Introduction

Methane mole fractions (CH4) in tropospheric air continue to increase but the main reason behind the acceleration in the methane growth rate in the last two decades remains still uncertain1. Different studies pinpoint anthropogenic emissions from the fossil fuel, waste and agricultural sectors as the main contributor to the global methane budget2,3, while others identify the main cause of the latest CH4 increase in the extension of tropical wetlands as a positive feedback to a warmer climate4,5,6. Addressing the main drivers of the atmospheric CH4 growth, and whether they are of anthropogenic origin rather than natural feedback, is essential to direct our efforts in reducing emissions. Methane relatively short lifetime in atmosphere (~10 years) compared to carbon dioxide makes it a good target for emission mitigation programmes, such as the Global Methane Pledge delivered in 2021 that aims at a reduction of at least 30% of emissions compared to 2020 by 2030. According to a 2021 report of the UN Enviroment Program – UNEP, a reduction of 45% of anthropogenic CH4 emissions by 2030 could prevent a global temperature increase of 0.3 °C by 20407. In this context, atmospheric stations that provide continuous and accurate CH4 mole fraction values are essential as they enable a continual monitoring of the CH4 burden in the atmosphere. In particular, remote and mountain stations, being mostly above the PBL (planetary boundary layer) and measuring lower free tropospheric air, are used for estimation of CH4 atmospheric background and global trends of atmospheric CH4. However, mole fractions enhancements due to sporadic local emissions or interception of air masses from the PBL may occur, which need to be filtered out by applying a background selection procedure specific for each site8,9.

There are several CH4 modelling studies including an ensemble of mountain sites for a top-down estimate of the regional CH4 budget and identification of CH4 emission areas in Europe10,11,12,13. However, the analysis of the mole fraction record at a single mountain site, combined with a description of air transport processes, can provide further insights into regional CH4 source areas. Few studies include a characterisation of a multiple-years CH4 mole fraction time series measured at a European mountain site and identify CH4 source regions within the station footprint14,15. One example is Necki et al.14 that covers six years of CH4 mole fractions at the mountain station Kasprowy Wierch, 1987 metres a.m.s.l., and attributes enhanced level of CH4 to local swamps, pit deposits and landfills. The atmospheric monitoring station at Plateau Rosa, situated in the north-western Italian Alps near Mt. Cervino, is part of the WMO/GAW (World Meteorological Organisation/Global Atmospheric Watch, Identification Code: PRS) programme since 1989 and part of the ICOS (Integrated Carbon Observation System) framework since 2021. The station, located at 3480 metres a.m.s.l., is the second highest greenhouse gases observatory in Europe after Jungfraujoch, Switzerland, and it is therefore a strategic site for atmospheric measurements of lower tropospheric background air. Given its altitude, PRS mole fractions measurements are influenced by long-range air masses that cross several European countries before reaching the station, making PRS a potential viewpoint for CH4 emission areas across Europe. The PRS station hosts a cavity ring down spectrometer (Picarro G2301) that provides continuous carbon dioxide (CO2) and methane (CH4) mole fractions since 2018 according to the highest quality standards required by the GAW program. A weather station enables continuous measurements of key meteorological parameters such as temperature, relative humidity, pressure, and wind speed and direction (via a 2D ultrasonic sensor).

Although mountain sites have a key role in greenhouse gases monitoring, tracing back emission areas that affect the mole fractions record at the receptor site by using atmospheric transport modelling remains a hard challenge for alpine stations, as the model might not capture entirely the air masses dynamics in very complex high-alpine terrains. In this study we provide an analysis of the seven-years timeseries of CH4 mole fraction from PRS since 2018, and a procedure for background selection based on a statistical approach. We use part of the mole fractions record (2021-2024, since PRS is being part of ICOS) for validation of the FLEXPART-COSMO atmospheric transport model11,16 coupled to the high-resolution (1 km × 1 km) output of the numerical weather prediction model COSMO. We use this modelling framework both to simulate background values at PRS, to be compared with the statistically selected background, and to produce concentration footprints, which represent the source area responsible for the mole fraction increments at the station, demonstrating the spatial representativeness of PRS. By convoluting footprints with both the global bottom-up inventories EDGAR v.202417 and the Copernicus Atmosphere Monitoring Service (CAMS) top-down flux estimates (2025)18, we simulate regional CH4 contributions and compare them to the observed mole fraction increments. We finally identify events of consistently enhanced mole fractions over background that are associated with pollution events at regional scale, i.e. covering part of the European continent, and we assess the ability of the modelling framework to capture such events.

Results and discussion

The whole timeseries of monthly CH4 mole fractions measured at PRS is shown in Fig. 1, together with monthly CH4 mole fractions recorded at the ICOS atmospheric station of Jungfraujoch (ICOS ID: JFJ, 3580 m altitude)19, Switzerland, and Zugspitze (ICOS ID: ZSF, 2666 m altitude)20, on the border between Germany and Austria, included to demonstrate the comparability of observations in the Alpine area. The CH4 baseline for Mace Head21, on the west coast of Ireland, is also included, as representative of the marine background in the northern hemisphere. In the period between 2021 and 2024 all stations recorded an overall increasing trend in mole fractions of 11 + /− 1 (2σ) ppb year−1.

Fig. 1: Monthly trend of CH4 mole fractions observed at Alpine stations and Mace Head.
Fig. 1: Monthly trend of CH4 mole fractions observed at Alpine stations and Mace Head.
Full size image

a Averaged monthly values of measured CH4 mole fractions, including enhancements, at Plateau Rosa, PRS (in black), Zugspitze, ZSF (in red), Jungfraujoch, JFJ (in orange) and monthly baseline for Mace Head, MHD (in blue). ZSF and JFJ mole fractions are taken from ICOS, MHD mole fractions are retrieved from AGAGE (Advanced Global Atmospheric Gases Experiment); b Seasonal cycle for PRS (grey diamonds) and MHD (blue circles). Error bars represent the σ of monthly values.

While PRS and JFJ monthly CH4 mole fractions are comparable due to the stations proximity, monthly values at ZSF are higher than at PRS, as a result of more frequent influence of air masses from CH₄-enriched source regions in Northern and Eastern Central Europe, whereas PRS is more often influenced by low-emission air masses in the Mediterranean region12,22. Monthly fluctuations at PRS, and so at JFJ and ZSF, are mostly due to changes in transport and different emission contributions, whereas average seasonality is driven by the reaction of CH4 with OH radicals, which is enhanced in summer causing an overall decrease in the atmospheric CH4 in June, July and August (Fig. 1b). This is confirmed by the seasonality of the CH4 baseline measured at MHD (blue in Fig. 1b), which is mainly linked to the reaction with OH radicals. Higher mole fractions in fall and winter at MHD can be explained by the CH4 latitudinal gradient, with higher mole fractions at higher northern latitudes caused by boreal wetlands fluxes23,24,25.

Background selection

Although the PRS station for most of the year is not influenced by local anthropogenic emissions, some sharp fluctuations due to local sources may occur. Therefore, the hourly CH4 background values were selected following a modified version of the BaDS (Background Data Selection) procedure in Apadula et al.8, which assumes that the background must have very little variability within the hourly averages and between two consecutive mean values. This selection process retains hourly averages with a standard deviation lower than 9.6 ppb and that differ from the next hour by less than 4.35 ppb. Retained values are used to compute a trend curve (21 days moving mean in Fig. 2), and the final background is given by hourly mole fraction measurements that differ from this trend by less than 22 ppb (see Supplementary Methods for the whole selection procedure). A similar statistical method is adopted for the Mauna Loa monitoring station26 and in other atmospheric stations that are part of ICOS Italy9, although with different cut-off threshold values, which are specific for each measurement site. Here we compare the statistically selected background values to the simulated background, Obs_Background and Sim_Background respectively in Fig. 2. Simulated background is extracted from CAMS CH4 mole fractions at the final location of FLEXPART model particles after 10 days of backward integration, which is comparable to boundary conditions at the interface of the European model domain (see Section “Methods”).

Fig. 2: CH4 mole fraction measurements, observation-based and simulated background at PRS.
Fig. 2: CH4 mole fraction measurements, observation-based and simulated background at PRS.
Full size image

CH4 mole fraction measurements at PRS (in grey), simulated background (in red), observation-based background (in blue), 21 days moving mean (in black) of retained values following the background selection procedure (see Supplementary Methods).

There are periods when the observation-based background is not captured by the simulated background, with a mean bias of 17 ppb. The observed values include some regional contribution even after filtering for background selection, which can be seen during April and May (see Fig. 3), when the observations-based background is systematically higher. In fact, the statistical filtering approach eliminates mainly sudden mole fraction peaks, retaining as background 73% of hourly means (see Supplementary Methods). In fall and winter, the observation-based background is lower than the simulated one. November 2023 shows the highest discrepancy, when there was an episode with air flow from the Northern Atlantic regions, indicating that CAMS mole fractions might be biased high in these areas.

Fig. 3: Monthly means of observed and simulated background.
Fig. 3: Monthly means of observed and simulated background.
Full size image

Monthly means of CH4 observation-based background (in blue) and simulated background (in red). Shaded areas are 2σ of the mean. Simulations for 2024 cover only the period from January until June.

Correlation between simulations and observations

The simulated mole fraction enhancements were calculated by multiplying FLEXPART footprints by CAMS and EDGAR fluxes. Besides having two different spatial resolutions, with CAMS at a coarser scale, CAMS fluxes are top-down emission estimates that include both anthropogenic and natural emissions, whereas EDGAR bottom-up emission inventories report only anthropogenic sources. The inclusion of natural emissions, given that they account for approximately 35% of global totals according to top-down estimates27, makes the use of CAMS fluxes more meaningful for comparison between simulations and observations. However, EDGAR fluxes are provided by source sectors, giving insights into the main anthropogenic source contributing to mole fraction enhancements at PRS (see Section “Individual pollution events”). Here we report the linear regression of simulated versus observed mole fractions using both fluxes as a means to evaluate the model uncertainty (Fig. 4). Simulated mole fractions are obtained by adding simulated increments to the simulated background rather than to the moving mean, to preserve the CH4 background short term variability that would be smoothed out by a statistical averaging. The linear regression between simulated and observed mole fraction values is shown in Fig. 4, where values in grey are computed by using the full simulated background series (Tot), and values in black are simulated using a subset of the simulated background (67%) that fits to the observation-based one (Fit to obs_bg), i.e. that falls within ±1σ of the moving mean (black line in Fig. 2). This comparison enables assessment of the uncertainty introduced by the modelled background variability. Using the Fit to obs_bg series improves the correlation coefficient from 0.7 to 0.8 and decreases the RMSE of 3 ppb, regardless of what flux dataset was used. The improved correlation coefficient and RMSE are similar to the one calculated in Bergamaschi et al.13 comparing model simulations to CH4 observations at Jungfraujoch. Overall, simulations using both EDGAR and CAMS fluxes tend to overestimate the measured values, with a mean residual of 13 ppb for CAMS and 9 ppb for EDGAR when using the Tot simulated background series, and 11 ppb and 7 ppb when using the Fit to obs_bg. However, the use of EDGAR fluxes leads to smaller discrepancies (RMSE ranging between 17 and 21 ppb for EDGAR vs 20 and 23 ppb for CAMS fluxes), suggesting that the higher spatial resolution of surface fluxes plays an important role in the modelling framework.

Fig. 4: Linear regressions of simulated mole fraction values versus measurements.
Fig. 4: Linear regressions of simulated mole fraction values versus measurements.
Full size image

Simulated values are computed by using the whole time series of the simulated background (in grey) and the simulated background that best fits to the observed one (in black); dotted red lines are 1:1 ratio; the ρ value is the Pearson correlation coefficient.

By using CAMS fluxes, we verify whether the model FLEXPART-COSMO captures the observed diurnal pattern by plotting the simulated CH4 excess and the observed CH4 mole fraction increments. The observed increment has been derived from the observations minus the simulated background, which retains the background short-term fluctuations. Figure 5 shows that in winter months (December, January, February) the CH4 diurnal cycle is characterised by very small diurnal variation, indicating that most of the air measured at PRS during winter is free tropospheric air. In summer, particularly in July, August and September, the diurnal cycle is more pronounced due to the advection of CH4 enriched air masses with recent PBL contact during daytime, which is enhanced due to strong solar insolation28. Such diurnal variation in fluxes is somewhat simulated in August and September, but not in July, suggesting that the upwards venting mechanism driven by the complex topography is not sufficiently captured at the present horizontal resolution of FLEXPART-COSMO, a similar result to a previous study for CO2 mole fractions at Jungfraujoch, which, however, relied on coarser meteorology29. Overall, simulated mole fraction enhancements are larger than the observed ones throughout the year, though to a lesser extent during the coldest months (i.e., December to April), which might indicate an overestimation of CH4 fluxes transported through advection caused by the low resolution of CAMS fluxes that smears high emissions into the Alpine area and Po Valley (see Supplementary Fig. 1).

Fig. 5: Diurnal cycle of simulated and observed CH4 increments.
Fig. 5: Diurnal cycle of simulated and observed CH4 increments.
Full size image

Diurnal cycle of CH4 mole fractions enhancements simulated using CAMS fluxes (in red) and diurnal cycle of observed CH4 mole fractions excess (in blue), derived from the observations minus the simulated background. Shaded area represents 2σ of the mean.

Individual pollution events

Individual events were selected based on the measured mole fraction enhancement over the moving mean and the associate wind speed. By selecting wind speeds higher than 3 m/s to avoid very local pollution peaks and enhancements larger than 48 ppb (4 σ of the mean difference between the observed mole fraction and the moving mean) we found 30 pollution events, each of them spanning more than 6 h (Table 1). Based on the FLEXPART footprints, we identified the provenance of air masses at peak time. All the FLEXPART footprints are included in Supplementary Fig. 2. We then compared the simulated mole fractions using CAMS surface fluxes to the observed mole fractions. Table 1 includes the Pearson correlation coefficients and root mean square errors of the comparison between simulated mole fractions and observed mole fractions for each event. The mole fraction event whose pattern was not captured by the modelling framework (correlation coefficient near to zero) is event 12, with footprint constrained on Switzerland, France, Belgium, the Netherlands and the UK. Most of the events are associated with air provenance from Northern Italy, including the north-western Italian Alps and the Po Valley, but the best simulated mole fraction events, with a Pearson coefficient higher than 0.8 and RMSE lower than 25 ppb, do not include Northern Italy in their footprint (i.e. events 11, 18, 19, 23, 25, 30), except for event 20, indicating that this region represents a large source of uncertainty when trying to simulate CH4 mole fractions at PRS. Event 28, with footprint constrained on Spain and South of France, shows the highest discrepancy between simulated and observed mole fractions (RMSE higher than 40 ppb).

Table 1 List of events including dates, countries covered by the footprint at peak time, RMSE and Pearson correlation coefficients
Full size table

In Fig. 6 we compare the observed mole fractions peaks, i.e. observations minus simulated background, with the simulated ones using EDGAR inventories, which are available by anthropogenic sectors, showing the source sectors’ contribution in terms of mole fractions added for each event’s peak. The highest peak (125 ppb) was simulated when the station footprint covered mainly the Po Valley (see footprint for event 16 in Supplementary Fig. 2), in Northern Italy, with the highest contribution from the agricultural and waste sector. This region is a hotspot for methane emissions from agricultural activities like enteric fermentation from livestock and rice cultivation. In addition, it is an important hub for natural gas distribution in Italy with a high potential for leakages from pipelines, storage facilities, and distribution networks. The fuel exploitation sector (blue in Fig. 6), covering emissions from production, transformation, and refining of fuels, also plays a big role, especially for events 11 and 24, which are both influenced by emissions from Switzerland, France, Germany and northern European countries such as the Netherlands, Belgium and the UK. The gas hub in the Netherlands might largely contribute to these pollution events. Also during event 30, with footprint including Czechia and Poland, the fuel exploitation sector looks more important relative to other events, which might be related to emissions from coal mines in the Upper Silesian Basin30. During event 24 that occurred on 2nd December 2023 and with air provenance from Northern Italy, Switzerland, France, Germany, the Netherlands, Belgium and the UK, the buildings sector, referring to emissions from on-site fuel combustion for buildings heating, also contributed largely to the CH4 enhancement. The buildings sector contribution is less pronounced for event 25, which occurred in the same period, but it does not include Northern Italy and South Germany in its footprint (see Supplementary Fig. 2). Note that some simulated peaks underestimate observed mole fraction enhancements (black dots in Fig. 6) by a fraction that cannot be solely explained by the model error (17 to 21 ppb). Events 2, 3, 16, 18 and 28 have the largest mismatch, with footprints constrained mainly on Northern Italy for event 2, 3, 16, Switzerland, Belgium and the UK for event 18 and South of France for event 28, indicating that methane fluxes from these regions might be highly underestimated.

Fig. 6: Source sectors’ contribution to the mole fraction peak for each event in Table 1.
Fig. 6: Source sectors’ contribution to the mole fraction peak for each event in Table 1.
Full size image

The source sector categories are taken from the EDGAR inventory. Black dots represent the measured increments over the simulated background. Error bars on simulated peaks are the RMSE of simulated vs measured mole fractions using EDGAR (20.8 ppb, see Fig. 4), error bars on the measured increments include both the σ of hourly measurements and the σ of simulated background for that hour.

Conclusions

Mountain sites offer an ideal spot for long-term monitoring of greenhouse gases, thanks to their altitude and distance from local anthropogenic emissions that enable measurements of well-mixed troposphere. At Plateau Rosa (PRS), continuous CH₄ mole fraction measurements since 2018 reveal a persistent increase in atmospheric methane. Using the high-resolution FLEXPART-COSMO model, we simulated CH₄ background values from 2021 to June 2024 to evaluate model performance, and we produced the station footprints, identifying the main European source regions affecting mole fractions at PRS. Regional CH4 contributions were simulated by coupling footprints with both CAMS and EDGAR fluxes and added to a simulated background for comparison with the observations.

While the modelling framework achieves a Pearson correlation of 0.7–0.8 and RMSE of 17–21 ppb, systematic biases remain. Comparison between the simulated background and the observation-based background showed that the model does not always capture background variations at PRS, with a mean bias of 17 ppb. Simulated mole fractions capture only partially the advection of air masses that are charged with CH4 emissions within the boundary layer, typical of summer months, stressing the need for an improved representation of boundary-layer advection in complex terrain, a similar conclusion to other modelling studies focused on the Alpine area.

Mole fraction peaks of most pollution events that we identified were underestimated by the modelling framework. A large part of the emissions during pollution events were missed, particularly when the footprint covered Northern Italy, including the Po Valley, the region that influences most mole fraction enhancements at PRS, suggesting that finer-scale CH4 emission inventories combined with reliable information on temporal variability are needed to reduce discrepancies between simulated and observed mole fractions at PRS.

Sectoral analysis using the EDGAR inventory indicates that primarily emissions from the agriculture and waste sectors affect the pollution events that have been detected, with fossil fuel contributions increasing when footprints extend to Central and Northern European countries, such as France, Germany, Austria, Belgium, the Netherlands and the UK. Future integration of isotopic measurements at PRS would help to further constrain source attribution.

Although footprints analysis improved our understanding on source regions over the European continent affecting the CH4 record at PRS, the small number of pollution events that have been detected throughout four years (30) confirms PRS as a background station, whose valuable record can be used both for analysis of CH4 global trends and as baseline for CH4 studies in Europe. Finally, discrepancies between the simulated and observed mole fractions should be further investigated, as modelling of greenhouse gas transport in complex terrain remains a challenge but is crucial for inverse modelling of greenhouse gas exchange in many world areas.

Methods

Mole fraction measurements

The atmospheric observatory of Plateau Rosa is part of the Testa Grigia laboratory managed by CNR (Italian National Research Council) on top of a large mountain plateau. The observatory is placed between a cable car and a refuge, operating during the day and for approximately eight months a year. The station is equipped with an electric heating system to avoid any use of fossil fuel and limit the influence of fossil emissions on the measurements. Unfortunately, works for construction of the new cableway Matterhorn Alpine Crossing since April 2019 have been produced several local pollution events, especially during central hours of the day and working days, interfering particularly on CO2 measurements. A data quality control is performed weekly to filter out these events and exclude them from the collected data series. CH4 mole fractions have been measured continuously at PRS since 2008 with a N.I.R.A. GC-FID (Gas Chromatography with Flame Ionisation Detector) analyser (data not included in this study because they have a precision of 30 ppb), and since 2018 with a Picarro cavity ring down spectrometer G2301. The PRS station is not manned and a system including a rotary valve (VICI Valco) has been installed to switch the inlet port automatically and control the measurements sequence. Air is drawn to the instrument from an air intake located at 10 m from the ground using a pump (KNF N820), enabling a flow of ca 120 cm3 min−1. Mole fractions are calibrated every 30 days against reference gases provided by the ICOS central analytical laboratory (ICOS-CAL) and are reported against the NOAA X2004A scale. Two working standards are measured for routine assessment of instrumental drift, every 20 h and 30 days respectively. The measurements reproducibility achieved is 0.01 ppm for CO2 and 0.2 ppb for CH4. A nafion dryer (model PermaPure MD-070) has been installed on 12th April 2024 to reduce humidity levels of outside air to values below 0.1 %. Before then, a water correction test was performed twice a year to remove the water interference on measurements following the ICOS procedure31. Mole fraction data are provided with a frequency of 2.8 s; the station PI performs routinely the data quality control, selecting data based on instrumentation parameters, such as Picarro cavity temperature and pressure, and flagging sudden peaks as non-background data, while the calibration process is all centralised and performed by ICOS. In this study we use hourly calibrated means for computing daily, yearly trends and for back trajectory analysis.

Simulated concentration footprints and model-based background estimation

FLEXPART-COSMO is a Lagrangian Particle Dispersion Model derived from the main version of FLEXPART32 for the use with output from the numerical weather prediction model COSMO11. For the current study, FLEXPART-COSMO simulations were based on 1 km COSMO operational analysis fields generated by MeteoSwiss. The setup is for 3-hourly releases of 50,000 model particles and integrating 4 days backward in time or until a model particle leaves the limited model domain. Afterwards, particle integration was continued up to a total of 10 days, using the standard version of FLEXPART (version 9.2.2-empa) and input from the operational runs of the Integrated Forecast System (IFS) of the European Centre for Medium Range Weather Forecast (ECMWF). More details on the model setup can be found in Katharopoulos et al. (2023)33, who employed it for sites in Switzerland. For PRS particles were released at a height of 3480 m a.s.l. For each 3-hourly release a concentration footprint (sampling height 40 m above ground) was calculated, which provides the source–receptor relationship between potential emission sources in the model domain and the enhancement in mole fraction at the measurement station. Concentration footprints were generated at two horizontal resolutions: 0.02° ×  0.015° for the Alpine domain and 0.16° × 0.12° for most of Europe.

Regional enhancements of CH4 mole fractions were simulated by convoluting the concentration footprints (units s m3 kg−1) with surface fluxes (units kg m−2 s−1) as provided by the inventory. We used the latest release of the EDGAR_2024_GHG, which are available by month and sector, and the latest release of CAMS global-inversion-optimised greenhouse gases fluxes (2025) available by month. The latter use surface air-samples and satellite observations as inputs for an inverse estimation of surface fluxes. Both surface fluxes have different spatial resolutions than FLEXPART footprints, 0.1° ×  0.1° and 1° × 1° for EDGAR and CAMS, respectively. We used a first order conservative regridding and bilinear interpolation for EDGAR and CAMS, respectively, to match the surface fluxes to the concentration footprints. The regridding method has been chosen based on the best match between simulated values and observations.

The total simulated CH4 mole fraction is the sum of the regional enhancements and a background mole fraction, similar to boundary conditions in a Eulerian model. The background mole fractions were constructed by sampling the CAMS CH4 mole fraction field at the location and time of each FLEXPART model particle’s final position. Averaging these mole fractions over all particles released in the same 3-hourly time window, results in the simulated background mole fraction.