Investigation of space-borne trace gas products over St. Petersburg and Yekaterinburg, Russia by using COCCON observations

This work employs groundand space-based observations, together with model data to study columnar abundances of atmospheric trace gases (XH2O, XCO2, XCH4, and XCO) in two high-latitude Russian cities, St. Petersburg and Yekaterinburg. Two portable COllaborative Column Carbon Observing Network (COCCON) spectrometers were used for continuous measurements at these locations during 2019 and 2020. Additionally, a subset of data of special interest (a strong gradient in XCH4 and XCO was detected) collected in the framework of a mobile city campaign performed in 2019 using both 20 instruments is investigated. All studied satellite products (TROPOMI, OCO-2, GOSAT, MUSICA IASI) show generally good agreement with COCCON observations. Satellite and ground-based observations at high latitude are much sparser than at low or mid latitude, which makes direct coincident comparisons between remote-sensing observations more difficult. Therefore, a method of scaling continuous CAMS model data to the ground-based observations is developed and used for creating virtual COCCON observations. These adjusted CAMS data are then used for satellite validation, showing good agreement in both 25 Peterhof and Yekaterinburg cities. The gradients between the two study sites (ΔXgas) are similar between CAMS and CAMSCOCCON data sets, indicating that the model gradients are in agreement with the gradients observed by COCCON. This is further supported by a few simultaneous COCCON and satellite ΔXgas measurements, which also agree with the model gradient. With respect to the city campaign observations recorded in St. Petersburg, the downwind COCCON station measured obvious enhancements for both XCH4 (10.6 ppb) and XCO (9.5 ppb), which is nicely reflected by TROPOMI observations, 30 which detect city-scale gradients of the order 9.4 ppb for XCH4 and 12.5 ppb XCO, respectively. https://doi.org/10.5194/amt-2021-237 Preprint. Discussion started: 7 September 2021 c © Author(s) 2021. CC BY 4.0 License.


Introduction
Since human beings exist on the Earth's surface, their activities have deteriorated the environment in several manners. The increase of the global population, the globalization of the economy, the growing industry and the transport sector are only some of the most important causes, which has increased the anthropogenic emission. These activities require the use of huge 35 amount of energy, among which the fossil fuels such as coal, oil and natural gas are the main sources since the industrial era.
Global warming is one of the most discussed negative effects caused by the anthropogenic emissions of GHGs. The effect is caused by the anthropogenic emissions of greenhouse gases (GHGs), mainly carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O). These gases absorb part of the infrared emission of the Earth, corresponding to their molecular structure.
Consequently, the Earth's surface temperature increases, resulting in melting of glaciers and the Greenland and Antarctic ice 40 sheets, sea level rise, droughts, and other negative effects. Global warming leads to a climate change which, in turn, leads to a disruption in the hydrological cycle, resulting in unpredictable weather patterns. Therefore, huge efforts are needed on all levels: local, national and global are required in order to slow down the GHGs emission tendency. Such efforts require not only a panel of scientists and engineers but also politics and policy/decision makers for implementing effective measures. On that regard countries have debated since more than three decades, and such meetings produced several important agreements. 45 In 1992, the first global deal that focused on climate change was created: the UN Framework Convention on Climate Change (UNFCCC), which established the annual Conference of the Parties (COP). Based on this meeting the Kyoto Protocol and the Paris Agreement were created. The first one began on 2005 and its main aim was committing industrialized economies to reduce the emission of GHGs for defined and agreed targets. Unfortunately, after more than one decade the global anthropogenic emissions of GHGs continued increasing (Harris et al., 2012). The second one came into force on November 50 2016, which aims to limit the global warming below 2 °C or even below 1.5 ° C. Such objective can be only possible through reducing the GHGs emitted into the atmosphere. Although the majority of cities have enacted initiatives to measure and control pollution, the majority of developed interventions are localized (Miller et al., 2013;Seinfeld and Pandis, 2016). In general, the governments of most countries globally have failed to enact effective measures of addressing anthropogenic pollution (Roger et al., 2016). 55 In summary, we need to know more about the natural sources and sinks of GHGs into the atmosphere to better understand climate change, which will in turn allow better projections of their future under climate change conditions. Additionally, we need to monitor the anthropogenic emissions, e.g., in the context of the Paris Agreement. Because CO2, which is the most important GHG, is long lived. Both applications require to measure relatively small changes over a large background concentration and this is only possible with high accuracy and state-of-the-art instrumentation, which nowadays has become 60 more crucial than ever. On that framework, national and international consortiums and agencies have been measuring GHGs in the atmosphere with different sampling methods, and different spatial-vertical resolutions and accuracies. Remote sensing is one of the approaches through which GHGs can be continuously measured on a global scale. Such measurements can be made with space-based techniques by using satellites, like the SCanning Imaging Absorption spectroMeter for Atmospheric https://doi.org/10.5194/amt-2021-237 Preprint. Discussion started: 7 September 2021 c Author(s) 2021. CC BY 4.0 License. surface fluxes are imprinted in the atmospheric concentrations, in order to learn about them it is imperative to accurately estimate their respective atmospheric gradients. On that regard, the gradients for XCO2, XCH4 and XCO are calculated between 100 both studied cities during the shared measurement period. Finally, a city-scale transport event occurred during the city campaign and tracked by TROPOMI is presented in this study.

Russian Campaign location and set-up
Within the VERIFY project, two cities in Russia (St. Petersburg and Yekaterinburg) were chosen as target regions. The main aim was to collect observations for evaluating XCO2 gradients and the XCO / XCO2 ratios in a very important region with 105 high emissions and large biosphere fluxes in Eastern Europe. To achieve the foreseen objectives two different activities were carried out: a mobile city campaign (see section 2.2) and continuous measurements in two fixed locations: Peterhof (15 months) and Yekaterinburg (6 months) (see section 2.3).

Stability of the COCCON spectrometers during the campaign period
Measurements of very high precision and accuracy are required for correctly retrieving the columnar GHG abundances in the 110 atmosphere. This can be well achieved with the portable FTIR spectrometers as the EM27/SUN spectrometer. For ensuring the optimum level of accuracy, prior to the campaign, the two instruments utilized in the campaign were checked, characterized and calibrated and the residual instrument-specific calibration factors of XCO2, XCO, XCH4 and XH2O with respect to the COCCON network reference were determined. For demonstrating the stability of the spectrometers, the calibration has been redone after the campaign. This calibration work is described in Sect. 2.1.1 and 2.1.2. 115

Instrumental Line Shape (ILS) characterization
An important step in order to find any kinds of instrumental malfunction is the laboratory calibration. Open-path measurements described by  are performed for recognizing channelling effects, increased noise levels, out-of-band artefacts, and for characterizing the instrumental line shape (ILS). The ILS for both instruments was determined at KIT before and after the campaign in order to track their stability and thus, their performance. The ILS is given in terms of modulation efficiency 120 (M. E.) and phase error (Table 1).

Side-by-side measurements
After the instruments were calibrated, solar side-by-side measurements between the instruments used in the campaign (FTS#80 125 and FTS#84), the COCCON reference and the TCCON spectrometer operated at the same location were carried out at KIT.
These measurements served to find the instrument-specific calibration factors for each retrieved gas. These factors are calculated with respect to the COCCON reference and help to harmonise the results for all COCCON spectrometers. Such measurements took place before (18 and 19 April, 2018) and during (12 April, 2019) the campaign. The later one served for crosschecking whether the instruments kept the same behaviour and performance. These results can be seen in Figure 1    From the measurements shown in Figure 1, the correction factors for XCO2, XCO and XCH4 measured by the two instruments are calculated as described in . These results are averaged and later used for scaling the results for each of the retrieved GHG analysed in this study as presented in Table 2.

EMME campaign
The EMME campaign is described in detail by Makarova et al. (2020), and here we summarize only the most relevant details of it. Because the aim of this campaign was to quantify the CO2 emissions, CO/CO2 emission ratios and the estimation of the CO2, CH4 and CO fluxes, two mobile COCCON FTIR spectrometers were used in order to retrieve the required GHG species.
Both instruments were located in the up-and downwind of the St. Petersburg city ring. This campaign was not made in a 145 https://doi.org/10.5194/amt-2021-237 Preprint. Discussion started: 7 September 2021 c Author(s) 2021. CC BY 4.0 License. continuous acquisition mode but the active phases were scheduled according to the weather forecast. The basic idea is to select the deployment position of each instrument one day before good meteorological conditions appeared. The wind forecast, and the orientation of the city's NO2 plume as modelled by HYSPLIT were used as prediction tools and the positions of the COCCON spectrometers were selected accordingly. In addition, during a measuring day, the Russian partners carried out mobile zenith DOAS measurements in order to measure the NO2 total column flux over the city in a near real time manner. 150 The second input helped to readjust the location of one or both spectrometers in case of deviations from the predicted plume orientation. Following this approach, a total of 11 successful measurement days were carried out during March to April 2019.
An overview of the collected COCCON data is presented in Figure 3, from that figure is remarkable the enhancement on 25-04-19. This measurement day is presented as a plume transport event in a city-scale domain tracked by TROPOMI as complement of the results shown by Makarova et al. (2020). 155

Ground-based FTIR measurements at Peterhof and Yekaterinburg
For the continuous, long-baseline campaign, the instrument FTS#80 remained at Peterhof station at the St. Petersburg State University and continued operation there, while the other spectrometer FTS#84 was moved to Yekaterinburg. time series for COCCON XCO2, XCH4 and XCO for that day and the enhancements are all observed in the three species. It seems that these large values could be related to a plume transport from a heavily industrialized area coming from Lappeenranta 170 city, which is located in the southeast of Finland and approximately 160 km away from Peterhof. In order to confirm this,    It was planned that immediately after the EMME campaign, the instrument FTS#84 would be transported to Yekaterinburg.
Unfortunately, unforeseen organizational problems significantly delayed moving the instrument from St. Petersburg to Yekaterinburg. The instrument was finally put in operation in Yekaterinburg in October 2019 and kept measuring until the very last day before being shipped back to KIT (April 2020). The instrument was operated at the Climate and Environmental 185 Physics Laboratory INSMA of the Ural Federal University (UrFU). The instrument was set up in an internal yard of UrFU building. However, the building structure, which blocked the sunlight, was a limitation. Sometimes high trucks passing through the yard blocked the field of view of the instrument (See Figure 6). The spectrometer rested on the windowsill of the basement, so it was located exactly at ground level ~260 m. Under good weather conditions, measurements were carried out approximately between 11:00 and 14:30, local time. In total, twenty-two days of measurements were collected as it can be 190 seen in Figure 7.   (Gisi et al., 2012). A second detector channel for XCO observations was added in 2015 (Hase et al., 2016).
The EM27/SUN spectrometers are widely used and there are currently about 78 instruments globally operated by different 205 research groups. It has been shown in several studies that the results for these GHGs observed by COCCON instruments are in good agreement with official TCCON results (Frey et al., 2020;Sha et al., 2020). With the characteristics of compactness, robustness and portability, these instruments have been successfully used in several field campaigns and continuous deployments Klappenbach et al., 2015;Chen et al., 2016;Butz et al., 2017;Sha et al., 2019;Vogel et al., 2019;Tu et al., 2020Tu et al., , 2021aJacobs et al., 2020;Frey et al., 2021;). A preprocessing tool and the PROFFAST non-linear 210 least squares fitting algorithm are used for data retrieval. This processing software was created in the framework of the ESA COCCON-PROCEEDS and COCCON-PROCEEDS II projects. The solar zenith angle (SZA) range of COCCON data used in this study is restricted to ≤70° in order to limit uncertainties connected to spectra recorded at very high air-mass.

TROPOMI 215
The Sentinel-5 Precursor (S5-P) satellite with the Tropospheric Monitoring Instrument (TROPOMI) on board as a single payload was launched in October 2017. S5-P is a low Earth orbit polar satellite. It aims at monitoring air quality, climate and ozone layer with high spatio-temporal resolution and daily global coverage during an operational lifespan of 7 years (Veefkind et al., 2012). TROPOMI is a nadir viewing grating-based imaging spectrometer, measuring back-scattered solar radiation spectra with an unprecedented resolution of 7×7 km 2 (upgraded to 5.5×7 km 2 in August 2019, Lorente et al., 2021). In this 220 study, we use the improved TROPOMI XCH4 product derived with the RemoTeC full-physics algorithm  and apply the recommended quality value (qa) = 1.0 to the data. For CO, the SICOR (short-wave infrared CO algorithm) is deployed to retrieve the total column density of CO from TROPOMI spectra at 2.3μm Borsdorff et al., 2018a, b). XCO is computed by dividing the CO total column by the dry air column extracted from co-located CH4 file, which reports the European Center for Medium-Range Weather Forecast (ECMWF) pressure fields. H2O retrievals are also performed 225 with SICOR algorithm. A similar quality filter is applied to the H2O product as used in Schneider et al., 2020.

OCO-2
The Orbiting Carbon Observatory-2 (OCO-2) is a NASA satellite, launched in July 2014, providing space-based measurements of atmospheric CO2 (Eldering et al., 2017). These observations have the potential capability to detect CO2 sources and sinks with unprecedented spatial and temporal coverage and resolution (Crisp, 2015). The OCO-2 mission carries 230 a single instrument incorporated with three high-resolution imaging grating spectrometers, collecting spectra from reflected sunlight by the surface of Earth in the molecular oxygen (O2) A band at 0.764 μm and two CO2 bands at 1.61 and 2.06 μm (Osterman et al., 2020). The OCO-2 satellite has three viewing modes (nadir, glint and target) and a near-repeat cycle of 16 days (98.8 min per orbit, 233 orbits in total). It samples at a local time of about 1:30 pm. The current version (V10r) of the OCO-2 Level 2 (L2) data product, containing bias-corrected XCO2 is used in this study. 235 In addition to the operational XCO2 product derived from OCO-2 observations described above, the data product generated using the Fast atmOspheric traCe gAs retrieval (FOCAL) algorithm described in Reuter et al. (2017aReuter et al. ( , 2017b had been used. Compared with collocated TCCON observations, the OCO-2 FOCAL data show a regional-scale bias of about 0.6 ppm and single measurement precision of 1.5 ppm (Reuter and Buchwitz, 2021). In this study, the latest version (v09) covering the time period of 2015 -2020 is utilized for further comparison with the COCCON results. 240

MUSICA IASI
The Infrared Atmospheric Sounding Interferometer (IASI) is a payload on board the EMETSAT Metop series of polar orbiting satellites (Clerbaux, 2009). The IASI instrument is a Fourier Transform Spectrometer that measures infrared radiation emitted from the Earth and emitted and absorbed by the atmosphere. It provides unprecedented accuracy and resolution on atmospheric humidity profile, as well as total column-integrated CO, CH4 and other compounds twice a day. There are 245 currently three IASI instruments in operation on Metop-A, B and C, launched in 2006, 2012 and 2018, respectively. The MUSICA IASI retrievals are based on a nadir version of PROFFIT (Schneider and Hase, 2009), which has been developed in support of the MUSICA project. More details can be found in Hase (2011) andSchneider et al. (2021b). A validation of the MUSICA IASI H2O profile data is presented by Borger et al. (2018).

GOSAT 250
The Greenhouse Gases Observing Satellite (GOSAT) was launched in January 2009, equipped with two instruments (the Thermal And Near-infrared Sensor for carbon Observation Fourier Transform Spectrometer (TANSO-FTS) and the TANSO Cloud and Aerosol Imager (TANSO-CAI)) (Kuze et al., 2009). The satellite is placed on a sun-synchronous orbit and passes the same point on Earth every three days. The GOSAT is the first mission to monitor the global distribution and sinks and sources of GHGs. For this study, GOSAT FTS Short Wave InfraRed (SWIR) Level 2 data version V02.90 from National 255 Institute of Environmental Studies (NIES) is used.

CAMS inversion
Copernicus Atmosphere Monitoring Service (CAMS) is operated by the European Centre for Medium-Range Weather Forecasts (ECMWF), providing global inversion-optimised GHG concentration products which are updated once or twice per 260 year. For XCO2 and XCH4, the latest version data sets (v20r1 for XCO2 and v19r1 for XCH4) using surface air-sample as observations input are used in this study. The CAMS global CO2 atmospheric inversion product is generated by the inversion system, called PyVAR (Python VARiational) with a horizontal resolution of 1.875º×3.75º and temporal resolution of 3 hours (Chevallier, 2020a, b). The latest version (V20r1) was released in December 2020, covering the period from January 1979 to May 2020. The V20r1 model data fits TCCON retrievals well with less than 1 ppm of absolute biases (Chevallier, 2020b). 265 For XCH4 we used the latest version V19r1 based on inversion of surface observations only, covering the period between January 1990 and December 2019. The CAMS XCH4 inversion product are based on the TM5-4DVAR (four-dimensional variational) inverse modelling system (Bergamaschi et al., 2010(Bergamaschi et al., , 2013Meirink et al., 2008) with a horizontal resolution of 2 º×3º and temporal resolution of 6 hours (Segers, 2020a, b). Compared to previous releases, v19r1 data has been adjusted by using new atmospheric CH4 sinks and updated wetland emissions, and the monthly bias is usually less than 10 ppb with respect 270 to the TCCON network (Segers, 2020b).

CAMS reanalysis (control run)
This study aims to compare XCO retrieved from the COCCON measurements with XCO from different satellite and CAMS data sets as well. However, no XCO is available from the before-mentioned CAMS data. Fortunately, CAMS also provides reanalysis data sets, covering the period of 2003 -June 2020. The standard CAMS reanalysis data uses 4DVar data assimilation 275 in CY42R1 of ECMWF's Integrated Forecast System (IFS) (Flemming et al., 2017;Inness et al., 2019). The CAMS reanalysis CO profiles under a control run, i.e. without any data assimilation, is obtained from Copernicus Support team. This control run reanalysis CO profiles are using only one IFS cycle with a 0.1º×0.1°latitude/longitude resolution, 3 hours of temporal resolution and 25 pressure levels. XCO is obtained when integrating the profiles from the lowest to the highest pressure level.

Peterhof
The seasonal patterns of the retrieved GHGs are shown in Figure 8, which illustrates the time series of daily mean of XCO2, XCH4, XCO and XH2O from different data products at Peterhof. The CAMS-COCCON data product presented in Figure 8 and Figure 9 are discussed in section 4.3. The TROPOMI satellite has a higher spatial resolution and therefore, the available 285 retrieved species from TROPOMI were daily averaged within a collocation radius of 50 km around Peterhof. For the GOSAT and MUSICA IASI data sets, a collocation radius of 100 km around Peterhof is used, and for OCO-2 data, a collocation radius of 200 km is used. The measurements from the different ground-and space-based observations and model data generally show good agreements and similar seasonal variability.
COCCON XCO2 is biased low by about 0.81 -3.1 ppm in comparison to CAMS and other satellite products. GOSAT ( Figure  290 8 (a)) also shows some obvious outliers compared to the other products, which have similar behaviours. The amount of XCO2 varies along the year and much of this variation is driven by respiration, which never stops but increases between fall and winter due to reduced uptake (no photosynthesis). In this case the atmospheric XCO2 concentration is stable between January and April. It started to decrease from May to end of July, during which the growing season and the photosynthetic activities increase. Similar behaviour in 2019 was also observed by Timofeyev et al. (2021) and in previous years by Timofeyev et al. 295 (2019) and Nikitenko et al. (2020). The amount of XCO2 stays around 403 ppm between end of July and middle of September and starts to increase afterwards.
For XCH4 COCCON shows similar a behaviour as TROPOMI and CAMS. Slightly higher mean values and variability can be seen in GOSAT XCH4 with a few outliers. Compared to XCO2, XCH4 shows generally less seasonal variabilities with more short-term enhancements. The seasonal variation is comparable to the results of Gavrilov et al. (2014), Makarova et al. (2015a, 300 2015b) and Timofeyev et al. (2016). A slightly higher XCH4 is observed at the end of 2019 for all data products.
XCO shows seasonal variability with a maximal value of 110 ppb in late April and decreases by nearly 40% to 70 ppb in the beginning of July. A secondary local maximal reaching ~95 ppb occurs in August. This feature needs further investigation.
The COCCON XCO matches well to the CAMS reanalysis data. Moreover, the COCCON agrees better with the TROPOMI data in summer than in spring and late autumn, when TROPOMI measured higher values. 305 XH2O shows a strong seasonal cycle with a maximal amount of ~4700 ppm in summer and minimal amount of ~320 ppm in winter. All products show quite similar behavior with high variability, which is similar to those in Semenov et al. (2015), Timofeyev et al. (2016) and Virolainen et al. (2016. The GOSAT data have higher mean values since the measurement period covers only the time period from later spring to summer, during which higher XH2O is observed.

Yekaterinburg
The measurement period covered winter and spring, from 5 October 2019 to 17 April 2020 at Yekaterinburg (Figure 9). Here we use a larger radius (100 km) to collect the TROPOMI observations because there are much less overpasses at Yekaterinburg during this period. 315 XCO2 shows a clearly increasing tendency from October of 408 ppm to a maximal value of 415 ppm in the middle of February, which covers later autumn and winter. This is because on top of the increase due to the anthropogenic emissions there are variations due to the photosynthesis and respiration (https://atmosphere.copernicus.eu/carbon-dioxide-levels-arerising-it-really-simple, last access: 2 July 2021). During that period the plants notably reduce or stop the photosynthesis processes which could increase the amount of CO2 in the atmosphere. Later this maximal value stays constant until mid of 320 March. It tends to decrease and a similar behavior is observed in Peterhof.
For XCH4, COCCON shows a good agreement with CAMS data, though there are not so many COCCON observations. XCH4 shows generally decreasing tendency but with more short-term variabilities. Such variabilities are observed in Peterhof as well. A few TROPOMI observations in October are deviating from the other two data sets and it seems that TROPOMI underestimates XCH4. This might be because most TROPOMI measurements are located in the rim of the collecting radius 325 and thus away from the location of Yekaterinburg, introducing some errors (see Figure A -5). Further, this underestimation could be due to the difficulty for retrieving CH4 in low-and high-albedo scenes .
XCO shows in general a similar behavior of XCO2, with a steady increase during late autumn and winter. It seems that the increasing behavior of XCO has an inverse relationship with XCH4. This is probably due to the fact that atmospheric CO is mainly produced by incomplete combustion of fossil fuels (Kasischke and Bruhwiler, 2002) and the oxidation of methane 330 (Cullis et al., 1983).
As expected, most of XH2O values are below 1000 ppm, similar to Peterhof in that period. This can be explained by the saturation concentration of water vapor in air, which reduces for lower temperatures.   Figure 10 to Figure 13 show the correlations between COCCON and different satellite products at Peterhof (triangle symbols) and at Yekaterinburg (dot symbols). The satellite products and CAMS generally agrees well with COCCON and the scaling factor (slope of the fitting line; intercept is forced to 0) varies from 0.9712 to 1.0842. Figure 14 illustrates the averaged bias and standard deviation of each product of the coincident Xgas (XCO2, XCH4 and XCO) values (in space-time) with respect to 340 COCCON for the available gases at both sites. In order to find the coincident COCCON data, the mean value of the observations 2 hours before and after a centralized time reference is taken. Such time reference differs for each of the products as follows: the overpass time for satellite, each of the timestamp for CAMS.

Correlation between COCCON and satellite products
The measuring period at Yekaterinburg for COCCON was mostly in winter and early spring, from October 2019 to April 2020, in which there were less sunny days. This results in less COCCON and satellite observations. There is only one 345 coincident point between COCCON and NASA operational OCO-2 ( Figure 11 (c)) and no coincident between COCCON and OCO-2 FOCAL and GOSAT products at Yekaterinburg. Even a much larger collection circle with a radius of 100 km is used for TROPOMI at Yekaterinburg, the coincidence measurements are lesser than those in Peterhof, where more than one year of measurements were performed.    At Peterhof OCO-2 FOCAL XCO2 data have the lowest bias with respect to COCCON, while GOSAT data show the highest bias and standard deviation (3.1 ppm ± 2.9 ppm, Figure 14). NASA operational OCO-2 and CAMS show similar biases.
CAMS, TROPOMI and GOSAT measure higher XCH4 than COCCON, among which GOSAT has the highest biases at 370 Peterhof. The high negative bias in TROPOMI at Yekaterinburg is mainly due to the underestimation of the TROPOMI product in October, 2019. At both sites TROPOMI XCO shows higher biases than CAMS with respect to COCCON, which can be

Using CAMS model fields for upscaling COCCON observations
Unfortunately, during the continuous campaign carried out at Peterhof and Yekaterinburg, there are just a few coincident 380 measurement days with satellite observations, especially in comparison with GOSAT and OCO-2 (see Figure 14). Although these satellites offer a global coverage, for our measurement period even with quite relaxed coincidence criteria, the comparisons do not use the majority of the ground-based observations. This is especially the case in Yekaterinburg during the observations from October 2019 to April 2020, i.e. GOSAT and OCO-2 have none or just a couple of measurements in winter and early spring period at high latitudes. Even in Peterhof where more than one year of measurements were taken, the 385 coincident measurements between the aforementioned satellites are rather few.
For that reason, we employ a novel method, which uses model fields for upscaling the ground-based FTIR measurements, thereby generating additional virtual coincidences. Such upscaling does not use one global scaling factor, but a time resolved one, as it is shown in Figure A consideration. Since the model considers all relevant aspects of dynamics (advection, changes in tropopause altitude) and attempts to even reproduce abundance changes due to sources and sinks, we expect that our approach is superior to ad-hoc 395 schemes typically used for enlarging the colocation area (as, e.g. using the potential temperature, see Keppel-Aleks et al., 2011). In order to avoid circular reasoning in the validation based on the adjusted model fields, the method should avoid model simulations which include the assimilation of satellite data.

Generation of the CAMS fields adjusted to COCCON observations
CAMS inversion results with surface air-sampled observations as input had been used for XCO2 and XCH4 (Segers, 2020a). 400 Unfortunately, no XCO is available on that model run. No XCO product from CAMS disable us to compare one of the main data product of S5-P (XCO), which offers a greater number of measurements with a high horizontal resolution in comparison with any other satellites. Instead, the CAMS team has provided special profiles of CO from CAMS reanalysis data (control run). On that run two important points have to be mentioned: (1) no total columns for CO2 and CH4 were available from this special data set and (2) no satellite data had been assimilated. Such results are available on a daily basis as described in Table  405 3. CAMS inversion is available on a daily basis for XCO2 and XCH4 but with different time frames. Unfortunately, there are no XCH4 results from CAMS for 2020, which adds a new constraint when simply comparing both results, especially for Yekaterinburg where approximately four out of six months were measured in 2020.  As explained before, the main idea is to adjust XCO2, XCH4 and XCO from CAMS by using COCCON results. This is achieved by performing a time-resolved scaling of the model data, which is informed by the available ground-based observations. The detailed workflow encompasses these steps: 1. As shown in Table 3, CAMS XCO2 and XCH4 are available on a daily basis in different prescribed time frames, while COCCON results are only available when specific conditions were fulfilled: good weather conditions (sunny or 415 almost sunny conditions), no mobile campaign and manpower available to start the measurements because the instruments were manually operated. These conditions made the measurements rather sparse but nevertheless there still is a significant number of measurements available. Therefore, the first step is to find the coincident days between CAMS and COCCON and then the COCCON results are averaged around each CAMS time if available. As the COCCON observations require sunlight, all CAMS points before 06:00 UTC and later than 18:00 UTC were filtered 420 out. For the aforementioned each averaged CAMS time was considered as reference and all the COCCON results ± 2 hours were averaged as the coincident data. After these steps, we have both results on the same time gridding.
2. The output from the first step are time series with coincident measurement days and time frames. These time series, which have the same date boundaries, are then divided into n smaller intervals or sub-windows. These sub-windows have the characteristics of being non-overlapping and they form equally sized bins on the time axis, as defined in the 425 Eq. 1. The user only needs to define the number of sub-windows "n".
Eq. 1 3. Additionally, a sliding-sub-window, with the same size described in step 2, is run over both time series with the main difference of being shifted by half of the size of the initial sub-window but still being not overlapping between them. Therefore, after step 2, the step 3 is done in order to look at the neighbours.
4. In each of these sub-windows (described above: step 2 and 3) a correlation analysis is carried out independently of 430 the other sub-windows. In order to make the COCCON time series adjust better to CAMS results, a linear correlation with the intercept forced to zero is carried out and therefore the slope gives the scaling factor for the CAMS data. 5. Each sub-window defined in step 2 is taken as a base with its slope calculated in step 4. After that, the slopes in the neighbourhood are also calculated in each overlapping sub-window defined in step 3, Finally, all the slopes are then averaged. Such averaged slope represents the scaling factor in that sub-window. It is important to mention that this 435 number of sub-windows (and then its size) was adjusted until good results were achieved as described below.
6. Finally, with the scaling factor calculated in step 5, the original CAMS fields keeping their original temporal sampling are scaled in the whole range of each sub-window.

Selection criteria for the best number of windows
In order to choose the best number of windows, the scaling code is run starting from windows=1 and stops when two different 440 conditions are fulfilled: 1. The Root-Mean-Square-Deviation (RMSD), which is calculated with the Eq. 2 between COCCON and the CAMS-COCCON data, must be the lowest possible.
Eq. 2 2. The number of measurements points in each of the windows must be larger than four.

Verification of the method
In order to test the method before it is applied to the study area, a much denser dataset in the COCCON network is used to proof its performance. Two years of measurements (January 2018 -December 2020) taken in Karlsruhe with the instrument FTS#37 which is the reference in COCCON were selected for this purpose. For the sensitivity study, three different sub-sets were generated from the original dataset. Such sub-sets consist of a percentage (40%, 60% and 80%) of the total amount of 455 measurement days, which are randomly selected. This is done in order to simulate the reduced number of observations available in the study area. The GHG used for this short sensitivity study is XCH4 because a comparison between each of the scaling results (for each dataset) can be compared with TROPOMI as well. The main results of this verification exercise are presented in the Figure  that some seasonal variation in the modelobservation difference can be corrected as well. Note that we do not require in our approach that the COCCON values are superior over the CAMS values. This test is performed to clarify that the CAMS fields 470 adjusted in the manner we described before provide the best prediction for what COCCON would have observed on a certain date.

Combined data results by using the scaling method
The scaling method described above is applied to XCO2, XCH4 and XCO at Peterhof and Yekaterinburg. The number of selected windows for XCO2, XCH4 and XCO was 11, 10, 11 at Peterhof and 5, 2, 4 at Yekaterinburg, respectively. These 475 scaled results are then compared with all the available satellite products as described in this study.

Peterhof
After using the scaling method, the COCCON-adjusted CAMS data show close agreement with COCCON for XCO2, XCH4 and XCO (see Figure A-3).
The CAMS-COCCON data fill the gap during the measurements, providing a continuous period of a new intermediate or 480 combined (CAMS-COCCON) data product, which helps to have more coincident data with satellites observations. Figure 16 to Figure 18 show the CAMS-COCCON data in comparison to the available observations from different satellite products.
There are more coincident data points for the operational OCO-2 product than OCO-2 FOCAL XCO2, which could be because the OCO-2 product has approximately three times more soundings (https://climate.esa.int/sites/default/files/ATBDv1_OCO2_FOCAL.pdf, last access 2 July 2021). However, their correlations 485 and patterns are quite similar, whereas OCO-2 FOCAL shows better agreement with CAMS-COCCON data. GOSAT XCO2 has a similar correlation with CAMS-COCCON as found for OCO-2 data but with some outliers. For XCH4, the CAMS-COCCON are mostly higher than TROPOMI but lower than GOSAT. The CAMS-COCCON XCO agrees well with TROPOMI data with a R 2 of 0.9968.

Yekaterinburg
The scaled data are much more important in Yekaterinburg because in this city there are just a few coincident measurement 500 days between COCCON spectrometer and satellite results, mainly because of the season of the measurements taken in winter and spring. That makes a real challenge in finding the best number of sub-windows to better adjust COCCON to CAMS results, which is rather small (between 2 and 3). Nevertheless, as it can be seen in Figure A-4, the CAMS-COCCON data agree better with the coincident COCCON observations, which indicates that the scaling improves the compatibility of CAMS data with COCCON, although the amount of sampling points is extremely small. 505 The correlations between the CAMS-COCCON and the OCO-2 and TROPOMI data are presented in Figure 19. There are not too many coincident data points than those at Peterhof due to the lesser COCCON and satellite observations and mostly poor weather condition in winter. The COCCON measurement ended in 17 April 2020. Here we use a larger radius (100 km) to collect TROPOMI data for coincident COCCON observations. The averaged biases between satellite products with respect to CAMS-COCCON are presented in Figure 20. Table 4 summarized selected biases and standard deviation of satellite products compared to  Here, only when the coincident data between satellite observations and COCCON and CAMS-COCCON are both available (at least at one site), are shown. For XCO2 the biases decrease slightly when OCO-2 is compared with COCCON and to CAMS-COCCON. The absolute bias between TROPOMI XCH4 and CAMS-COCCON increased mostly twice at both sites in comparison to the direct TROPOMI XCH4 to COCCON comparison. The increased low bias at Peterhof is mainly driven by the TROPOMI outliers in April (Figure 8 (b)). The increased low bias at Yekaterinburg is due to the fact that the CAMS-520 COCCON data are only available up to end of 2019 and all TROPOMI data in autumn 2019 are biased low (Figure 9 (b)). For XCO the bias increased slightly at Peterhof and decreased by nearly half at Yekaterinburg when using CAMS-COCCON as the reference instead of COCCON at both sites.

Gradients between Peterhof and Yekaterinburg 530
The gradients (ΔXgas) are the difference of each products between two sites during the same time period. The gradients between Peterhof and Yekaterinburg (Peterhof-Yekaterinburg) are presented in Figure 21. The measuring time of COCCON at Yekaterinburg is less than that at Peterhof. We therefore use monthly means at each site to compute the gradients. A collecting circle with a radius of 100 km is used for TROPOMI at both sites. The coincident measurement days at both sites start from October 2019 until April 2020. 535 For XCO2 the gradients between COCCON at both sites are mostly negative and lower than those of CAMS and CAMS-

St. Petersburg city emission transport event tracked by TROPOMI
The results of the EMME campaign are in detail described and analysed in Makarova et al., (2020) and Ionov et al. (2021), nevertheless none of these studies performed any satellite comparison so far. Therefore, in this sub-section we show how a satellite with a high temporal and spatial resolution can measure and track a large transport of pollutants in a megacity like St.
Petersburg. During EMME campaign, we have been lucky to have the overpassing of the TROPOMI satellite during one of 550 the days with strong transport gradient as presented in Makarova et al. (2020). Such results are presented in Figure 22, which illustrates the XCH4 and XCO observations on a sample day on April 25, 2019 when the wind flowed from northeast to east before noon. The coincident TROPOMI data are the mean value collected within a circle of 15 km radius. The downwind COCCON instrument FTS#84 measured significant enhancements of XCH4 and XCO around 9:00 UTC. The higher XCH4 measured by FTS#84 than that by FTS#80 is later observed by TROPOMI as well at 10:40 UTC, though the absolute values 555 are lower in TROPOMI than the corresponding COCCON observations. When comparing observations at two locations, the difference between them at 10:40 UTC is about 10.6 ppb measured by COCCON and 9.4 ppb by TROPOMI (Figure 22 -(e)). For XCO, TROPOMI observes higher values than COCCON. The difference between two locations at 10:40 UTC is 9.5 ppb for COCCON and 12.5 ppb for TROPOMI. The increase of XCO at FTS#80 location measured by COCCON can also be detected by TROPOMI, as it increased from 107.0 ppb to 115.7 ppb. 560 All the data products in Peterhof show similar seasonal variability. However, for XCO2, the COCCON data set is generally lower than the other available data sets among which GOSAT has a highest standard deviation than the other datasets.
TROPOMI observes slightly lower XCH4 but slightly higher XCO than the other products. The largest seasonal variability is seen in XH2O. Higher amounts of XH2O are observed in summer mostly due to higher evaporation and precipitation, which is expected. The averaged GOSAT XH2O is higher than the other products due to its short measurement period, which is mostly 575 in summer. There is shorter measurement period in Yekaterinburg, covering mostly winter and spring, from October 2019 to April 2020. Similar seasonality and concentrations are observed to that in Peterhof at the same time period.
The satellite observations are sparser in the high latitude regions than in mid and low latitude regions, while models provide continuous data sets. The ground-based COCCON observations have been proved to be highly accurate by many previous studies. To combine the advantages of CAMS and COCCON data sets, we developed an upscaling method by adjusting CAMS 580 data to the COCCON observations collected at Peterhof and Yekaterinburg to obtain a continuous data of virtual COCCON observations (as demonstrated using different sub-sets of COCCON measurements at Karlsruhe). This method is more important for Yekaterinburg, where we face three different problems: 1. less amounts of measurements in general (around 6 months compared to 15 months in Peterhof), 2. less measurement days per month (mostly in winter), and 3. shorter daily period of measurements. As expected, the CAMS-COCCON data show better correlations with COCCON observations than the 585 original CAMS data sets. The CAMS-COCCON data are then compared with satellite products, showing good agreements as well and generally similar biases to that between satellite products and COCCON observations. This method was also used for the observations at Yekaterinburg where less COCCON measurements were taken. The gradients between the two study sites (ΔXgas) are similar between CAMS and CAMS-COCCON data sets. There are a few COCCON and satellite ΔXgas measurements, fitting well to that of CAMS-COCCON. These results presented in this study indicate that our scaling method 590 is working reliably.
In addition, the XCH4 and XCO observations recorded during one of the mobile city campaign days (April 25, 2019) was analyzed. In the city campaign, two COCCON instruments were set up in the upwind and downwind sites and the wind flowed from northeast to east before noon on the sample day. The downwind COCCON instrument measured obvious enhancements in both XCH4 (10.6 ppb) and XCO (9.5 ppb), which is also observed by TROPOMI (9.4 ppb in XCH4 and 12.5 ppb XCO, 595 respectively).