Atmospheric CO2 and CH4 abundances on regional scales in boreal areas using CAMS reanalysis, COCCON spectrometers and Sentinel- 5 Precursor satellite observations

We compare the atmospheric column-averaged dry-air mole fractions of carbon dioxide (XCO2) and methane (XCH4) 15 measured with a pair of COCCON spectrometers at Kiruna and Sodankylä sites in boreal areas with model data provided by the Copernicus Atmosphere Monitoring Service (CAMS) and with XCH4 from the recently launched Sentinel-5 Precursor (S5P) satellite. In addition, measured and modeled gradients of XCO2 and XCH4 (ΔXCO2 and ΔXCH4) on regional scales are investigated. Both sites show a similar and very good correlation between COCCON retrievals and the modeled CAMS XCO2 data, while CAMS data are biased high with respect to COCCON by 3.72 ppm (±1.80 ppm) in Kiruna and 3.46 ppm (±1.73 20 ppm) in Sodankylä on average. For XCH4 CAMS values are higher than the COCCON observations by 0.33 ppb (±11.93 ppb) in Kiruna, and 7.39 ppb (±10.92 ppb) in Sodankylä. In contrast, the S5P satellite generally measures lower atmospheric XCH4 than the COCCON spectrometers, with a mean difference of 9.69 ppb (±20.51 ppb) in Kiruna and 3.36 ppb (±17.05 ppb) in Sodankylä. We compare the gradients of XCO2 and XCH4 (ΔXCO2 and ΔXCH4) between Kiruna and Sodankylä derived from CAMS reanalysis and COCCON and S5P measurements to study the capability of detecting sources and sinks on regional 25 scales. The correlations in ΔXCO2 and ΔXCH4 between the different datasets are generally smaller than the correlations in XCO2 and XCH4 between the datasets at either site. The ΔXCO2 predicted by CAMS are generally higher than those observed with COCCON with a slope of 0.51. The ΔXCH4 predicted by CAMS is mostly higher than that observed with COCCON with a slope of 0.65, covering a larger dataset than the comparison between S5P and COCCON. When comparing CAMS ΔXCH4 with COCCON ΔXCH4 only in S5P overpass days (slope = 0.53), the correlation is close to that between S5P and COCCON 30 (slope = 0.51). CAMS, COCCON and S5P predict gradients in reasonable agreement. However, the small number of observations coinciding with S5P limits our ability to verify the performance of this sensor. We detect no significant impact of ground albedo and viewing zenith angle on the S5P results. Both sites show similar situation with the average ratio of XCH4 https://doi.org/10.5194/amt-2020-19 Preprint. Discussion started: 12 February 2020 c © Author(s) 2020. CC BY 4.0 License.

properties and aerosols on the measurements are minimal (Wunch et al., 2017). The measurements are scaled to the World Meteorological Organization (WMO) reference scale applying a post correction and thereby guaranteeing high accuracy (Wunch et al., 2015). The high-resolution TCCON sites are distributed globally, however, many of these are concentrated in Europe, Northern America and eastern Asia. The costs, logistic requirements and the need of qualified personnel on site have hindered the expansion of the network e.g. to the African continent, South America and central Asia (Wunch et al., 2011). 70 Remote sites and regions with high or low surface albedo are generally poorly covered by the TCCON network. Ground-based measurement stations in the above mentioned regions are needed for satellite and model validation and carbon cycle science.
Recently, cheaper and portable spectrometers have been developed and are now available for GHG measurements, with the potential to complement the TCCON network (Frey et al., 2019;Sha et al., 2019). The EM27/SUN FTIR spectrometer was developed by Karlsruhe Institute of Technology (KIT) (Gisi et al., 2012), in cooperation with Bruker Optics GmbH, Ettlingen, 75 Germany. It is available from Bruker as a commercial device since spring 2014. The EM27/SUN instrument is a portable ground-based FTIR spectrometer, consisting of a spectrometer body with dimensions of 35 × 40 × 27 cm and a solar tracker which is directly mounted on the spectrometer. The whole weight is approximately 25 kg and can be carried by one person.
This solar-viewing FTIR instrument has a resolution of 0.5 cm -1 , similar to that of TROPOMI. This compact and mobile EM27/SUN instrument is appropriate for field campaigns as well as for long-term deployment at a site with the potential to 80 complement the TCCON network. In addition, its excellent robust and reliable characteristics have been demonstrated in several successful field campaigns (Frey et al., 2015;Klappenbach et al., 2015;Butz et al., 2017;Vogel et al., 2019;Kille et al., 2019;Sha et al., 2019b). KIT performs final optimisations, an expert review of instrument performance and a final calibration of each unit with respect to the reference EM27/SUN spectrometer operated at KIT and the TCCON site in Karlsruhe. In the framework of ESA recent projects, codes required for the data analysis of EM27/SUN measurements have 85 been developed by KIT, which are open source and freely available (https://www.imk-asf.kit.edu/english/3225.php). If the operation of EM27/SUN spectrometers adheres to the described standards (use of calibrated units, processing using the provided codes), then this practice is compatible with the requirements of COCCON (Collaborative Carbon Column Observing Network, see Frey et al., 2019). The data presented in this paper have been generated using a pair of EM27/SUN spectrometers following these requirements. For this reason, we refer to these as COCCON spectrometers in the following. This paper 90 compares S5P observations to the ground-based observations from two COCCON spectrometers operated in the high latitude regions in Sodankylä, Finland and Kiruna, Sweden. The measurements from these two sites are highly valuable for investigating the gradients of the greenhouse gas distribution on regional scales near the Arctic Circle.
In addition to the COCCON and the TROPOMI datasets, we investigate the CO2 and CH4 products from Copernicus Atmosphere Monitoring Service (CAMS). CAMS services are operated by the European Centre for Medium-Range Weather 95 Forecasts (ECMWF), providing near-real-time analysis and forecast data with a spatial resolution of approximately 80 km.
The CAMS reanalysis dataset is the latest global reanalysis dataset of atmospheric composition, though a reanalysis for the greenhouse gases (CO2, CH4) is being produced separately and will only be released in 2019 (Inness et al., 2019). This work uses 6-hourly analyses CAMS data of CO2 and CH4, initialized daily from analyses at 00:00 UTC. CAMS profiles of CO2 and https://doi.org/10.5194/amt-2020-19 Preprint. Discussion started: 12 February 2020 c Author(s) 2020. CC BY 4.0 License.
CO4 are also used to study the quality of a-priori profiles used for the trace gas retrievals, and compared with the TCCON 100 official MAP a-priori profiles.
In this study, we provide an intercomparison of atmospheric CO2 and CH4 column-averaged abundances derived from ground-based COCCON spectrometers and the CAMS reanalysis dataset, and CH4 from S5P satellite within the Arctic Circle.
Another objective is to explore the ability of the S5P satellite and the COCCON in observing the gradients of GHGs on regional scales in high latitude regions. The following section gives a description of the sites and data sources. The results and 105 discussions are given in section 3 and the final conclusions are discussed in section 4. The public S5P CH4 data are only available since May 2018. The comparison between the S5P and the COCCON measurements starts since the beginning of the public data. Currently, the Level-2 (L2) products of S5P are released, including the column-average dry-air mole fraction of methane, XCH4. This value presents the total column of methane in the atmosphere 120 from the surface up to the top of the atmosphere divided by the corresponding dry-air column (Apituley et al., 2017). S5P L2 products provide bias corrected XCH4 retrievals, which are used in this work. The quality control value (qa_value) is given as part of the CH4 data product and it is recommended use only data with qa value above 0.5 to exclude data of questionable quality. To compare with the COCCON data, S5P data are collected from the average value within a radius of 100 km around each station. The radius criterion of 100 km was the best tested case as discussed in Sha et al., 2019a. When comparing the 125 bias corrected S5P XCH4 product with the NDACC and TCCON FTIR products, it shows slightly higher correlation in using the radius criterion of 100 km than those of using 50 km. A 10-minute average value of COCCON data (retrieved from approximate 10 spectra) is obtained at the coincident S5P overpass time. The overpass time over Kiruna and Sodankylä stations is between 9 UTC to 12 UTC. The standard error of mean is used as error bar, as it presents the estimation of the standard deviation of its sampling distribution and is calculated by using:

Sites and data sources
here the is single measurement in the defined area or time range, ̅ is mean value of data sample, n is the number of data points. This method is useful to distinguish highly scattered dataset, especially in S5P and CAMS data, which come from large areas.
The comparison between the CAMS reanalysis and the COCCON observations starts from the beginning of the field 135 campaign (March 2017). CAMS provides a 6-hourly reanalysis data in defined areas around Kiruna and Sodankylä. These defined areas resemble rectangles of 100 km × 100 km, covering 67°N -69°N and 18°E -23°E around Kiruna; 66.5°N -68.3°N and 24°E -29°E around Sodankylä. In these defined areas there are 476 data points in total in the area of Kiruna and 442 data points in the area of Sodankylä within their respective measuring periods. We use the average value from these points as hourly CAMS reanalysis data. The coincident COCCON data are collected from one-hour average at 6 UTC or 12 UTC, 140 because the spectrometer measures only at daytime. Additionally, selection criteria are applied to the COCCON data as described in the work of Frey et al. (2015). Measurements are filtered out as they were measured at solar zenith angle (SZA) > 80° to reduce uncertainties connected to spectra recorded at very high airmasses. The data are also filtered based on Xair (column-averaged amount of dry air) and Xair range between 0.995 and 1.005 is required.
Any practical choice of the a-priori volume mixing ratio (VMR) profiles is generally based on model data. To assess the 145 quality of the model data, knowledge of the actual profiles is required and might be obtainable from in-situ instruments onboard aircrafts performing profile measurements or from in-situ AirCore balloon launches. The AirCore, which was an auxiliary activity in the FRM4GHG campaign, is a simple and viable atmospheric sampling system to measure vertical profiles of greenhouse gases (Karion et al., 2010). The AirCore system that was used in Sodankylä was built at the University of Groningen (UG) and at the FMI. It consists of a 100 m long coiled stainless steel tube, combining ~40 m of 0.25 inch (6.35 150 mm) tube and ~60 m of 0.125 inch (3.175 mm) tube, along with an automatic shut-off valve and home-made data logger to record temperature and pressure during the flight. A 3 kg meteorological balloon was used to launch the AirCore along with a radiosonde and the payload positioning system. The air is evacuated from the tube during ascent to an altitude of ~30 km due to the pressure difference, while ambient air flushes into the tube as it descends. Upon landing the automatic valve shuts off to prevent any further exchange of the sampled air inside the tube with ambient air. A cavity ring-down spectrometer (CRDS) 155 manufactured by Picarro Inc. is used afterwards to quantify the mole fractions of the target gases (e.g. CO2 and CH4) in the AirCore sample.

Quality of a-priori profiles and their influence on the retrieval results
The choice of a-priori VMR vertical profiles for the target gases is important for retrieving correct column abundances from 160 ground-based FTIR spectra. The PROFFAST algorithm for retrieving COCCON observations is a nonlinear least squares https://doi.org/10.5194/amt-2020-19 Preprint. Discussion started: 12 February 2020 c Author(s) 2020. CC BY 4.0 License. spectral fitting algorithm, scaling an a-priori dry-air mole fraction of profile to generate the best spectral fit to the measured spectrum by using a single scale factor.
In the following section two different sets of a-priori profiles are used for investigating the sensitivity of the retrievals with respect to the choice of the profiles. One set of VMR profiles (MAP) is the one used by TCCON. The profiles are 165 derived from a stand-alone program to generate profiles as described in Toon et al., 2015. These profiles are based on temperature, pressure and humidity generated by National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR), empirically derived from MkIV FTS balloon flights (Toon, 1991)

Comparison of the MAP and the CAMS profiles for the Sodankylä campaign site
The CO2  The difference drops to -14.4 ppm on March 9, then increases up to 11.5 ppm on July 23, 2017, while it is -17.6 ppm on February 7, then increases up to 11.6 ppm August 5, 2018. The MAP profiles become underestimated in autumn and the largest difference at ground level is -14.9 ppm occurring on September 5, 2017 and is -21 ppm on August 9, 2018. In the stratosphere 185 the CAMS CO2 profiles show smaller vertical changes compared to the MAP profiles over the year, however they are generally higher in concentration than the MAP profiles over 40 km in 2017 and over 30 km in 2018. Altogether, the MAP a-priori profiles agree quite well with the CAMS profiles.
A much larger difference exists in CH4 between the MAP and the CAMS profiles. CAMS shows a significant seasonal change, especially in the stratosphere, while MAP is more constant and overestimated compared to CAMS over the whole 190 year. In contrast to CO2, the highest differences between MAP and CAMS appear in the lower stratosphere between 20 km and 40 km as seen for both 2017 and 2018 plots (Figure 1). In the beginning of the year, the difference between MAP and CAMS profiles is around 0.9 ppm at 28 km, and reaches to nearly 1 ppm at a lower level of 20 km in spring. being close to CAMS and the highest difference is around 0.35 ppm at 33 km in summer 2017 and 2018. In winter, the 195 differences are also obvious and near to 0.9 ppm at around 30 km. The steeper vertical gradients together with the dynamical processes occurring in the polar atmosphere make a climatological guess of a-profile shape much harder than for carbon dioxide and therefore the MAP a-priori profiles are less realistic for methane. We will investigate in the next section using AirCore soundings to which degree CAMS is capable of following the actual profile variability.

Comparison of in-situ AirCore profiles and CAMS profiles for the Sodankylä campaign site 200
The in-situ profiles are derived from the AirCore balloon launches at the Sodankylä campaign site and up to an altitude of approximate 30 km. Figure 2 shows the differences of CO2 and CH4 between the AirCore and the CAMS profiles for 10 measurement days in 2017 and 9 measurement days in 2018. The AirCore launches cover the spring to autumn period.
The CAMS CO2 profiles are generally overestimated compared to the AirCore profiles, with a mean difference on average of 1.35 ppm in 2017 and 3.33 ppm in 2018. In the 10 AirCore launched days in 2017, CAMS profiles are slightly overestimated 205 in the stratosphere, while the tropospheric CAMS profiles are closer to the AirCore profiles in summer than those in autumn 2017. Two peak differences are found at altitude around 9 km with -5.98 ppm (AirCore -CAMS) on April 24 and -9.46 ppm on April 26 and another peak at around 1 km with -5.76 ppm on September 5, 2017. The CAMS profiles show a slightly higher overestimation in CO2 profiles in summer than another days in 2018 and which is likely due to the seasonal bias in CAMS. In general, CAMS profiles are overestimated over the whole vertical altitude range and differences in the stratospheric part are 210 quite constant throughout the year. The averaged difference over 10 km is about -1.7 ppm in 2017 and -2.9 ppm in 2018. The significant differences for CH4 in the stratosphere can be seen in the early year when comparing CAMS with in-situ AirCore profiles. Two obvious differences occur on April 21 and 26, whose plots are highlight with additional red and green dots in the right corner of Figure  profiles increases with height and reaches up to 0.13 ppm at 21 km, however, CAMS show an underestimation at higher levels, with a peak value of 0.15 ppm at 27 km.
Despite the remaining discrepancies, CAMS CH4 profiles approximate the true state of the polar atmosphere considerably better than the MAP profiles used as a-priori for TCCON. 230

Comparison of COCCON and TCCON datasets with different a-priori profiles
When directly comparing the measurements of different remote sounders, it is necessary to account for differing observing systems characteristics, particularly the a-priori profiles used and the different sensitivity characteristic (Rodgers and Connor, 2003). In the following, we discuss the impact of the a-priori profile choice. Figure 4 shows the comparison of XCO2 and XCH4 between COCCON and co-located TCCON as a reference in Sodankylä 235 in 2017 and 2018. Since the same a-priori profiles are used, the differences between these two datasets are mainly from the different smoothing error characteristics. The partial column sensitivities of TCCON and COCCON both are imperfect and differ from each other, the kernels are shown in Figure 5 in the work by Hedelius et al. (2017). Therefore we expect that a more realistic a-priori profile will bring the results in better agreement. The left panel of Figure  When using CAMS profiles as the a-priori information, COCCON data show better correlations with TCCON data than using MAP a-priori profiles, especially in XCH4. This is mainly because CAMS profiles have better seasonal variations, especially for CH4. A significant bias of XCH4 in April 2017 and in March 2018 were found when using MAP a-priori profiles, 250 which is mainly caused by the polar vortex (see Figure 3). The stratospheric subsidence was not included in the MAP profiles, resulting in high biases. However, these biases disappeared in the data comparison when using CAMS profiles and the correlation improved due to the better modeled profile information from CAMS. Ostler et al. (2014) investigated the stratospheric subsidence caused by the influence of the polar vortex and found different impacts on mid-infrared and nearinfrared retrievals because of the differing sensitivity depending on the altitude, although the same a-priori VMR profiles were 255 used. Here, a similar mechanism is at work, the different sensitivities between TCCON and COCCON generate different smoothing errors. The more realistic CAMS a-priori information reduces these discrepancies. The COCCON data discussed below are using the CAMS profiles as a-prior profiles. https://doi.org/10.5194/amt-2020-19 Preprint. Discussion started: 12 February 2020 c Author(s) 2020. CC BY 4.0 License.

XCH4
The correlation of XCH4 between CAMS and COCCON measurements is more scattered than that of XCO2 (see Figure 6,

Effects of albedo and viewing zenith angle on XCH4 300
The officially released S5P data also contains other parameters, like albedo retrieved in the same SWIR region and viewing zenith angle (VZA). The sensitivities of the ratio of XCH4 (S5P measurements divided by COCCON) to albedo and VZA at each site are presented in Figure 8. period. The sensitivity analysis shows that there are negligible changes in measuring XCH4 when VZA changes.

Comparison of gradients measurement at two sites between CAMS/S5P and COCCON
To study the capability to measure the gradients of XCO2 (ΔXCO2) and XCH4 (ΔXCH4) on reginal scales of both S5P and COCCON, the differences of XCO2 and XCH4 with respect to CAMS between Kiruna and Sodankylä are presented in Figure   9 and Figure 10. As S5P does not measure CO2, the ΔXCO2 results are shown only for CAMS vs COCCON. 310 The ΔXCO2 comparison between CAMS and COCCON show a much poorer correlation (R 2 = 0.3322) than the comparison of XCO2 between two sites (R 2 = 0.9643, mean value of both sites over the whole measurements), as to be expected: the ΔXCO2 signals are very small (on the order of 0.5 ppm). Still, a positive correlation in ΔXCO2 and similar amplitudes are found in CAMS and COCCON data. If the comparison would be dominated either by horizontal smoothing effects due to the limited resolution of the model (would reduce the spread along the y-axis of Figure 10 or by the uncertainties of the COCCON 315 measurement (would amplify the spread along the x-axis of Figure 10, the variability ranges would differ significantly.
For ΔXCH4 the comparison between CAMS and COCCON measurements (Figure 10 left panel) shows a better correlation (R 2 = 0.4117) than that between S5P and COCCON (Figure 10 Figure 10, middle panel shows the agreement between CAMS and COCCON for the subset of days with S5P observations. The correlation in the restricted days is similar to the correlation between S5P and COCCON. CAMS, COCCON and S5P seem to be able to detect methane gradients on regional scales. Table A. 1 lists the statistics of S5P data coincident with COCCON data when S5P overpasses 325 both sites in one day.

Conclusions
In this study, two COCCON instruments are used to perform multi-year measurements at Kiruna and Sodankylä. The instruments demonstrate useful performance for accuracy, measuring gradients of column-averaged greenhouse gas abundances on regional scales. We first compared the profiles derived from CAMS with the TCCON official profiles (MAP). 330 For CO2 vertical profiles, both CAMS and MAP present similar seasonal variations, though CAMS profiles show higher vertical variability and more obvious seasonal changes over the whole time period of analysis. The main differences between them dominate in the troposphere, with peak-to-peak variability of about 25 ppm. However, the CH4 profiles derived from CAMS show a significant seasonal change, especially in the stratosphere, while MAP is more constant and overestimated over the whole year. The CH4 difference reaches up to 1 ppm at around 25 km height in April 2018. The AirCore balloon launches 335 were performed as an auxiliary activity during the Finland campaign. CAMS profiles show a better agreement with the in-situ measurements derived from AirCore launches than the official TCCON MAP a-priori profiles. Especially, CAMS presents better profiles for CH4 in April, while the MAP profiles do not show the stratospheric subsidence caused by the polar vortex.
MAP and CAMS profiles are used as apriori information in processing COCCON and TCCON data at the Sodankylä and Kiruna sites. The correlation between COCCON and TCCON data improved for both XCO2 and XCH4 when using CAMS a-340 priori profiles. R 2 increased to 0.9925 in 2017 and 0.9863 in 2018 for XCO2 and 0.9708 in 2017 and 0.9635 in 2018 for XCH4.
The obvious biases in April 2017 when comparing COCCON to the TCCON data (using MAP profiles) is mainly caused by the polar vortex. However, these outliers disappeared in the data comparison when data are processed with CAMS profiles.
Different instrument show different sensitivity to the a-priori profiles and the CAMS profiles might be a good choice to improve the data accuracy. 345 We also compared XCO2 and XCH4 between COCCON and CAMS and XCH4 between COCCON and the S5P satellite in When studying the possibility of measuring gradients of XCO2 and XCH4 for the region between Kiruna and Sodankylä, we compared the COCCON results with CAMS and S5P (only for XCH4). For ΔXCO2 CAMS show higher values and has a R 2 value of 0.3322. For ΔXCH4 COCCON shows a better correlation with the CAMS (slope = 0.6482, R 2 = 0.4117) than with the S5P (slope = 0.5791, R 2 = 0.2078). When limiting the COCCON and CAMS data to the S5P overpass days, the correlation of ΔXCH4 between them decreased (slope = 0.5304, R 2 = 0.2242) and is close to the correlation between S5P and COCCON. 360 The lower correlation between COCCON and S5P results is probably due to the smaller dataset. COCCON observations can be used for the quantification of sources and sinks of greenhouse gases and for the validation of space borne observations. Data availability. The data is accessible by contacting the corresponding author (qiansi.tu@kit.edu). 365      Qiansi Tu, Thomas Blumenstock, Pauli Heikkinen, Rigel Kivi and Uwe Raffalski took an active part in the field campaign by 440 operating the COCCON spectrometers and collecting data. Rigel Kivi and Pauli Heikkinen also operate the TCCON station at Sodankylä site and provided data. Jochen Landgraf, Alba Lorente and Tobias Borsdorff offered technical support in analyzing S5P satellite data. Huilin Chen provided the AirCore data. All authors discussed the results and contributed to the final manuscript.

445
Competing interests. The authors declare that they have no conflict of interest.

Acknowledgements. 450
We thank Xiaobo Yang in the Copernicus User Support Team at ECMWF providing the CAMS model data, which were