5a, a clear seasonal cycle in biospheric CO2 can be observed (Fig. 5b) in agreement with biospheric exchange, but both positive and negative extremes are less frequently observed (-12 to +22 ppm).

The variation in CO2bio during afternoon (12:00–15:00 UTC) was recently estimated at this site at a range of -20 to +20 ppm by combining observations and model simulations for the year 2013 (Oney et al., 2017). Our estimates are more positive than those in their study, due to the higher RCO which results in lower CO2ff and correspondingly higher CO2bio values.

Biospheric CO2 shows a seasonally dependent diurnal variation as shown in Fig. 6. During winter (December–February), the biospheric CO2 component remains consistently positive (+2 to +5 ppm) throughout the day, implying net respiration fluxes. In summer, a clear feature with increasing CO2bio values during the night peaking between 07:00 and 08:00 UTC (i.e., between 08:00 and 09:00 local time) can be observed. This buildup during the night can be explained by CO2 from respiration fluxes accumulating in the stable and shallow nocturnal boundary layer. Then, after sunrise, the early morning CO2bio peak starts to gradually decrease due to a combination of onset of photosynthesis and enhanced vertical mixing due to the growth of the boundary layer. At Beromünster, a decrease in CO2 mixing ratios from both processes is visible more or less at the same time at the 212.5 m height level. As reported by Satar et al. (2016), this decrease in early morning CO2 concentrations at the 212 m inlet lags behind the decrease at the lowest sampling level of 12.5 m by approximately 1 h. Between 12:00 and 15:00 UTC, when the daytime convective boundary layer is fully established, the biospheric CO2 continues to become more negative, implying net photosynthetic uptake, which eventually stabilizes for 3–5 h until nighttime CO2bio accumulation starts.