The High-Altitude Research Station Jungfraujoch (stretching from 3455 to 3585 m a.s.l., 46°32^′51^′′ N, 7°59^′7^′′ E, JFJ), established in 1931, is located on a mountain ridge in the Swiss Alps. Among others, this station is part of the Global Atmosphere Watch (GAW) network as well as of the Network for the Detection of Atmospheric Composition Change (NDACC). Furthermore, JFJ is labelled as a Class 1 Station of the European-wide Integrated Carbon Observation System (ICOS) Research Infrastructure (Heiskanen et al., 2022) since May 2018 (Yver-Kwok et al., 2021). Because of its high elevation, JFJ is a well-recognized international background station (Leuenberger and Flückiger, 2008). However, it is also intermittently impacted by direct transport from the polluted planetary boundary layer, most frequently during daytime from April to September (Henne et al., 2010).

A fortnightly air sampling program measuring ¹⁴CH₄ and ¹⁴CO₂ at JFJ every two weeks started in January 2019. Until March 2023, two morning grab air samples were taken using a membrane pump (N022AN.18, KNF Neuberger AG, Germany) that directly pumped air from the Sphinx terrasse (3580 m a.s.l.) at JFJ into two PE-Al-PE 120 L bags (Tecobag, Tesseraux Spezialverpackungen GmbH, Germany) through a dedicated sampling line (Synflex Decabon 1300 Tubing, OD =6 mm). The sampled air was dried with a magnesium perchlorate dryer (Mg(ClO₄)₂, ACS reagent, ThermoFisher, USA) to avoid condensation and further reactions between water vapor and other sampled gas species. The sampling duration per bag was commonly between 20 and 55 min depending on the ambient pressure, temperature and the flow resistance during sampling due to the dryer and the long length and small diameter of the sampling line. The morning grab samplings were performed manually and as early as possible in the morning to avoid the influence of the daytime planetary boundary layer; in practice, it commonly occurred between 07:00 and 10:00 UTC (i.e., 08:00 and 12:00 local time, UTC+1/UTC+2), depending on the seasonal train schedule ensuring the public transport up to JFJ.

To increase the reproducibility of the air sampling procedure and its representativeness as a background measurement, an automated air sampling system was developed replacing the manual sampling from 18 April 2023 onwards (Fig. 1). Its development was guided by two principles: (1) air sampling should be restricted to nighttime when the station mostly resides within the free troposphere, allowing for the collection of temporally integrated samples over multiple days; and (2) air sampling should occur as far as possible passively, i.e., without any mechanical pumps in contact with the sampled air, thereby minimizing the risk of contamination from membrane outgassing. The novel Jungfraujoch Air Sampling System (JASS) consisted of a custom-made electropolished steel sampling tank (volum = 260 L, length (L) = 140 cm, outer diameter (OD) =51 cm, Bechtiger Edelstahl AG, Switzerland) coupled to a sampling box containing a series of valves, all controlled by a home-made control unit including a Raspberry Pi 4B module (Raspberry Pi Ltd, United Kingdom) (Fig. 1). The whole system was installed under the roof of the research station at JFJ, at about 3462 m a.s.l. (Fig. S1 in the Supplement). Its design allows a routine integrated nighttime air sampling procedure lasting 6 h every night (between 00:00 and 06:00 UTC) over 14 d, so integrating 84 h in total. Compared to the initial fortnightly morning grab sampling strategy, integrated sampling gives a more representative air mixture for every two-week period of interest (Fig. S2).

The inlet of the sampling line (Synflex Decabon 1300 Tubing, OD =12 mm, L∼10 m) was protected by a dust filter (Fig. 1). A flushing membrane pump (N022AN.18, KNF Neuberger AG, Germany) continuously conveyed outside air through this sampling line and a water separator to the sampling box. Through two T-pieces (SS-12M0-3, Swagelok, USA) integrated in this sampling line, part of the sampled air could be further directed into the sampling box, either automatically using line (1), or manually using line (2) (Fig. 1). At the beginning of a sampling period, the tank was evacuated using a vacuum pump (MV 2 NT, Vacuubrand GmbH + Co KG, Germany); the end pressure was ≤0.7 mbar. During automatic air sampling, the sampling line (1) was open and worked as follows: both 2/2-way solenoid valves A1 and A2 (0330-A-02, Bürkert, Germany) were open and the 3/2-way solenoid valve B (0330-F-02, Bürkert, Germany) was in the position connecting this line with the tank (blue connection, Fig. 1); the mass-flow controller (MFC, F-201DV, Bronkhorst, The Netherlands) between A1 and A2 ensured a controlled air flow into the system and the increasing pressure in the tank was monitored by a pressure gauge (RPT 200 AR, Pfeiffer Vacuum, Germany); in front of the MFC, a Mg(ClO₄)₂ dryer dried the sampled air and a 0.5 µm filter (SS-4FWS-05, Swagelok, USA) protected the MFC from particulate matter. After 6 h of sampling, both A1 and A2 valves were closed. This sampling procedure was repeated every night. After 14 nighttime samplings, the pressure in the tank reached 600–650 mbar. The mean ambient atmospheric pressure at JFJ was 656 mbar over the period 2019–2024 (Emmenegger et al., 2025d); we configured our passive sampling system in a way where the tank pressure was always lower than ambient pressure, otherwise the passive sampling based on the pressure difference between the tank and the ambient pressure would not work. On-site, the transfer of the sampled air from the tank into a new bag was prepared in this manner: a dedicated transfer pump (N922SPE, KNF Neuberger AG, Germany) was turned on and the manual 3/2-way valve D2 was turned to the left (Fig. 1); the position of the 3/2-way valve D1 stayed by default turned to the left. The Python script running on the Raspberry Pi was then restarted to turn on the vacuum pump and switch the 3/2-way solenoid valve B in the position connecting the tank with the pump line (red connection, Fig. 1). At this stage, the transfer pump was evacuating the tank and after ∼30 s of line flushing, a bag was connected to the line for the transfer of the sampled air. After about one hour, enough air was transferred to the bag for further ¹⁴C analyses. It was then closed and disconnected from the transfer pump (tank pressure at around 70–90 mbar). To accelerate the evacuation of the tank and condition it again for the next sampling, the valve D2 was then turned to the right, and the vacuum pump was evacuating the tank. The system was left like this for the rest of the day and, at 23:59 UTC, valve B switched automatically back to its sampling position (blue connection, Fig. 1) (final tank pressure ≤0.7 mbar), the vacuum pump turned off and the whole system was ready to start the next sampling period. The whole sampling system was validated to be leak-tight in our laboratory in Bern before being transferred to JFJ. The low pressure achieved in the tank before each new sampling every two weeks regularly validated the tightness of the sampling system.

For both morning grab and integrated nighttime sampling strategies, the air collected in bags was transferred to the LARA laboratory at the University of Bern (Szidat, 2020), where a dedicated extraction line for ¹⁴CH₄ and ¹⁴CO₂ analysis was available (see Espic et al., 2019 for details). The extraction was generally done within one week after sampling and most of the time the day after. The typical air volume for one analysis was 60 L. During a preconcentration step, this air was pumped successively through three traps filled either with fiber glass or activated charcoal and cooled down with liquid nitrogen; this allowed us to preserve the whole amount of CH₄ contained in the initial 60 L of air by trapping or pumping away the main other gas species, especially nitrogen and oxygen. CO₂ was quantitatively trapped in the first trap and could be collected separately (see below). After that, the gas sample volume was about 10 mL, small enough to be run through a gas chromatograph (GC, 7890B, Agilent, USA; ShinCarbon ST 80/100 packed column; thermal conductivity detector, He as carrier gas) to purify the sample. CH₄ was isolated quantitatively from the remaining CO and CO₂ present in trace quantities due to different elution times in the GC (CO: ∼ 2 min; CH₄: ∼ 8 min; CO₂: ∼ 13 min). Downstream of the GC, the extracted pure CH₄ was converted into CO₂ by combustion in a flow oven at 950 °C using copper oxide wires as a catalyst. The CH₄-derived CO₂ (∼ 60 to 70 µg of carbon, µgC) was transferred into glass ampoules (OD =4 mm) and sealed for gas radiocarbon measurements. After the CH₄ extraction procedure, the CO₂ of the sample held in the first trap was recovered by cryogenic transfer into a ∼ 55 mL glass flask. The carbon mass of the recovered CO₂ was typically greater than 1 mgC, which was transformed into graphite using an automated graphitization equipment (AGE) (Němec et al., 2010). Concerning the quality control of the ¹⁴CH₄ extraction system, Espic et al. (2019) reported a cross-contamination of 0.4 ± 0.2 % (n=2) from the previous sample and an extraction yield of 101.2 ± 1.4 (n=13).

Radiocarbon analyses of the gaseous CH₄-derived CO₂ samples and the graphite CO₂ samples were performed either at LARA or at the Laboratory of Ion Beam Physics, ETH Zürich, Switzerland, using the same type of AMS (MIni CArbon DAting System, MICADAS), equipped with a gas ion source (Ruff et al., 2007; Synal et al., 2007). Glass ampoules were cracked in a dedicated gas inlet system (Wacker et al., 2013) and the CH₄-derived CO₂ was diluted with He to ∼ 5 %, transferred into a syringe, and then fed into the ion source using a constant gas flow. Graphite samples were introduced directly to the MICADAS source. For both types of samples, raw ¹⁴C / ¹²C as well as ¹³C / ¹²C ratios were converted into F¹⁴C and δ13C values, respectively, by performing a blank subtraction as well as a standard normalization and correction for isotope fractionation (only for F¹⁴C) using ¹⁴C-free CO₂ (Blank) (F¹⁴C =0) and CO₂ produced from the primary NIST standard oxalic acid II (OX-II) (SRM 4990C) (F¹⁴C =1.34066, δ13C =-17.8‰), respectively. For the gas measurements, standards came from two gas bottles directly attached to the gas inlet system; both mixtures consisted of 5 % CO₂ with the F¹⁴C value of interest and 95 % He. For the graphite measurements, both standards underwent the same sample preparation as the CO₂ samples and were measured with them in the same magazine. Gas samples were commonly measured starting with 2–4 OX-II and 2 Blank standards, then with the samples in chronological sampling order and finishing with 2–4 OX-II and 2 Blank to account for potential drifts. The dataset presented here was built over 40 measurement days in 6 years. Per measurement day, about 24–32 standards and samples, including about 5–8 from JFJ, were measured; the remaining samples came from other sampling sites or further projects. The average standard deviation of all the OX-II standards used for normalization within a single measurement day was ± 0.007 (in F¹⁴C) whereas the average single OX-II precision derived from counting statistics was ± 0.010 (n=204). Graphite samples were generally measured in a 39-position magazine containing 3–4 OX-II and 3 Blank samples. The dataset presented here was built over 35 magazines. The average standard deviation of all the OX-II standards used for normalization within a single measurement day was ± 0.0018 (in F¹⁴C) whereas the average single OX-II precision derived from counting statistics was ±0.0025 (n=117). The final data evaluation was done using the BATS tool (Wacker et al., 2010). The typical measurement precision is 8 ‰ for ¹⁴CH₄ and 1.5 ‰ for ¹⁴CO₂ (see below). Throughout the current work, ¹⁴C results are reported using the notation Δ14C, and calculated with age correction as the parameter Δ in Stuiver and Polach (1977).

At JFJ, in the framework of the ICOS measurement program, CO₂ and CH₄ concentrations are measured continuously using cavity ringdown spectroscopy; corresponding hourly average values were used in the present study (Emmenegger et al., 2025b, c). These values were used, among other purposes, to determine the stability of CH₄ and CO₂ concentrations in the transfer bags between sampling and extraction. CH₄ and CO₂ concentrations were measured in the remaining sampled air after extraction with a cavity ringdown spectrometer (Picarro G2401, Picarro Inc., USA) calibrated with three gas bottles with known CH₄ and CO₂ concentrations (1861.2, 1955.9, and 2206.7 ppb for CH₄; 379.21, 418.38, and 458.62 ppm for CO₂) (Carbagas, Switzerland). Mean differences between concentrations measured after extraction and measured in situ were 1.3 ± 2.8 ppb and 0.7 ± 1 ppm, for CH₄ and CO₂, respectively (n=13, between January and July 2024).

Furthermore, integrated samples have been collected at JFJ since 1986 to analyze atmospheric Δ14CO₂, first started by the University of Heidelberg, now run as part of the ICOS Research Infrastructure (Emmenegger et al., 2025a; Hammer et al., 2017; Levin and Kromer, 2004). The dedicated setup draws ambient air throughout 14 d through a sodium hydroxide solution, in which CO₂ is chemically absorbed. We chose the same 14 d schedule for our nighttime sampling as the ICOS schedule to enable a comparison between both datasets. ICOS Δ14CO₂values are also reported as Δ values according to Stuiver and Polach (1977). Furthermore, atmospheric Δ14CO₂ values from the Mace Head Atmospheric Research Station (MHD, Ireland) are used here for comparison (Hammer and Levin, 2023); due to its location exposed to westerly winds from the North Atlantic Ocean, MHD is often used as European background station.

We also made use of continuous ²²²Radon (in the following referred as Rn) measurements at JFJ with a two-filter dual loop alpha particle detector, operated by the University of Basel since 2009 (Griffiths et al., 2014) and which is meanwhile part of ICOS (Fig. S1). Rn is emitted from land surfaces into the atmosphere and because of its half-life of 3.8 d, it is a potential tracer of recent land contact. In this study, Rn was used as a proxy to distinguish free tropospheric samples from those originating from the planetary boundary layer (i.e., with recent land contact). To account for variations in ambient temperature and pressure, the raw hourly Rn concentrations (Conen, 2025) were converted into Rn values at STP conditions (i.e., T=0 °C, P=101325 Pa).

To evaluate the accuracy and stability of our Δ14CH₄ and Δ14CO₂ measurements over time, regular measurements of a pressurized air bottle (PAB) (Carbagas, Switzerland) considered as an internal standard have been performed since March 2022 in parallel to the measurements of the JFJ samples; this air strictly underwent the same extraction and measurement procedures as the JFJ samples, commonly just before them. A first bottle was measured until end of July 2023 (CH₄ =1997 ± 2 ppb, and CO₂ =434.5 ± 0.1 ppm), replaced by a second one in August 2023 (CH₄ =2178 ± 2 ppb, and CO₂ =455.8 ± 0.1 ppm). It should be noted that the Δ14CH₄ and Δ14CO₂ values of both bottles were a priori unknown and that we used them to evaluate the stability of the values derived from our measurement procedure.

One goal of the present study was to evaluate the influence of nuclear ¹⁴CH₄ and ¹⁴CO₂ emissions on our atmospheric Δ14CH₄ and Δ14CO₂ measurements at JFJ. For this purpose, the amount and transport of ¹⁴CH₄ and ¹⁴CO₂ molecules from nuclear emissions were simulated using the Lagrangian particle dispersion model FLEXible PARTicle (FLEXPART) (Pisso et al., 2019) in the version adopted for inputs from the numerical weather prediction model COSMO (Henne et al., 2016). Here, we use the COSMO analysis product of the Swiss Federal Office of Meteorology and Climatology (MeteoSwiss), based on high spatial resolution (1 km × 1 km) model simulations and using a local ensemble transform Kalman filter (LETKF) meteorological data assimilation system (Schraff et al., 2016). Meteorological analysis fields from this product were available at 1 h temporal resolution for the Alpine domain (approximately 0–17° E, 42–50° N). The FLEXPART-COSMO model was used in previous atmospheric studies including the verification of the Swiss CH₄ emission inventory (Henne et al., 2016), the influence of nuclear ¹⁴CO₂ emissions on Δ14CO₂ at a Swiss tall tower (Berhanu et al., 2017), the analysis of CO₂ and δ13CO₂ at JFJ (Pieber et al., 2022), and for inverse modeling of halocarbon emissions over Switzerland (Katharopoulos et al., 2023). Here, the model was operated in the same way as in Katharopoulos et al. (2023). In short, 50 000 model particles were released continuously from JFJ for every 3 h interval and traced backwards in time for 4 d or until they reached the domain boundaries. Afterwards, the integration of the particles' path was continued for up to 10 d in a European scale FLEXPART-IFS simulation driven by hourly inputs from the European Centre for Medium-Range Weather Forecasts (ECMWF) HRES operational forecast/analysis product available at 0.1°×0.1° resolution. The residence time of particles below a height of 50 m above model ground is then estimated in 3 h intervals and divided by air density provides so called source sensitivities (or concentration footprints).

Nuclear ¹⁴C emissions between 2019 and 2023 from the nuclear power plants (NPPs) located in the modeling area (Fig. S3) were mostly estimated using the compilation of nuclear ¹⁴C emissions of Laemmel et al. (2025). The atmospheric transport modeling was performed only for the period 2019–2023, as the input parameters for 2024 were not yet fully available. For most countries (i.e. Bulgaria, the Czech Republic, Romania, Slovakia, Slovenia, Spain, the Netherlands, and the United Kingdom) only annual total ¹⁴C emissions for the NPPs were available. For some countries, more detailed information could be used including quarterly ¹⁴C emissions from NPPs in France and Germany and monthly emissions for the Swiss NPPs Leibstadt and Gösgen, the Swedish NPPs Forsmark, Oskarshamn, and Ringhals, some NPPs in the United Kingdom and the French nuclear fuel reprocessing plant (NFRP) La Hague. Monthly inorganic and organic ¹⁴C emissions from the Hungarian NPP Paks were also available for 2019 in this dataset. In addition to these published values, monthly inorganic and organic ¹⁴C emissions for 2020–2023 were kindly made available by the operating company of the NPP Paks for this simulation. Reported emissions from the NPPs located in Belarus, Belgium, Russia, and Ukraine were not available; we estimated the annual ¹⁴C emissions for each of these plants by multiplying the annual electricity production (Laemmel and Szidat, 2025) by reactor-specific emission factors (EF) of 0.19, 0.41, and 1.3 TBq GWa⁻¹ (GWa = gigawatt ×1 year) for PWR, VVER (water-cooled water-moderated energy reactor), and LWGR (light water graphite reactor), respectively. EF values for PWR and VVER were derived from the work of Fujita et al. (2025) and EF value for LWGR from the work of Zazzeri et al. (2018).

In a further step, ¹⁴CH₄ and ¹⁴CO₂ emissions per NPP were derived, as the reported values were mainly for total ¹⁴C. We used data for inorganic and organic ¹⁴C emissions that were reported in some countries (e.g., for NPPs in Germany, Hungary, Spain, Sweden, and Swizerland). For other countries we assumed that PWRs and VVERs emit 75 % of the total ¹⁴C amount in organic form (so 25 % in inorganic form) and that all the other reactors (e.g. boiling water reactor (BWR), LWGR, and pressurized heavy water reactor (PHWR)) emit ¹⁴C entirely in inorganic form, as also was assumed by Fujita et al. (2025). Furthermore, we assumed that the inorganic ¹⁴C form is entirely composed of ¹⁴CO₂ and that ¹⁴CH₄ represents 72.5 % of the organic ¹⁴C form; this last value was derived from the study of Kunz (1985) who analysed the composition of hydrocarbons of gaseous effluents at two PWRs in the USA. They reported values of 68 % and 77 % at the PWR Ginna and PWR Indian Point, respectively. Finally, the annual, quarterly or monthly emission amount of ¹⁴CO₂ and ¹⁴CH₄ available for each plant was equally divided into three-hour intervals to compute an emission rate per output step of the FLEXPART transport simulation. FLEXPART-derived source sensitivities, convolved with these emission rates yield ¹⁴C mole fractions at the sampling site and a final nuclear correction Δ14C^Nuc can be calculated assuming a mass balance model for C and ¹⁴C (Graven et al., 2019) (See detailed equations in the Supplement).

Another goal of the present study was to evaluate the representativeness of our atmospheric Δ14CH₄ measurements at JFJ on a global level. For this goal, we compared our data with calculated atmospheric Δ14CH₄ values from a one-box model developed by Fujita et al. (2025). In this study, we extended their simulation until 2024 (Fig. S4) by updating their posterior CH₄ emission scenarios since 2013 (i.e., average of posterior CEDS, EDGARv5, and EDGARv6 scenarios; see Fujita et al., 2025). The anthropogenic CH₄ emissions for 2013–2022 were extended by using EDGARv8 (European Commission Joint Research Centre, 2023). To match the consistency with the posterior anthropogenic biogenic (BIO) and fossil fuel (FF) emissions in Fujita et al. (2025), the mean differences between EDGARv8 and Fujita et al. (2025) were calculated for 2008–2012 (BIO: 2.1 Tg yr⁻¹, FF: 9.8 Tg yr⁻¹) and then added to the values of EDGARv8 after 2013, respectively. For 2023–2024, the emissions of the Shared Socioeconomic Pathways (SSPs) were used. Here, we adopted SSP5-8.5 scenario in 2030 (Gidden et al., 2019) and linearly interpolated it between the EDGARv8 2022 emissions and the 2030 scenario emissions to the years 2023 and 2024. Natural biogenic CH₄ emissions were optimized based on the CH₄ mass balance equation to keep the consistency between our simulations and observed global mean CH₄ mole fractions by NOAA/GML (Lan et al., 2025a). To evaluate the ¹⁴C signature of the biospheric CH₄ sources, Δ14CO₂ values for the time period 2013–2024 were derived from the SSP5-8.5 scenario (Graven et al., 2020). To extend the nuclear ¹⁴CH₄ emissions beyond 2013, we considered the posterior annual nuclear ¹⁴CH₄ emissions derived by Fujita et al. (2025). Assuming that only PWRs and VVERs are emitting ¹⁴CH₄, a mean annual emission factor Φ (GBq GWa⁻¹) was computed by dividing these emissions by the annual total electricity production by PWRs and VVERs derived from the data compilation by Laemmel and Szidat (2025). Considering the years 2008–2012, the mean Φ value was 250 GBq GWa⁻¹, the min Φ value was 243 GBq GWa⁻¹, and the max Φ value was 259 GBq GWa⁻¹. Considering these three Φ values and the annual total electricity production by PWRs and VVERs in 2013–2024 (the value for 2024 was chosen equal to 2023 as the real value was not yet available), three projections of annual nuclear ¹⁴CH₄ emissions were computed and used as variable input parameter in three different simulations of global atmospheric Δ14CH₄ values (Fig. S4g). Posterior geologic emissions, biospheric turnover time, total CH₄ lifetime, carbon and hydrogen kinetic isotope effects, and carbon and hydrogen CH₄ isotopic signatures for respective sources in Fujita et al. (2025) were repeated by the values in 2012 over 2013–2024.

The simulated Δ14CH₄ values were compared to our annual mean Δ14CH₄ values at JFJ for 2019–2024 and previously reported measurements from samples (ice cores and/or firn air) of Greenland and Antarctica (Hmiel et al., 2020) and from atmospheric samples (Gonzalez Moguel et al., 2022; Lassey et al., 2007b; Levin et al., 1992; Quay et al., 1999; Sparrow et al., 2018; Townsend-Small et al., 2012; Wahlen et al., 1989). Note that the data in Hmiel et al. (2020) was used in Fujita et al. (2025) as observational constraints. Several other studies were found but not used here because of the large scatter in the reported data (Lowe et al., 1988; Manning et al., 1990) or difficulties in the unit conversion into current radiocarbon parameters (Ehhalt, 1974).

Δ14CH₄ values at JFJ between 2019 and 2024 were between 322 ‰ and 767 ‰ and showed a slightly increasing trend over the last six years (Fig. 2a, b). Some exceptionally high Δ14CH₄ values were measured (i.e., on 13 June and 25 July 2019; 13 May, 10 June and 24 June 2020; 12 May 2021) and were attributed to local and strong nuclear ¹⁴CH₄ releases. Rn values were lower than 5 Bq m⁻³ over 2019–2024 with generally lower values during winter months compared to summer months (Fig. 2c), which is consistent with a potentially stronger influence of the planetary boundary layer in summer. Annual mean CH₄ concentrations increased from 1930 ppb in 2019 to 1996 ppb in 2024 (Fig. 2d). Rn and CH₄ concentration data shown here represent the corresponding average values during the sampling intervals.

Reduced scatter in Δ14CH₄ values since the installation of the JASS in April 2023 is visible (Fig. 2a, b). Also, hourly Rn and CH₄ concentrations averaged over the JASS sampling periods show less amplitude variation after this date. For both parameters, corresponding values are closer to the monthly means measured in situ by the ICOS instruments (blue curves in Fig. 2c, d). The temporal stability of our Δ14CH₄ measurements from March 2022 onwards is evaluated using the regular Δ14CH₄ measurements of our two internal standard PAB bottles. The standard deviation of the Δ14CH₄ value for all 40 CH₄ measurements over 15 months for each PAB bottle (so about 80 CH₄ measurements in total) is 8 ‰ (Fig. S5a, b), which is lower than the instrumental uncertainty of a single Δ14CH₄ measurement (12 ‰) indicating the satisfactory long-term reproducibility of our Δ14CH₄ measurements.

For 2019–2024, mean annual atmospheric Δ14CH₄ values can be deduced from our fortnightly air sampling program in different ways (Table 1). Firstly, all Δ14CH₄ measurements clearly influenced by nuclear contamination were excluded. Choosing a threshold of 420 ‰ (Δ14CH4≤420‰, Figs. 2b and 3) 129 of the 137 measurements initially available are retained (94 %). Over the six years, a slightly increasing Δ14CH4≤420‰ tendency rising from 350 ± 19 ‰ in 2019 to 381 ± 13 ‰ in 2024 (i.e., by a rate of ∼+6 ‰ yr⁻¹) is observed (Table 1, Fig. 3).

Secondly, only Δ14CH₄ measurements whose corresponding Rn values are lower than 1.5 Bq m⁻³ STP (Δ14CH4Rn-filt) were retained; analyzing the probability density function of more than five years of Rn values at JFJ (November 2015 to December 2020), Conen and Zimmermann (2020) found that air masses with corresponding Rn values below 1.5 Bq m⁻³ STP belong mostly to the free troposphere (with 77 % confidence). Compared to the first set of mean annual values (n=129), mean annual Δ14CH₄ values are 1 ‰–7 ‰ lower and the total number of points used for these mean values is almost halved (n=71) (Table 1, Fig. 3). During the period of integrated sampling the observations removed with the Rn threshold are mainly from the summer.

The third processing step additionally corrects for the influence of NPPs by subtracting the nuclear Δ14CH₄ signal simulated with FLEXPART-COSMO from the Δ14CH₄ measurements at JFJ. Figure 3 shows for each Δ14CH4Rn-filt value (red point) the corresponding value corrected for the nuclear influence (Δ14CH4Rn-filt + Nuc-corr) (blue point). Overall, the mean nuclear Δ14CH₄ influence is 7 ± 9 ‰ (min =0 ‰, max =58‰, n=60).

Our annual mean Δ14CH₄ values at JFJ are similar to previous observations in the Northern Hemisphere from 2005–2020 (Fig. 4). Direct observations from Los Angeles, Canada and Alaska showed values of 340 ‰–350 ‰, comparable to our JFJ data of 338 ‰–366 ‰ for Δ14CH4Rn-filt + Nuc-corr from 2019–2023 (Fig. 4, Table 1). Firn air observations from Greenland for 2005–2013 were slightly higher, 350 ‰–380 ‰, but still consistent with our data.

A rather stable global background Δ14CH₄ is simulated for the period 2005–2024, supporting the consistency between the observations over this period. This stable period followed an increase from ∼ 140 ‰ in 1980 to ∼ 350 ‰ in 2005 (see Fig. 4 and Fujita et al., 2025), mainly driven by increasing nuclear ¹⁴CH₄ emissions since 1970 (Fig. S4h).

All the Northern Hemisphere measurement data is higher than the simulated Δ14CH₄ (Fig. 4), which reflects global atmospheric Δ14CH₄ accounting for contributions from both hemispheres (Fujita et al., 2025). Due to a lack of data, the current difference between the hemispheres is presently not well-known, but previous data indicate an excess Δ14CH₄ in the Northern Hemisphere, which is consistent with stronger nuclear power plant emissions there (Fig. 4). The observational targets from Fujita et al. (2025) were constructed to allow for this hemispheric difference and for uncertainty due to lack of data. The mean offset between the observations in Greenland for 2005–2013 and the simulated global Δ14CH₄ value is 22 ± 9 ‰, comparable to the offset between our JFJ observations for 2019–2024 and the global simulation of 0 ‰ to 33 ‰.

The three simulated Δ14CH₄ trends from 2013 to 2024 (blue lines in Fig. 4) based on three different emission factors Φ (Sect. 2.5 and Fig. S4g) show a decrease followed by a stabilization (for Φ=243 GBq GWa⁻¹), or a slight increase (for Φ ≥250 GBq GWa⁻¹). The initial decrease was caused by a decrease in nuclear ¹⁴CH₄ emissions following the Fukushima accident in 2011 (Fig. S4h). Afterwards, the stabilization or slight increase has arisen from nuclear ¹⁴CH₄ emissions that have increased again in particular after 2017 (Fig. S4h). Our measurement data show a slight positive trend that is reduced after accounting for regional influences from nuclear power plant emissions (Fig. 4, Sect. 3.1). The simulations seem to be more consistent with this slight positive trend from our measurement data using the higher emission factors than the lowest emission factor, where a slight decrease in Δ14CH₄ is simulated.

We can also compare with observations at JFJ in 1988–1991 by Levin et al. (1992) (Fig. 4). A large spread in individual measurements of 210 ‰–255 ‰ was found at that time. We also found a large scatter in samples collected in the morning before the installation of integrated nighttime sampling (Fig. 2, Sect. 3.1). The number of operating PWRs worldwide passed from about 240 in 1991 to about 310 in 2024 (Laemmel and Szidat, 2025), suggesting that the influence of nuclear ¹⁴CH₄ emissions has increased. Overall, the increase from ∼ 230 ‰ in the 1980s to 360 ‰ in the early 2020s is consistent with other data and with the simulated change (Fig. 4).

Between 2019 and 2024, Δ14CO₂ values measured at JFJ ranged from +3 ‰ to -17 ‰, following a decreasing trend over these six years; this trend is mostly due to the emissions of ¹⁴C-free fossil fuel CO₂ which depletes the global atmospheric Δ14CO₂ signal (Fig. 5a). Annual mean CO₂ concentrations increased from 411.9 ppm in 2019 to 424.7 ppm in 2024 (Fig. 5b). The standard deviation of the Δ14CO₂ value for all 40 CO₂ measurements over 15 months for each PAB bottle (so about 80 CO₂ measurements in total) is 1.5 ‰ (Fig. S5c, d), which is lower than the instrumental uncertainty of a single Δ14CO₂ measurement (2 ‰) indicating the satisfactory long-term reproducibility of our Δ14CO₂ measurements.

We compare our annual mean Δ14CO₂ and individual Δ14CO₂ observations with measurements from ICOS at JFJ (Emmenegger et al., 2025a) in Figs. 5a and 6 and in Table 2. The average trend in Δ14CO₂ is similar in both datasets: -2.1 ± 1.9 ‰ yr⁻¹ for our data and -2.8 ± 1.9 ‰ yr⁻¹ for ICOS data. Annual mean values differ by less than 1.5 ‰ except for 2019, when our annual mean Δ14CO₂ value was 3.6 ‰ lower than the ICOS annual mean. Larger individual differences are visible especially in the first half of 2019 (Fig. 5a). Both annual mean ICOS Δ14CO₂ values for 2019 and 2020 at JFJ are consistent with the equivalent means at the MHD station (Table 2), indicating that the offset in 2019 between our values and the ICOS ones was probably due to a small fossil contamination in our early Δ14CO₂ measurements that was remediated during the year.

After the JASS installation in 2023, a more detailed comparison of Δ14CO₂ with ICOS became possible, as we use the same 14 d sampling as the ICOS integrated sodium hydroxide solution sampling (Fig. 6). Over about 20 months the mean difference was -0.04 ± 1.74 ‰ (n=20) with (insignificantly) lower values in our data. A few large differences up to ± 4 ‰ were observed. We emphasize that even though the fortnightly periods are the same, our JASS sampler integrates only nighttime hours whereas the ICOS sampler integrates all day; moreover, the air inlet of our system is situated about 3 m higher than the one from the ICOS sampler. Therefore, the measurements are not conducted on the exact same air. However, the insignificant mean difference suggests that the different sampling conditions affect the measured Δ14CO₂ only marginally.

Similar to the correction we made using the simulated ¹⁴CH₄ nuclear influence on Δ14CH₄ measurements, we also make a correction using the simulated ¹⁴CO₂ nuclear influence on Δ14CO₂ measurements (Fig. 7). The mean nuclear Δ14CO₂ influence is 0.2 ± 0.4 ‰ (min =0‰, max =2.3‰). The overall effect of nuclear ¹⁴C emissions on atmospheric Δ14CO₂ is less than for Δ14CH₄, i.e., 0.2 ± 0.4 ‰ compared to 7 ± 9 ‰, respectively. Moreover, nuclear ¹⁴CO₂ emissions (from BWRs and NFRP) are known to be less sporadic than ¹⁴CH₄ emissions from PWR reactors (Stenström et al., 1995b), so our atmospheric model based on monthly nuclear ¹⁴C releases simulates better nuclear ¹⁴CO₂ contributions than ¹⁴CH₄ contributions.