Airborne quantification of net methane and carbon dioxide fluxes from European Arctic wetlands in Summer 2019

Arctic wetlands and surrounding ecosystems are both a significant source of methane (CH4) and a sink of carbon dioxide (CO2) during summer months. However, precise quantification of this regional CH4 source and CO2 sink remains poorly characterized. A research flight using the UK Facility for Airborne Atmospheric Measurement was conducted in July 2019 over an area (approx. 78 000 km2) of mixed peatland and forest in northern Sweden and Finland. Area-averaged fluxes of CH4 and carbon dioxide were calculated using an aircraft mass balance approach. Net CH4 fluxes normalized to wetland area ranged between 5.93 ± 1.87 mg m−2 h−1 and 4.44 ± 0.64 mg m−2 h−1 (largest to smallest) over the region with a meridional gradient across three discrete areas enclosed by the flight survey. From largest to smallest, net CO2 sinks ranged between −513 ± 74 mg m−2 h−1 and −284 ± 89 mg m−2 h−1 and result from net uptake of CO2 by vegetation and soils in the biosphere. A clear gradient of decreasing bulk and area-averaged CH4 flux was identified from north to south across the study region, correlated with decreasing peat bog land area from north to south identified from CORINE land cover classifications. While N2O mole fraction was measured, no discernible gradient was measured over the flight track, but a minimum flux threshold using this mass balance method was calculated. Bulk (total area) CH4 fluxes determined via mass balance were compared with area-weighted upscaled chamber fluxes from the same study area and were found to agree well within measurement uncertainty. The mass balance CH4 fluxes were found to be significantly higher than the CH4 fluxes reported by many land-surface process models compiled as part of the Global Carbon Project. There was high variability in both flux distribution and magnitude between the individual models. This further supports previous studies that suggest that land-surface models are currently ill-equipped to accurately capture carbon fluxes inthe region. This article is part of a discussion meeting issue 'Rising methane: is warming feeding warming? (part 2)'.


Introduction
As of 2020, atmospheric abundances of the greenhouse gases methane (CH 4 ) and carbon dioxide (CO 2 ) have increased by approximately 1155 ppb and 132 ppm, respectively, since 1850 AD, and continue to rise at an estimated rate of 9 ppb per year for CH 4 and 2 ppm per year for CO 2 [1]. The global atmospheric emission budgets of both CH 4 and CO 2 still remain uncertain, with the balance between total anthropogenic and biospheric sources and sinks yet to be fully understood and accounted for. A temporary stagnation in CH 4 growth between 1998 and 2007 [2], and renewed growth with a concurrent shift in carbon-13 isotopic ratio to lighter bulk abundance since 2007, further compound the current uncertainties associated with CH 4 source and sink apportionment [3][4][5].
Wetlands are understood to be a key ecosystem in terms of the surface exchange of climaterelevant trace gases. CH 4 is produced by methanogenic archaea under anoxic conditions with high soil organic carbon (SOC) in wetland soils. The magnitude of CH 4 production within wetland soils is highly sensitive to temperature, SOC availability, presence of vegetation, and water table height and hence oxygen content of the soil [6][7][8][9]. Consequently, the number of variables affecting CH 4 production, as well as their spatial and temporal variability, cause significant difficulty in parametrizing and predicting current and future CH 4 emissions from wetlands accurately [10]. Global wetlands are thought to represent the largest single natural source of atmospheric CH 4 , contributing approximately 101-179 Tg CH 4 yr −1 to the global CH 4 budget which represents 20% of the global yearly CH 4 source to the atmosphere [9,11,12]. In addition to producing CH 4 , well-drained mineral soils under aerobic conditions can facilitate oxidation of CH 4 to CO 2 by methanotrophic microorganisms [13], while the surface exchange of CO 2 is controlled by the balance between respiratory CO 2 production from soil carbon stocks and photosynthetic CO 2 uptake by vegetation [14]. The Arctic is currently a net CO 2 sink, with an average of −0.13 Pg CO 2 year −1 taken up by the terrestrial arctic; this CO 2 sink is highest in the summer months, when gross primary productivity is at a maximum [15]. Recent research has identified that longer Arctic growing seasons, increased precipitation and evapotranspiraton rates may be driving increases in the Arctic CO 2  higher ecosystem respiration rates and hence CO 2 emission have been linked to higher air temperatures in the Arctic. It is therefore evident that the rapidly changing climate in the Arctic has the potential to significantly impact the source-sink dynamics of CO 2 exchange in this area, and continuous in situ monitoring is crucial to assess the impact of climate change on Arctic CO 2 fluxes [16][17][18] Approximately 53% of global wetland area is situated in northern latitudes above 50°N [19]. Therefore, Arctic and Boreal wetlands contribute significantly to the global CH 4 budget [20]. In addition to the current high CH 4 emission from high-latitude wetlands, these areas are sensitive to increasing CH 4 emission from positive climate feedbacks and Arctic climate amplification. Arctic mean air temperatures have increased at more than twice the rate of the global average, with current arctic temperature growth over 1.5°C higher than the 1971-2000 global average temperature growth with further warming predicted for the future [21,22]. Higher temperature may result in increased microbial activity in wetland ecosystems, leading to enhanced methanogenesis [23]. Furthermore, thawing of permafrost as a result of increasing temperature may result in an increase in arctic wetland extent as well as enabling the release of organic carbon from the estimated approximately 1700 Pg of stored SOC in arctic permafrost [24][25][26][27]. It is therefore clear that the CH 4 emissions from high-latitude wetlands may become increasingly important over time due to their high sensitivity to climate change.
Wetland trace gas emissions are commonly determined via top-down inversion modelling, bottom-up process-based modelling [7], or upscaling of eddy covariance or chamber fluxes to a wider wetland area. [28,29]. Process modelling of CH 4 fluxes from the wider Arctic often feature large uncertainty ranges due to the highly complex set of variables that influence microbial CH 4 production and emission processes to atmosphere. Additionally, the spatial and temporal heterogeneity of wetland environments, as well as the poorly defined boundaries of wetlands that often change seasonally, add significant uncertainty to annualized emission estimates as process models often account poorly for fine spatial and temporal variability in emissions [7,9]. In situ measurements of wetland emissions can be used to evaluate and improve process model estimates. However, the majority of in situ flux measurements are on a much smaller spatial scale than typical model outputs (typically on a 0.5°grid), and there are currently few in situ measurements on an appropriate scale for more direct model comparison [29,30]. Aircraft measurement platforms allow in situ measurements of trace gas emissions to be carried out on a similar spatial scale to process models, albeit as discrete snapshots of flux, and may allow the uncertainties on emission estimates from such models to be better constrained [31][32][33].
The Methane Observations and Yearly Assessments (MOYA) project aimed to use in situ measurements, targeted field campaigns and modelling to constrain global CH 4 sources and sinks from a variety of key CH 4 emission hotspots, such as African biomass burning [34] and Tropical wetlands (Shaw et al. in review). In situ measurements of CH 4 fluxes in these key areas will aid in reducing the uncertainty in their contribution to the global CH 4 budget and may provide a clearer explanation for currently rising atmospheric CH 4 mole fractions (MFs). As part of the MOYA project, the MOYA-Arctic field campaign was conducted from 29 July 2019 to 2 August 2019 based in Kiruna, Sweden. This field campaign used in situ aircraft measurements to quantify emissions of CH 4 and other trace gases from northern Swedish and Finnish (Fennoscandian) wetlands (66-69°N, 22-28°E) during the summer period. This work presents in situ aircraft measurements of CH 4 , CO 2 and N 2 O MF during one of the survey flights carried out during the MOYA-Arctic campaign. From these measurements, mass balance flux estimates for CH 4 and CO 2 were calculated and compared with previous similar aircraft studies in the region by O'Shea et al. [33]. Despite no direct flux being attainable from the N 2 O data, a minimum flux threshold using this mass balance method was calculated for N 2 O. Additionally, this study compares the fluxes obtained via aircraft mass balance with fluxes from Global Carbon Project (GCP) wetland process models, where both the magnitude and spatial distribution of CH 4

Methods (a) Airborne instrumentation
The FAAM BAe 146-301 Atmospheric Research Aircraft (FAAM ARA) was operated for in situ sampling during the MOYA-Arctic campaign. Thermodynamic and meteorological parameters such as temperature, pressure and three-dimensional wind vector were measured by the FAAM ARA core instrument suite [35]. Temperature was measured by a Rosemount 102 sensor, with an estimated uncertainty of 0.1 K. Static pressure was measured by a series of pitot tubes distributed across the aircraft surface, with an uncertainty of 0.3 hPa. The three-dimensional wind vector is measured by a nose-mounted five port turbulence probe, with an uncertainty of 0.2 m s −1 .
A Los Gatos Research Fast Greenhouse Gas Analyser (FGGA) was used for 10 Hz measurements of CO2 and CH 4 MF. The FGGA instrument uses a Cavity-Enhanced Absorption Spectroscopy technique and two continuous-wave near-IR diode lasers. The FGGA is mounted within a 19-inch rack in the cabin of the aircraft with ambient air pumped via a rearward-facing 3/8' stainless steel inlet mounted to a window blank. The FGGA was calibrated using three calibration gas standards: high-and low-concentration calibrations to account for instrument drift over the course of a flight, and a target calibration to assess long-term instrument precision and bias over multiple flights. All three calibration standards were traceable to the National Oceanic and Atmospheric Administration/Earth System Research Laboratory (NOAA/ESRL) World Meteorological Organisation (WMO) X2004A scale for CH 4 and X2007 scale for CO 2 [34,36]. Accounting for all sources of uncertainty, the mean (calibrated) biases and associated 1σ overall uncertainties are estimated to be −0.048 ± 0.626 ppm and −1.22 ± 2.93 ppb, respectively for 10 Hz CO 2 and CH 4 sampling during MOYA-Arctic. One hertz measurements of N 2 O MF were sampled by an Aerodyne Quantum Cascade Laser Absorption Spectrometer (QCLAS). The QCLAS was calibrated by means of three calibration gas standards, which were traceable to the WMO X2006 calibration scale [34,37]. An overall 1σ uncertainty of 0.58 ppb was estimated for 1 Hz N 2 O MF measurements during the MOYA-Arctic flights.

(b) Aircraft mass balance flux technique
Aircraft mass balance flux techniques are well established in their ability to quantify trace gases fluxes from various sources, including regional-scale city emissions [38][39][40], point-source oil and gas emissions [41][42][43] and regional-scale biospheric trace gas emission/uptake [33,44]. For reliable flux quantification using aircraft mass balance, several criteria must be satisfied. First, MF measurements must be made downwind of a targeted emission source. Second, background measurements should be made, either within the centre of the well-mixed boundary layer upwind of the targeted emission source, or from downwind measurements either side of the emissions plume from the targeted emission source. These background measurements represent an estimate of the MF that would have been measured downwind of the targeted source in the absence of any emissions from that source. Additionally, wind direction should ideally be perpendicular to upwind and downwind sampling to ensure the measured airmass advects over the emission source, and wind speed should be constant for mass balance calculations. The meteorological conditions at the time of the survey flight reported here were highly favourable for this approach and the survey design was optimized to sample accordingly (described in §3). Flux determination by aircraft mass balance is expressed by equation (2.1). The flux of a trace gas species in g s −1 is defined as the enhancement in trace gas concentration (C Enh is the enhanced concentration downwind in this case, C 0 is the background concentration). MFs are first converted to concentrations in units of g m −3 using equation (2.2), where ρ air is the molar density of air, M x is the molar mass of CH 4 or CO 2 and M air is the molar mass of air. (C Enh -C 0 ) is then multiplied by the windspeed perpendicular to the flight track in m s −1 , U ⊥ , integrated over the length of the downwind flight transect, x, and the height of the convective boundary layer, z. Measured statistical variability in the background concentration and wind vector, as well as measurement uncertainty and quantified systematic uncertainty in the height of boundary layer mixing (diagnosed from thermodynamic profiles), are propagated through equation (2.1) to determine flux uncertainty [33]. In addition to mass balance flux techniques, the FAAM ARA is capable of quantifying trace gas fluxes using the eddy covariance technique [45,46]; however, the magnitude of vertical windspeed during flight C195 was not sufficient for reliable calculation of CH 4 or CO 2 fluxes using eddy covariance in this study.

(c) Chamber fluxes
The mass balance fluxes derived from airborne measurements have also been compared to area-weighted chamber flux measurements, which were carried out in the same study area investigated here as part of the CH 4 and other greenhouse gases in the Arctic-Measurements, process studies and Modelling (MAMM) project. These chamber experiments were carried out daily between 12 July 2012 and 2 August 2012 and yielded area fluxes by specific land type for wetland (4.5 ± 3.7 mg m −2 h −1 ) and forest (0.05 ± 0.07 mg m −2 h −1 ) for summer. These area fluxes were scaled using the total wetland and forested area fraction with each of the three flux areas surveyed here according to the CORINE land cover map. The total wetland area was calculated as the sum of the peat bog and inland marsh grid cells within each area, and the total forested area was determined as the sum of all forest subclasses (broadleafed, coniferous and mixed forest) cells for each area. The chamber area fluxes were then multiplied by the total wetland or forest areas to give a bulk flux value for each of the three distinct flux areas.

(d) Flight description and strategy
The target area of FAAM ARA Flight C195 (figure 1) is mostly Northern Finnish Lapland, but also encompasses parts of Northeast Sweden (Norbotten County) and North Norway (Finnmark County). The area surveyed was comprised boreal (Taiga) forest interspersed with peat bogs and lakes. Seasonal thaw of accumulated winter snow and ice typically results in the high prevalence of semi-permanent water bodies and peatland mires in the summer months. Flight C195 was carried out on 31 July 2019 between 10 : 00 and 14 : 30 CEST and involved four straight aircraft transects of approximately 200 km length across the wetland area, each at constant latitude. The first of these transects was the northernmost upwind leg at 69°N latitude, and the legs step down southwards in increments of 1°N with the final southernmost downwind leg at 66°N (as shown in figure 1). These constant latitude transects at 69°N, 68°N, 67°N and 66°N are referred to as transects 1, 2, 3 and 4 throughout. All transects across the wetland were conducted at altitudes between 300 m and 600 m above ground level (agl). Six deeper profiles (three ascents, three descents) from approximately 300 m agl to approximately 2500 m agl were carried out at the start, middle and end of the flight in order to assess planetary boundary layer (PBL) height and development used to derive mixing height for equation (2.1) over the course of the sampling period. A single biomass burning plume was intercepted at approximately 12:22 CEST, but this was removed from the CO 2 and CH 4 data prior to flux calculations. Measurements of carbon monoxide (CO) remained constant during the flight (with the exception of the single fire plume), strongly suggesting that this biomass burning event as well as any other anthropogenic sources did not have any impact on CH 4 MFs further downwind.   figure S2). However, a theoretical 'limit of detection' for N 2 O mass balance fluxes using the aircraft instrumentation was derived using the standard deviation of the N 2 O MF over transect 1, and this is detailed in the electronic supplementary material. Figure 2 shows the potential temperature (θ ), CH 4 and CO 2 MFs during the six altitude profiles carried out in flight C195. All profiles were conducted within the near vicinity of the study area at the start, middle and end of the flight, and the profiles bracket the four longitudinal transects across the study area (see electronic supplementary material, figure S3). There was very little change in PBL height between the first and second set of vertical profiles as diagnosed from the characteristic sharp change in potential temperature gradient seen at PBL top (dashed blue lines in figure 2). However, there is a significant difference between the final profile ascent (figure 2e) and the final profile descent (figure 2f ), as the PBL height is observed to be approximately 1000 m agl, whereas the descent shows a PBL height approximately 450 m higher at approximately 1450 m AGL, this final descent profile is therefore not used in PBL determination for mass balance calculations. To account for this change in mixing height used in the mass balance approach, the nearest available thermodynamic profile to each transect was used to determine PBL height in the flux calculations (i.e. only the ascending profile in figure 2e is used). The relatively small increase in PBL height over the course of flight C195 suggests that any entrainment of free tropospheric air into the PBL can be considered to be negligible and therefore will not significantly affect the uncertainty of flux estimates calculated here. Furthermore, MFs of CH 4 and CO 2 within the PBL were observed to be constant within each of the deep profiles, suggesting that the PBL was well mixed throughout the study region.
Wind direction over the course of flight C195 was predominantly northerly during transect 1 and transect 2 as shown in table 1 and by the wind barbs in figure 1a. As the flight progressed, the average wind direction changed from northerly to more north-easterly winds towards the southern end of the flight track was also confirmed by HYSPLIT back-trajectories with trajectory Vertical profiles of potential temperature, CH 4 mixing ratio and CO 2 mixing ratio during the six profiles (labelled chronologically (a-f )) carried out by the FAAM ARA during flight C195. Approximate convective mixing heights, determined by a change in vertical gradient in potential temperature, for each profile are also displayed as blue dashed lines. (Online version in colour.) endpoints calculated for each transect shown in figure 3. Owing to this gradual change in wind direction over the course of the flight, a mass balance flux calculation across the entire flight area (i.e. using transect 1 as the background and transect 4 as the enhanced run) would be inappropriate, as transect 1 does not sample the same airmass as transect 4. Therefore individual fluxes were calculated between parallel meridional transect pairs, with the northern transect of each pair used to determine the upwind background, and the southern transect to determine the CH 4 gradient over the distance between each pair. The three areas between the meridional transect pairs are referred to as Area 1, Area 2 and Area 3 throughout. Area 1 is between transect 1 (69°N) and transect 2 (68°N), Area 2 is between transect 2 (68°N) and transect 3 (67°N), and Area 3 is between transect 3 (67°N) and transect 4 (66°N).  The net CO 2 uptake observed during this study is higher than that reported in previous work. The CO 2 sink reported from Arctic wetlands by O'Shea et al. is −350 ± 143 mg m −2 h −1 , which agrees within overlapping 1σ uncertainty for CO 2 area fluxes reported here for each area (table 1), despite the maximum average CO 2 flux value calculated in this study being 24% higher than that determined in O'Shea et al. However, the Christensen et al. chamber CO 2 flux of −96 ± 33 mg m −2 h −1 is significantly lower than the CO 2 area fluxes for Area 1 and Area 2 [24]. Biospheric CO 2 fluxes are known to exhibit strong spatio-temporal variability that is highly sensitive to temperature, precipitation, insolation and leaf area index of the vegetation types studied, and therefore a close agreement between studies conducted on different days and years is not expected. Table 1 shows that the mean CH 4 emission rate and area flux decreases with decreasing latitude from Area 1 to Area 3. Figure 4 shows the 2018 Copernicus Land Monitoring Service    CORINE land cover classification of the study area (https://land.copernicus.eu/pan-european/ corine-land-cover/clc2018), and table 2 shows the most abundant land classes within each flux area by percentage. It can be seen from both table 2 and visually from figure 4 that the abundance of peat bogs decreases towards the south of the survey area. Peat bogs comprise 22.9% of the land cover within the northernmost Area 1 between transect 1 and transect 2, this decreases slightly to 20.0% within Area 2 and decreases further to 7.28% within the southernmost Area 3. The decreasing peat bog abundance towards the southern end of the survey area provides a likely explanation for the gradually decreasing CH 4 flux seen from north to south in table 2. Additionally, there is a positive correlation between CO 2 sink magnitude and CORINE vegetation cover within the three areas of the flight (electronic supplementary material, figure S5). However, the correlation between CO 2 sink and vegetation cover is weaker than the CH 4 flux-peatland area correlation. This is likely due to the differing CO 2 uptake capacities of specific vegetated land types (e.g. dense forest will sequester more CO 2 than an equivalent area of cropland).
The mass balance fluxes of CH 4 derived in this study are compared to upscaled chamber CH 4 flux measurements that were previously taken in the same study area. A description of the chamber measurements as well as the method of upscaling these fluxes can be found in §2.3. The chamber flux results are shown in table 1 and figure 5. It can be seen that the scaled chamber fluxes have a larger relative error of between 84% and 88% of the flux value when compared to the mass balance fluxes (between 14% and 31%); however, the mean mass balance and chamber bulk fluxes agree very well within overlapping 1σ uncertainty for all three flux areas. The mean bulk fluxes from mass balance for Areas 1 and 2 (7.85 ± 1.06 kg s −1 and 6.37 ± 2.01 kg s −1 , respectively) are approximately 15-25% higher than the scaled chamber fluxes of 6.86 ± 5.75 kg s −1 for Area 1 and 5.04 ± 4.26 kg s −1 for Area 2. The slightly higher fluxes from mass balance could be associated with the presence of plant-mediated wetland CH 4 emission via the transport of CH 4 through specialized plant tissues. Emission from this pathway would be detectable using mass balance techniques but may be missed when using flux chamber apparatus mounted at ground level. However, the agreement between the two techniques provides support for the efficacy of the mass balance technique compared to ground-based flux quantification techniques and demonstrates the potential for spatial scalability and interpretation of point measurements such as chamber fluxes.

(b) Comparison with land-surface model methane fluxes
Top-down in situ flux estimates such as those derived in this work can provide an important comparison to emission estimates from bottom-up biogeochemical process models and anthropogenic emission inventories and can provide crucial validation of such models. Wetland CH 4 fluxes are typically derived by land-surface models by parametrizing key biogeochemical characteristics, such as CH 4 production, transport and oxidation within wetland soils, as well as the amount and type of vegetation present. These initial parameters are then forced by environmental variables such as precipitation, temperature, respiration and atmospheric CO 2 concentration in order to account for seasonal and interannual differences in CH 4 emission [48,49]. The CH 4 flux density output from biogeochemical parametrization is then combined with a wetland distribution map for a given area to spatially distribute the CH 4 flux and produce a flux map [7]. here use a diagnostic means of prescribing wetland cover, namely the Wetland Area Dynamics for Methane Modelling (WAD2 M) which uses satellite microwave remote-sensing inputs [50]. Seven of the 13 process models also include prognostic with internal wetland prescription in each model. Figure 6a,b show modelled CH 4 flux distribution maps for the diagnostic and prognostic models, respectively. The majority of diagnostic models share a common spatial distribution of CH 4 flux due to the WAD2M product that these models use to prescribe wetland cover. This common flux pattern shows flux hotspots in the northern two-thirds of the study area, particularly to the northeast. The prognostic model outputs do not show a common pattern of CH 4 flux distribution as with the diagnostic models, and flux distributions are much more variable in these model variants. The majority of diagnostic GCP models and a select few of the prognostic models (namely LPX-Bern and ORCHIDEE) show peak fluxes in the northern two-thirds of the study area, which broadly agrees with the mass balance flux observations where the highest fluxes were also measured in the northern two-thirds. Additionally, the aforementioned models also appear to show flux hotspots towards the eastern end of the study area. The mass balance technique could not explicitly resolve west to east flux gradient in this case; however, figure 1 appears to show higher CH 4 MFs towards the eastern end of the flight track suggesting that CH 4 fluxes may be higher towards this eastern end. Despite the differences and similarities in flux distribution between models and mass balance, it should be noted that the model outputs presented here are July averages over period of 17 years. It is therefore highly probable that wetland distribution in this area has changed over this time period, and good agreement between model flux distribution and mass balance flux distribution is not necessarily expected due to this. Figure 7 shows bar plots of the CH 4 flux from the diagnostic and prognostic process models along with the mean mass balance fluxes from the three distinct flux areas identified in table 1. Most of the diagnostic and prognostic models report significantly lower CH 4 fluxes for all three study areas compared to the mass balance results. In general, the prognostic models report higher CH 4 fluxes for all three areas than the diagnostic models, most notably with the ORCHIDEE diagnostic model where fluxes for Areas 1 and 2 (1.32 ± 0.47 mg m −2 h −1 and 1.29 ± 0.54 mg m −2 h −1 , respectively) agree well within overlapping uncertainty with the mass balance fluxes for Areas 1 and 2. Despite the general disagreement between modelled and mass balance CH 4 fluxes in this case, it is worth noting that the mass balance results represent a single temporal snapshot from a single daytime flux from July 2019, whereas the model outputs are July monthly averages from 2000 to 2017. A likely source of disagreement between mass balance and process modelling in this case is that the mass balance may not be truly representative of the monthly average model output over multiple years. Having said this, an average air temperature of 12.2°C was measured on 31 July 2019 from the Sodankylä Lokka weather station during the time of the flight, which was slightly lower than the July mean temperature between 2000 and 2017 for the same weather station (14.5 ± 1.6°C). In addition, the average precipitation for July 2019 (33.4 mm) was also significantly lower than the July average precipitation between 2000 and 2017 (77.4 ± 28.2 mm) [51]. Lower temperature and precipitation for July 2019 suggest that the mass balance CH 4 fluxes reported in this work may actually be lower than previous years, yet many of the GCP process models report significantly lower fluxes for the years previous to this study.
In summary, the land-surface models assessed here generally provide a lower estimate of wetland CH 4 flux than top-down aircraft mass balance techniques for the study area of northern Sweden and Finland. However, the mass balance flux was measured during the daytime whereas the July average model outputs are comprised 24 h flux outputs. Wetland fluxes in the Arctic are known to exhibit a diurnal cycle with daytime flux maxima and night-time minima [52]; the inclusion of nocturnal low flux periods within the model outputs may partially account for the lower estimates of CH 4 flux compared to mass balance. In addition to this, the GCP models only account for CH 4 emission from areas classified as wetlands and do not account for lake, riverine or other biogenic CH 4 sources. Mass balance will capture the flux footprint from all sources in the study area, not solely wetland. Therefore this could also account for the higher mass balance   (a) shows models with the diagnostic WAD2M wetland prescription and (b) shows models with the prognostic wetland cover information determined within the models. The model data are obtained from the supplementary data of Saunois et al. [9]. (Online version in colour.) magnitude and distribution, and the models likely estimate lower CH 4 flux even when taking the previous points into account. In order to provide model fluxes that are both more precise and more accurate, improvements in model inputs that more successfully estimate CH 4 flux in comparison to in situ measurements, as well as standardized, accurate estimates of wetland cover and dynamics, are clearly both required. More frequent observational flux measurements are also ideally needed to provide important intercomparison and evaluation for model techniques.
Put simply, the GCP models disagree markedly with one another, and with the measurements reported here for the region studied. It is imperative that this is addressed as a priority in order to more meaningfully use GCP models for Arctic carbon emissions, especially given the Arctic's rapidly changing climate. 5.93 ± 0.72 mg m −2 h −1 was obtained for CH 4 , and a peak total land area-normalized flux of −513 ± 74 mg m −2 h −1 was obtained for CO 2 using the aircraft mass balance flux method for this area of northern Sweden and Finland (approximately 78 000 km 2 ). The bulk CH 4 fluxes determined via mass balance were found to agree well with upscaled chamber fluxes for the same study area. These results indicate that the wetlands in this area are a significant net source of CH 4 , and the area also represents a notable biospheric CO 2 sink. A clear gradient of decreasing CH 4 flux was identified between the northern and southern end of the flight track, which appears to correlate with decreasing peat bog land cover percentage from north to south. The mass balance fluxes were also compared with a variety of GCP land-surface process model fluxes, the majority of which were found to significantly underestimate CH 4 emission in this area when compared to the mass balance. The results from this study provide an important wetland trace gas emission dataset that will aid validation of global land-surface models and will help further constrain the contribution of Arctic wetland and vegetation to global CH 4 and CO 2 budgets. Furthermore, the results highlight the sensitivity of bottom-up process models to accurate wetland cover and dynamics estimations and other input parameters when quantifying flux using these methods. This study also highlights an urgent need to improve land-surface models by using high-accuracy observational wetland cover datasets as model inputs, and by continuing in situ measurements as a means to evaluate the performance of these models. Continued improvements to land-surface models will allow them to more accurately predict summer CH 4 emissions in the Arctic.

Summary and Conclusion
Data accessibility. Aircraft data: FAAM ARA data from the MOYA project can be found on the CEDA archive