In situ observations of greenhouse gases over Europe during the CoMet 1.0 campaign aboard the HALO aircraft

The intensive measurement campaign CoMet 1.0 (Carbon dioxide and Methane mission) took place during May and June 2018, with a focus on greenhouse gases over Europe. CoMet 1.0 aimed at characterising the distribution of CH4 and CO2 over significant regional sources with the use of a fleet of research aircraft, as well as validating remote sensing measurements from state-of-the-art instrumentation installed on-board against a set of independent in-situ observations. Here we present the results 5 of over 55 hours of accurate and precise in situ measurements of CO2, CH4 and CO mixing ratios made during CoMet 1.0 flights with a cavity ring-down spectrometer aboard the German research aircraft HALO, together with results from analyses of 96 discrete air samples collected aboard the same platform. A careful in-flight calibration strategy together with post-flight quality assessment made it possible to determine both the single measurement precision as well as biases against respective WMO scales. We compare the result of greenhouse gas observations against two of the available global modelling systems, 10 namely Jena CarboScope and CAMS (Copernicus Atmosphere Monitoring Service). We find overall good agreement between the global models and the observed mixing ratios in the free-tropospheric range, characterised by very low bias values for the CAMS CH4 and the CarboScope CO2 products, with a mean free tropospheric offset of 0 (14) ppb and 0.8 (1.3) ppm respectively, with the quoted number giving the standard uncertainty in the final digits for the numerical value. Higher bias is observed for CAMS CO2 (equal to 3.7 (1.5) ppm), and for CO the model-observation mismatch is variable with height 15 (with offset equal to -1.0 (8.8) ppb). We also present laboratory analyses of air samples collected throughout the flights, which include information on the isotopic composition of CH4, and we demonstrate the potential of simultaneously measuring δC−CH4 and δH−CH4 from air to determine the sources of enhanced methane signals using even a limited amount of discrete samples. Using flasks collected during two flights over the Upper Silesian Coal Basin (USCB, southern Poland), one of the strongest methane-emitting regions in the European Union, we were able to use the Miller-Tans approach to derive the 20 isotopic signature of the measured source, with values of δH equal to -224.7 (6.6) ‰ and δC to -50.9 (1.1) ‰, giving significantly lower δH values compared to previous studies in the area. 1 https://doi.org/10.5194/amt-2020-287 Preprint. Discussion started: 17 August 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
Increased mixing ratios of atmospheric greenhouse gases (GHGs) are recognised as the primary cause of the warming observed in the climate system over the past 70 years. Of these, the most important are carbon dioxide (CO 2 ) and methane (CH 4 ), 25 respectively responsible for approximately 56 % and 32 % of the globally-averaged increase in radiative forcing caused by greenhouse gases, as compared to the pre-industrial period (IPCC et al., 2013). Further increases in the atmospheric burden Compound Precision Uncertainty of scale link Unit  A significant drift in CO mixing ratios was observed over the period between sample collection and the subsequent measurement in the GasLab. The resulting enhancement in the mixing ratio was not easily correctable, therefore the results had to 170 be discarded. Careful quality control and additional tests did not show any signs of other gases being affected.

Flight patterns
Depending on the scientific goals set out before each research flight, different flight patterns were executed in order to obtain the most valuable data. The main strategies adopted for the CoMet 1.0 mission were: i) long-range gradient observations, designed to maximise the amount of observations for active lidar measurements with CHARM-F operated on HALO, ii) vertical profiles, 175 aimed mainly at the intercomparison between the lidar and in situ observations and iii) low-altitude legs, performed to assess the enhancements of CO 2 and CH 4 downwind of their sources (plume-chasing).

Large-scale variability in upper troposphere and lower stratosphere
Due to the constraints related to using other instruments (the active lidar), a significant amount of flight time was spent flying level at altitudes higher than 4 km, in order to emulate a flight path similar to that of a satellite system. Typical variability of in 180 situ greenhouse gas mixing ratios was low in these cases and is usually considered to be caused by intermixing of air masses coming from different regional source areas. In situ data obtained in this manner are well-suited for validation of Global Chemistry Models. As an example, in sec. 3.2 we compared JIG observations against well established modelling products: CAMS greenhouse gas forecasts. A detailed model description is given in sec. 2.4.

185
Multiple vertical profiles of the atmosphere were carried out during the campaign in order to establish the connection between column-integrated remote sensing and in situ measurements, thus also linking remote sensing observations to common WMO scales for greenhouse gases. The typical strategy consisted of i) a high-altitude overflight over a selected target, ii) descent in the form of a spiral to the lowest possible altitude above the target, iii) subsequent ascent back to high altitude, usually flown along the shortest path in the direction of the next planned way-point.

190
Usually two or three vertical profiles were executed during a given flight, depending on the availability of points of interest and airspace accessibility. Wherever possible, profiles were executed above a) ICOS stations, b) TCCON stations (Total Carbon Column Observing Network), c) Sentinel 5P or GOSAT overpass locations. Flasks were also collected during vertical soundings, at levels distributed between the minimum and maximum altitude, typically consisting of six samples per profile (in some cases reduced to four).

195
Measurements of vertical profiles are also of high interest for model validation exercises, as the availability of highly precise data on greenhouse gases over Europe is currently still limited. In the scope of the current study, we have assessed the performance of two well-established modelling products (CAMS and Jena CarboScope, see sec. 2.4) against CoMet 1.0 in situ observations. Additional to the in situ measurements, flasks were also collected to gather information about additional compounds and the stable isotopic composition of CH 4 . For the cases where sufficient data were available, we have applied the method of Miller and Tans (2003), a variation of a classic Keeling model (Keeling, 1958), in order to obtain the isotopic mean source signature (δ 0 ), expressed using relative delta notation. The method assumes a two-factor mixing of background air and methane-enhanced 210 plume: where δ obs is the measured isotopic signature, δ bg is the background signature, χ obs is the mixing ratio of the analysed compound, and χ bg is the background mixing ratio. Here, similar to the Keeling approach, information on δ 0 can be gleaned from the application of linear regression, however the source signature is calculated from the slope, rather than intercept, of the 215 linear fit formula. Following the study by Cantrell (2008), we have applied a Williamson-York regression, which allows one to take into account uncorrelated errors in both the X-and Y-axes of the data.
The Miller-Tans method relies on the appropriate assignment of the background signature (i.e. of the atmospheric air outside of the plume). Long term data available from atmospheric observations show that the isotopic composition of methane in the atmosphere is variable (Röckmann et al., 2016;Nisbet et al., 2019) in both space and time. In order to estimate the background 220 signature, here we have used measurements from air samples collected in the immediate vicinity of the target plume, either from i) the upwind air masses when possible or ii) air outside of the main plume when not.

Models
As part of the Copernicus Atmosphere Monitoring Service (CAMS), the European Centre for Medium-Range Weather Forecasts (ECMWF) performs greenhouse gas simulations based on its Integrated Forecasting System (IFS) and provides opera-225 tional global forecasting products focused on greenhouse gases. In this work, we have used the five-day high resolution greenhouse gas forecast product from CAMS (experiment ID: gqpe, downloaded in April 2020; see Agusti-Panareda et al., 2017; Agustí-Panareda et al., 2019) in order to validate the model using our observations. Further in the text, we will refer to these data as CAMS for simplicity. Satellite data were used for initialization of the forecast, namely TANSO-GOSAT for CO 2 and CH 4 and additionally MetOp-IASI for CH 4 (Massart et al., 2014(Massart et al., , 2016. For CO, CAMS operational analysis (Inness et al., octahedral grid (equivalent to approximately 9-km horizontal resolution) was interpolated to 0.125°x 0.125°. The frequency of the analysed CAMS data was 3-hourly, and vertical resolution was the regular L137 ECMWF configuration.
Additionally, CO 2 data were also compared to the Jena CarboScope product (version s04oc_v4.3, Rödenbeck, 2005), further referred to as CarboScope. While the resolution of the driving CarboScope model output fields is much lower in this case (4°x 235 5°horizontal), the system benefits from using the fluxes of CO 2 optimized using a Bayesian inversion framework. A detailed description of the modelling system is given in Rödenbeck et al. (2003) and Rödenbeck (2005). The transport model TM3, which is used by CarboScope, is described in Heimann and Körner (2003). and CO, respectively.
The comparison between flask and in situ measurements is available for all except one flight (no. 5). From the total 96 250 samples collected and analysed, 84 had simultaneous in situ measurements available from JIG that could be used for a bias assessment. As shown in Fig. 3, the average bias for flights 1-7 was equal to -0.131 (30) ppm for CO 2 and -2.93 (32) ppb for CH 4 , where number in brackets represent standard uncertainty in the final digits quoted for the numerical value.
Larger spread when independent measurements are considered (Fig. 2) stems mainly from the imperfect match between the temporal coordinates of the two instruments, which can be considered random and does not cause systematical shift. After the 255 malfunction, i.e. for flights no. 8 and 9, these mean offsets were equal to 0.127 (68) ppm and -0.64 (91) ppb for the respective gases. While the difference of values caused by the broken mounting is statistically significant, the difference is still close to the WMO compatibility goal.

Large Scale Variability
Out of the total amount of observations during CoMet 1.0, 84 % were performed at altitudes above 4 km and are of particular 260 interest for model validation. To demonstrate the utility of the observations to validate model results, as well as to help under-  stand the patterns in measured mixing ratios, we analyse and compare JIG measurements to CAMS high-resolution products for CO 2 , CH 4 and CO. Flight no. 2, shown in light red in Fig. 1, is discussed as an example.
The flight (Fig. 4,    In the following section, we analyse the model-data mismatch more closely using the subset of CoMet 1.0 data collected 300 only during the vertical soundings.

Vertical structure of the atmosphere
All profiles of CO 2 , CH 4 and CO collected with JIG are presented in Fig. 5, together with comparison to the CarboScope and CAMS model products. Individual comparisons are available in the supplement (Fig. S3-S10 and S11-S18). It should be noted that the mean profile for the lowest altitudes is dominated by a limited amount of cases when the ground level was reached. This  Above the altitude of 10 km, a more pronounced decrease in the mixing ratios is observed, which is directly related to occasional crossings into the tropopause region and the lowermost stratosphere. The variability of the observed decrease is large and follows the variability in the tropopause height. On average we have observed a 4 ppm decrease for CO 2 between 10 320 km and 13 km, which is most probably caused the increasing age of the slow-mixing stratospheric air (Andrews et al., 2001).
Decreases of CH 4 and CO are more pronounced (on average 150 ppb and 70 ppb, i.e. 8 % and 45 % relative to the value at 10 km), underlining an increased oxidative breakdown of these tracers (added to the age effect in case of CH 4 ).
While the observed gradient is similar to previously reported studies (e.g. Wofsy, 2011;Sweeney et al., 2015;Umezawa et al., 2018), measurements from CoMet 1.0 also clearly indicate the increase in atmospheric concentrations over the past

Model validation
The vertical profile subset of the measurements was the basis of the comparison to the well-established global modelling 330 systems CAMS and CarboScope. Here, we focus on describing the vertical structure of the model-data mismatch, defined as the difference between the modelled results and in situ observations from JIG, presented in Fig. 6. Mirroring previous discussion The variability in the mismatch is highest closest to the surface (bottom 3 km), which is related to influences from local sources/sinks, as well as variability of atmospheric mixing and transport in the PBL, which are hard to represent at respective model resolutions (0.125°x 0.125°for CAMS, 4°x 5°for CarboScope). Another source of mismatch is related to uncertainties 340 in the emissions data used by the models. Validation of individual emission sources, while of critical importance, remains challenging. In addition, in the case of biospheric CO 2 , the prediction of fluxes on scales relevant for direct comparison of mixing ratios on regional scales also remains a difficult task. This is true for all the analysed compounds and both models, with a markedly larger discrepancy in the CarboScope product that clearly suffers as a result of its low spatial resolution. As the in situ measurements from CoMet 1.0 are not numerous enough to give a robust estimate for the European region, and differences 345 between the model predictions and observations will be heavily dependent on a specific synoptic range and distribution of sources in the vicinity, we do not provide any general statistics for this lowest part of the atmosphere.
In the free-tropospheric range, the mismatch represents large-scale offset between the model and observations better, and is only weakly dependent on the spatial distribution of the emissions sources. Under this assumption, the mismatch is mostly caused either by i) large (i.e. at least national) scale offsets in emission strengths, ii) bias in the initialisation of the forecasted 350 fields (with CAMS GHG and operational analysis fields which are a combination of model simulation and satellite observations), iii) errors in chemistry parameterisations (OH radical reaction chains, CH 4 and CO).
In the CAMS product, the offset between the modelled values and observations in the troposphere becomes stable with height for CH 4 and CO 2 , with a symmetric distribution around a mean value (CH 4 : 0 (14) ppb; CO 2 : 3.7 (1.5) ppm, where standard uncertainty in the final digits is given in brackets. For CO 2 , a substantial offset is still present, most probably connected with

360
For altitudes above 10 km, the mismatch between CAMS and observations shows larger variability for CH 4 and CO, with CO 2 discrepancies similar to those observed in the free troposphere. While the number of observations at these higher altitudes is relatively low compared to those below 10 km, we believe that these differences are also caused by errors in both transport and chemistry schemes in the IFS system. These have been investigated in some detail in the case of CH 4 , for which the errors in the stratosphere have been found to be larger than those observed in the troposphere (Verma et al., 2017).

365
Optimised CO 2 mixing ratios from CarboScope also show overall good agreement when compared to observations, despite lower model resolution compared to CAMS. The model-data mismatch is dominated by a random term in the free tropospheric range (0.8 (1.3) ppm). Interestingly, the distribution of the mismatch in this altitude range is a positively skewed Gaussian curve (Fig. 6, bottom-right panel), with the main peak almost symmetric around 0 ppm, and the tail being responsible for most of the offset in the 3-10 km range. The most probable cause is the inability of the model to represent convective uplifting of 370 CO 2 -depleted air from the PBL. It should also be noted that in the CarboScope product, a systematic over-prediction of CO 2 mixing ratios above 10 km (up to 5 ppm) is observed, which might be caused either by i) significant errors in the tropopause height or ii) too fast vertical mixing in the lower stratosphere, leading to underestimation of the gradient and the chemical age of CO 2 . In the PBL range, the mixing ratios are generally underestimated, sometimes by more than 10 ppm, albeit such a large discrepancy is only visible for the lowest altitude range (less than 1 km), where the sample size is low. Where the 375 observation set is more robust, the bulk of observations is characterised by discrepancies smaller than 10 ppm and can have either positive or negative sign. Such behaviour is to be expected when trying to compare local plume enhancements to the low-resolution model results that averages over large, inhomogeneous areas characterised by a dynamic spatio-temporal diurnal cycle of fluxes.
3.5 Additional data from discrete samples -JAS 380 Figure 7 presents additional data acquired throughout the campaign with discrete samples, with a detailed overview provided in the supplement ( Fig. S19 and Table S1). Apart from CH 4 and CO 2 , for which the flask data were used for validation, important constituents were monitored, offering further insights into the state of the atmosphere over Europe during the CoMet 1.0 mission. The general nature of the collected data follows the patterns described for in situ data, with three distinct abundance regimes: i) PBL, ii) residual layer / free troposphere and iii) tropopause and lower stratosphere, however with some marked 385 differences.
For N 2 O and SF 6 , both potent greenhouse gases (IPCC et al., 2013), there is no clearly visible mixing ratio gradient between the PBL and the free troposphere. For both gases, the variability is known to be dominated by the slow stratospheric transport, effectively causing the "age" of air masses to be higher than the tropospheric air below (Andrews et al., 2001). For N 2 O, this effect is superimposed on the additional signal caused by its photo-chemical destruction in the stratosphere. Notably, during 390 CoMet 1.0, two samples were collected with SF 6 mixing ratios elevated by approximately 0.2 ppt. The first was filled on June 7 th , at 9.2 km altitude, over Czechia, and second on June 12 th , at 7.6 km, during the downward profile over the Po Valley. The potential source of these two observations might be worth investigating, especially in light of the constant atmospheric increase of the SF 6 , despite substantial efforts to curb emissions of this potent greenhouse gas (Weiss and Prinn, 2011). Some attention was also given to molecular hydrogen (H 2 ) due to its potential feedbacks to the atmosphere oxidative capacity and stratospheric 395 ozone levels (see Batenburg et al. (2012) and references therein). Values measured during the mission, namely 540 ppb near the surface, approximately 550-560 ppb throughout the free troposphere and approx. 570 ppb in the lower stratosphere, are comparable to previously reported values, e.g. in the scope of the CARIBIC project (Batenburg et al., 2012). This structure is driven by the presence of a relatively strong soil sink in the latitude band covered during CoMet 1.0, as has been confirmed by modelling studies (e.g. Pieterse et al., 2011). O 2 /N 2 and Ar/N 2 ratios are presented for completeness, but are not discussed 400 in the present study.
Of particular interest during CoMet 1.0 was the stable isotopic composition of methane. Abundances of both δ 13 C−CH 4 and δ 2 H−CH 4 are strongly and negatively correlated (R = -0.88 and R = -0.96, respectively) with mixing ratios of methane, signifying the potential to use the isotopes as a marker of the source processes. Indeed, in the next section we present an application of using isotopic composition to differentiate between specific source types in the study area of the USCB.