Quantifying hard coal mines CH4 emissions from TROPOMI and IASI observations using high-resolution CAMS forecast data and the wind-assigned anomaly method

Intensive coal mining activities are in the Upper Silesian Coal Basin (USCB) in southern Poland, resulting in large amounts of methane (CH4) emissions. Annual CH4 emission reached to 448 kt according to the European Pollutant Release and Transfer Register (E-PRTR, 2017). As a CH4 emission hot spot in Europe, it is of importance to investigate its emission sources and accurate emission estimates. 15 In this study, we use satellite-based column-averaged dry-air molar fraction observations of CH4 (XCH4) from the TROPOspheric Monitoring Instrument (TROPOMI) and tropospheric XCH4 (TXCH4) from the Infrared Atmospheric Sounding Interferometer (IASI), together with the high-resolution model forecast XCH4 and TXCH4 from the Copernicus Atmosphere Monitoring Service (CAMS) to estimate the CH4 emission rate averaged over three years in the USCB region (49.3° 50.8° N and 18° 20° E). Using the CAMS inventory as the a priori knowledge of the sources, together with ERA5 20 wind at 330 m, the wind-assigned XCH4 anomalies for two opposite wind directions are calculated, which yields an estimated CH4 emission of 9.6E26 ± 1.4E25 molec./s for CAMS XCH4 and 9.1E26 ± 1.2E25 molec./s for CAMS TXCH4. These values are very close to the total emission of the CAMS inventory (9.7E26 molec./s). Very good agreements between CAMS and the wind-assigned model results (R2=0.89 for XCH4 and TXCH4) indicate that our wind-assigned method is quite robust. The similar estimates of XCH4 and TXCH4 also imply that for a strong source, the dynamically induced variations of the CH4 25 mixing ratio in the upper troposphere and lower stratosphere region is of secondary importance. This wind-assigned method is further applied to the TROPOMI XCH4 and TROPOMI+IASI TXCH4 with using the Carbon dioxide and Methane (CoMet) inventory performed in 2018. The calculated averaged total CH4 emission over the USCB region is about 5.7E26 ± 4.9E24 molec./s for TROPOMI XCH4 and 5.2E26 ± 2.2E25 molec./s for TROPOMI+IASI TXCH4. These results are very close and thus in agreement to the emissions given in the E-PRTR inventory (5.33E26 molec./s) and the CoMet 30 inventory (6.6E26 molec./s). https://doi.org/10.5194/acp-2022-41 Preprint. Discussion started: 11 February 2022 c © Author(s) 2022. CC BY 4.0 License.

00:00 UTC. Here we use the daily averaged CAMS forecasts during 9:00 UTC -18:00 UTC at each resolution grid point. The 95 corresponding standard deviation (STD) is considered as the noise/error: where XCH + , is the CAMS XCH4 (or CAMS TXCH4) in each resolution grid at each time step, XCH + ((((((( is the daily average (9:00 UTC -18:00 UTC), and n is the number of CAMS forecasts of each day. The time resolution of CAMS forecasts is 3h and thus, n = 4.
The CAMS-GLOB-ANT inventory is based on the emissions provided by the EDGARv4.3.2 inventory for the time period 2000(Crippa et al., 2018 and linearly extrapolated to 2020 using the trends from the CEDA global inventory in 2011-2014 (Hoesly et al., 2018). The latest version (CAMS-GLOB-ANT v4.2) was released in March 2020, using the same set-up as v4.1 except for adding the emissions in 2020. The anthropogenic sources in the standard v4.2 are divided into 12 sectors 105 and the agriculture sections are split into three sectors, including livestock, soils and waste burning (https://eccad3.sedoo.fr/, last access: 27 October, 2021). The inventory is provided as monthly mean with the same spatial resolution (0.1° × 0.1°) as the CAMS forecast data (Granier et al., 2019).
The monthly averages of the CAMS global anthropogenic emissions for different sectors in the study area of USCB are presented in Figure 1. The emissions from the sectors "agriculture soils" and "solvents" are zeros, and emissions from ships 110 with a magnitude of 19 are much lower than the other sectors. Thus, these three sectors are not shown here. The fugitive sources tend to be from energy production and distribution (e.g. fuel use) and are the dominant CH4 sources in this region with a mean value of 7.9E26 molec./s and a standard deviation of 2.2E25 molec./s. Compared to its high amount, the seasonal variations of the fugitives sector can be ignored. Though the sources from agriculture livestock (1.7E25 ± 4.0E25 molec./s) show an obvious seasonal cycle, these amount only 4% of the total emissions in this region. Whereas the CH4 emitted from 115 the fugitives sector occupies 82%. The spatial distribution of the CH4 inventory of the CAMS-GLOB-ANT from all anthropogenic sources and from fugitives are quite similar in the USCB region ( Figure 2). Therefore, we apply the three-year mean of total emissions at grids with significant emissions without considering seasonal variations in the simple plume model (see Sect. 2.3). The total emissions amount to 9.7E26 molec./s over this study area.

TROPOMI and IASI data sets
The TROPOMI instrument is a nadir-viewing, imaging spectrometer, which uses passive remote sensing techniques to perform measurements of the solar radiation reflected by and radiated from the earth in the ultraviolet, the visible, the near-infrared and the shortwave infrared spectral bands (Veefkind et al., 2012). The algorithm for CH4 column retrieval is called RemoTeC algorithm and it has been extensively used to derive CO2 and CH4 retrievals from Greenhouse Gases Observing Satellite 130 (GOSAT) and Orbiting Carbon Observatory-2 (OCO-2; Boesch et al., 2011;Butz et al., 2009Butz et al., , 2011Hasekamp and Butz, 2008;Schepers et al., 2012). An updated retrieval algorithm has been implemented by Lorente et al. (2021) to obtain a data suit with less scatter and a higher resolution surface altitude database. This updated TROPOMI XCH4 dataset has been validated with the Total Carbon Column Observing Network (TCCON) (-3.4 ± 5.6 ppb) and GOSAT (-10.3 ± 16.8 ppb), showing very good agreements. In this study the TROPOMI XCH4 during November 2017 and December 2020 within the 135 study area over the USCB region is investigated. The data provided by Lorente et al. (2021) includes an additional quality filter parameter (quality value, q). TROPOMI XCH4 with qa=1.0 represents the data under clear-sky and low-cloud atmospheric conditions and the problematic data points are removed as well, which is applied in this study.
The IASI instrument is a nadir viewing Fourier-transform spectrometer that measures the infrared part of the electromagnetic spectrum. IASI measurements are performed with a horizontal resolution of 12 km and a full swath width of about 2200 km 140 on the ground. It is the key payload element of the polar-orbiting Metop-A -B and -C satellites. These satellites overpass the equator at 09:30 in the morning and 21:30 local time in the evening with a little more than 14 orbits per day. It provides unprecedented accurate vertical information of the atmospheric temperature and humidity, which helps to improve numerical weather prediction (NWP). The thermal infrared nadir spectra of IASI have been successfully used in retrieving different atmospheric trace gas profiles and these retrievals are especially sensitive between the middle troposphere and the stratosphere 145 (Schneider et al., 2021). By combining these IASI profiles and the TROPOMI CH4 total column which has a higher sensitivity near ground, it is able to detect the tropospheric XCH4 (TXCH4) independently from CH4 at higher altitudes. The combined product cannot be obtained by either the TROPOMI or IASI product independently. It shows a weak positive bias of about 1 % with respect to the references (Schneider et al., 2021). We refer to this product in the following as the TROPOMI+IASI TXCH4. 150

Simple plume model and wind-assigned anomaly method
The averaged distribution of emitted CH4 over a long-term period can be modeled simply as an evenly-distributed cone-shape dispersion based on the wind and source strength. Since CH4 is a long-lived gas, its decay is negligible for short periods and not considered in the model. This model is referred to as simple plume model (see Figure 2, Tu et al., 2021). We use the model wind from ERA5, which is the fifth generation ECMWF reanalysis product using 4D-Var data assimilation and model forecasts 155 in Cycle 41R2 of the ECMWF IFS model (Copernicus Climate Change Service, C3S, 2017, Hersbach et al., 2020. It provides hourly estimates on 137 pressure levels in the vertical covering the atmosphere from the surface up to 0.01 hPa, with a spatial resolution of 0.25º×0.25º (Hersbach et al., 2020). where the emission strength is the a priori knowledge from the CAMS-GLOB-ANT data set or from the coal mine ventilation shafts in this study (see Sect. 3.2). Their emission rates are assumed to be constant with time from 2017 to 2020. is the angle of the emission cone and has an empirical value of 60°. is the wind speed from ERA5, and is the distance between the downwind location and the CH4 emission source. Each individual source either from the CAMS-GLOB-ANT inventory or from the knowledge of the ventilation shafts is considered as an individual point source. The plumes computed from different 165 point sources over daytime (8:00 UTC -18:00 UTC) are super-positioned and then averaged to a daily plume.
The wind distributions at different height levels (10 m, ~330 m, ~500 m) over the USCB region are presented in Figure 3.
The wind speed increases with increasing altitude (see Table A-1). The ERA5 wind is divided into two opposite wind regimes based on directions (e.g., 135°-315° for SW and the rest for NE). For each wind sector, an averaged plume is computed and the difference of the two plumes are therefore the wind-assigned anomalies. The estimated emission strengths can be calculated 170 by fitting the modelled anomalies to the known anomalies from e.g. CAMS XCH4/TXCH4, and TROPOMI and TROPOMI+IASI observations. Note that CH4 has a lifetime of around 12 years, which results in a high background compared to the newly emitted CH4. Thus, the contributions from the background should be removed for correctly estimating emissions (Liu et al., 2021). The background is considered to consist of a constant value, a linear increase with time, a seasonal cycle, a daily anomaly and a horizontal anomaly. For more details, see the Appendix in Tu et al., 2021. 175 This method was firstly used to estimate the CH4 emission from landfills in Madrid, Spain and yielded a CH4 emission rate of 7.4×10 25 ± 6.4×10 24 molec./s based on the TROPOMI XCH4 (Tu et al., 2021).

Estimated emissions derived from CAMS forecasts (evaluation of the method)
The CAMS forecast XCH4 data from November 2017 to December 2020 within the study area are illustrated in Figure 4 left.
The areas with high XCH4 amounts fit well with the CAMS anthropogenic CH4 emissions (square symbols). Similar to the 185 CoMet inventory, high sources in the CAMS-GLOB-ANT inventory are centered in this region, but there are other weaker sources outside. The total emission rate of the CoMet inventory is 6.6E26 molec./s, which is slightly less than the CAMS- GLOB-ANT emissions (9.7E26 molec./s). This is probably because the CAMS inventory includes more CH4 emission sources, e.g., wastes, and combustion from residential, commercial, which account to about 20%. Based on the CAMS emissions, the wind-assigned method is applied to CAMS XCH4. The XCH4 anomalies (raw-195 background) and the wind-assigned anomalies are presented in Figure 5a and b, respectively. Note, that the CAMS XCH4 is coincided with TROPOMI XCH4 for better comparison. Some data are thus missing here mostly due to the quality filter of TROPOMI observations. After removing the XCH4 background, the XCH4 anomalies well represent the CAMS sources. The highest CH4 sources from the CAMS-GLOB-ANT inventory are also obviously seen in the 2D anomalies. In addition, the spatial distributions of the three XCH4 data products show different patterns (Figure 4 (d)). This indicates that the background removal is of importance for XCH4 and our method works well.
The CAMS ΔXCH4 and modelled ΔXCH4 show a very good agreement with a slope of 0.97 and a R 2 of 0.89 ( Figure 5c).
Our results are derived from the CAMS emission information, and they agree very good with the CAMS model data. The estimated emission rate is about 9.6E26 ± 1.4E25 molec./s when using the ERA5 wind at 975 hPa (~330 m) and this value is 205 very close to the CAMS inventory (estimated emission rate at other levels are presented in Sec. 3.2). Therefore, we use ERA5 wind at this level in the following. Note that the points whose distances to the nearest dominant sources are less than 10 km, are removed here, because they are very close to the significant sources and small changes in wind (either speed or direction) can result in high uncertainties.
The retrieved CH4 from satellite observations are based on total columns and therefore, these are strongly affected by the 210 stratospheric abundance, i.e., by the changing tropopause altitude (Liu et al, 2021;Schneider et al., 2021). The model simulation uncertainties in representing XCH4 in the stratosphere might introduce biases in investigating CH4 sources and sinks (Pandey et al., 2016). To remove this influence, we calculate the tropospheric CAMS forecasts CH4 (TXCH4) from the surface up to 7 km. The results are presented in Figure 5d-f. The CAMS TXCH4 anomalies have similar distribution as CAMS XCH4, showing that background removing also works for the tropospheric CH4. The wind-assigned plume and the correlation between CAMS and the wind-assigned model results are very similar between XCH4 and TXCH4. The estimated CH4 emission strength derived from CAMS TXCH4 is 9.1E24 ± 1.2E24 molec./s.

Estimated emission derived from satellite observations
The high-resolution TROPOMI XCH4 provides the ability to detect and quantify the CH4 emissions (e.g., oil and gas sector, 225 coal mining) on fine and large scales (Pandey et al., 2019;Varon et al., 2019;Gouw et al., 2020;Schneising et al., 2020).  Three-year averaged TROPOMI XCH4 observations presented in Figure 4b shows scattered high XCH4 amounts, whereas CAMS XCH4 is more concentrated on the center of the study area, and they agree well with its anthropogenic emission sources (CAMS-GLOB-ANT inventory). This might be because TROPOMI detects other real CH4 sources that are not included in the 240 CAMS forecast model data.
For better comparison with other studies discussing the coal mine emissions in the USCB region, we apply the CoMet inventory as the a priori known sources in the wind-assigned method to estimate the CH4 emission. The results are illustrated in Figure 7. The TROPOMI XCH4 and TROPOMI+IASI TXCH4 anomalies show high amounts around the areas where the ventilation shafts are located and the region in the northeast of Katowice. Though the anomalies of the satellite observations 245 are lower than the CAMS results (Figure 5a), their spatial distributions are similar. Positive and negative plumes can be clearly seen in Figure 7b and e. The ΔXCH4 correlation between the TROPOMI and model have a very good agreement with a R 2 value of 0.72. Similar results are also derived from TROPOMI+IASI TXCH4 with a R 2 value of 0.62. Compared to CAMS data, higher scatter is expected, because satellite observations suffer from observational errors and might contain more CH4 sources (e.g., landfills, gas distribution network). Though none of these sources is at the same level of magnitude of coal 250 mining emission, they might still bring some errors.
The estimated CH4 emission strengths in molec./s are 5.7E26 ± 4.9E24 for XCH4 and 5.2E26 ± 2.2E25 for TXCH4, and both are close to the E-PRTR inventory (5.33E26 molec./s). The TROPOMI+IASI result has a slightly higher uncertainty than the TROPOMI result, because (1) the vertical distribution of CH4 is in general much more difficult to measure than the total column of CH4 and (2) the vertical distribution is derived by considering two independent measurements, each with its own 255 noise error. This might change for a larger number of data points (e.g., by using data from more years or by applying the method to IASI and TROPOMI successors on the upcoming METOP-SG satellite, which offers much more collocated observations).
However, in our study using TXCH4 data in addition to XCH4 data nicely documents the robustness of the method. Important for a correct estimation of the emission is the correct removal of the methane background signal. For XCH4 the stratospheric 260 and the tropospheric backgrounds have to be removed, whereas only the tropospheric background has to be removed for   (10 m, 330m, 500m). Square symbols represent the a priori emission sources from the CAMS-GLOB-ANT inventory and triangle symbols represent the a priori emission sources from the CoMet inventory. The two horizontal lines represent the number of total emissions for the CAMS-GLOB-ANT inventory (lavender color) and for the CoMet inventory (orange color), respectively. Note, the error bars are much smaller than the results and they are not 275 visible here. For specific values see Table A

Uncertainty analysis
Winds, particularly near the surface, are significantly altered by topography, which yields uncertainties in knowing the transport pathway from emission sources to the measurement location (Chen et al., 2016;Babenhauserheide et al., 2020). Thus, wind is one of the most important factors in correctly estimating the emission rates. Here we investigate the wind uncertainties 280 based on the CAMS XCH4 and the CAMS-GLOB-ANT inventory. The wind used in Sect. 3.3.2 and 3.3.3 are ERA5 wind at 10 m.

Vertical wind shear
Compared to the wind at 330 m, the wind distributions are similar at lower or higher altitude (10 m and 500 m) but the speed increases with higher altitude (Figure 3). Wind at 10 m is 19 % slower than that at 330 m (Table A-1), which yields a 285 corresponding lower estimates of 7.4E26 ± 1.1E25 molec./s (-23%) based on CAMS XCH4 and CAMS emission inventory ( Figure A-1a).
Considering the height of the Planetary Boundary Layer (PBL), we use the ERA5 wind at 500 m above the ground ( Figure   3c). The wind speed at 500 m increases by 26% and 37% for NE and SW sectors, respectively, compared to the wind at 330 m. The share of SW directed winds is slightly larger at the 500 m level. These differences result in an increase of 13% of the 290 estimated emission rate (1.1E27 ± 1.7E25 molec./s).

Use of narrowed angular wind regimes
The long-term wind comes from all directions (0°-360°) (Figure 3). To define the uncertainty of wind regimes' coverage, the wind is separated into two groups with narrow coverage fields: NE_narrow (0°-90°) -SW_narrow (180°-270°) and NW_narrow (270°-360°) -SE_narrow (90°-180°). The final estimated emission strength is weighted by the number of the 295 valid binning data in the plume maps under different wind regimes (i.e. 171 for narrow NE-SW and 26 for narrow NW-SE, respectively). The XCH4 anomalies and the plume for narrow NE-SW regime are quite similar to those with using widercoverage NE and SW fields (Figure 9a-c). CAMS ΔXCH4 and modelled ΔXCH4 show very good agreement as well. Slightly less data points are found here because of the choice of narrower wind fields, especially for NW-SE wind fields. The estimated emission rate is about 9.8E26 ± 1.5E25 molec./s for the narrow NE-SW field. This indicates that the effect of the section in 300 the wind field coverage is negligible when there are enough measurements. The use of narrow NW-SE wind fields yield an emission strength of 1.4E27 ± 5.40E25 molec./s. The higher uncertainty is probably due to less measurements in these wind fields. The weighted rate is therefore about 1.0E27 molec./s, 4.2% higher than based on the wider NE-SW wind regime (Sec. 3.1).

Investigation of different choices for wind field segmentation
The wind category here is based on its predominant wind fields over the USCB region and is divided into two opposite sectors 310 (SW and NE). To investigate its uncertainty, we apply another kind of segmentation: N (<90° or >270°) and S (90° -270°) categories. Similar results are found and are shown in Figure 10. Though the 2D distribution of the plume changes due to the new wind category, an obvious plume can be seen. The estimated emission rate is 9.4E26 ± 1.7E25 molec./s, which is only 2.1% less than that using NE and SW wind categories. The correlation between the CAMS ΔXCH4 and the wind-assignmodeled ΔXCH4 shows a very good agreement as well, with a similar R 2 value of 0.9 to that in the NE-SW wind category. 315 This result demonstrates that our method is not significantly influenced by the wind regime division.

Conclusion 320
Intensive mining activities are the dominant CH4 emission sources in the USCB region, Poland, where one of the largest coal mining areas in Europe is located. It is thus of importance to quantify the CH4 emissions from this area. In this study we use https://doi.org/10.5194/acp-2022-41 Preprint. Discussion started: 11 February 2022 c Author(s) 2022. CC BY 4.0 License. the combination of a simple plume model and a novel wind-assigned model to estimate CH4 emission rates from highresolution CAMS forecast XCH4 and TXCH4, along with satellite data (TROPOMI XCH4 and TROPOMI+IASI TXCH4) over the USCB region (49. 3°N-50.8°N and 18°E-20°E) from November 2017 to December 2020. 325 Based on the CAMS-GLOB-ANT inventory, the dominant CH4 source is emitted from energy production and distribution, and the significant sources are spread around the city of Katowice and its southwest region. We firstly apply the wind-assigned method to the CAMS forecasts based on the a priori knowledge of CAMS-GLOB-ANT inventory (9.7E26 molec./s in total) and ERA5 wind at ~330 m. We use ΔXCH4/ΔTXCH4 to represent the difference of XCH4/TXCH4 between the conditions of two opposite wind fields (NE and SW). The CAMS ΔXCH4/ΔTXCH4 data show very good agreements with the output of the 330 wind-assigned anomalies with a R 2 value of 0.89 for CAMS XCH4 and CAMS TXCH4. This nice correlation indicates that our background removal works well. In addition, similar estimates are derived from CAMS XCH4 (9.6E26 ± 1.4E25 molec./s) and TXCH4 (9.1E26 ± 1.2E24 molec./s).
To investigate the CH4 emissions from this hot spot, the CoMet campaign was performed in 2018. Locations and emission rates of the ventilation shafts of the coal mine used in this study are based on this inventory. Based on this knowledge, the 335 estimated CH4 emissions are 5.7E26 ± 4.9E24 molec./s and 5.2E26 ± 2.2E25 molec./s derived from the TROPOMI XCH4 and combined TROPOMI+IASI TXCH4, respectively. These results are 40% less than that derived from the CAMS model and CAMS-GLOB-ANT inventory. It is probably because the CAMS inventory includes many sectors of anthropogenic sources, like wastes, and combustion from residential and commercial, which account for about 20%. Nevertheless, our results derived from satellite observations are close to the E-PRTR inventory of 5.33E26 molec./s and reasonablely compared to the CoMet 340 inventory (6.6E26 molec./s), and to previous studies over the USCB region (ranging from 1.05E25 molec./s to 9.38E25 molec./s for a sub-clusters of shafts (Krautwurst et al., 2021) up to 5.68E26 molec./s derived from one flight (Kostinek et al. (2021)). Similar 2D anomalies and plumes are also observed for TROPOMI XCH4 and TROPOMI+IASI TXCH4. This nicely documents the robustness of the method. The TROPOMI+IASI result has a slightly higher uncertainty than the TROPOMI result, because (1) the vertical distribution of CH4 is in general much more difficult to measure than the total column of CH4 345 and (2) the vertical distribution is derived by considering two independent measurements, each with its own noise error. This might change for a larger number of data points (e.g. by using data from more years or by applying the method to IASI and TROPOMI successors on the upcoming METOP-SG satellite, which offers much more collocated observations). Nonetheless, the uncertainties are insignificant compared to the estimated emission rates.
Wind contains uncertainties in knowing the transport pathway from emission sources to the measurement location and thus, 350 we analyze the effects in selecting wind at lower and higher altitude (10 m and 500 m), wind field coverage and wind category.
Wind distributions at higher levels are similar to that at 330 m. However, their speeds decrease by 19% at 10 m and increase by 32% at 500 m, which results in higher emission rates by -23% and 13 %, respectively. Narrower wind field coverage (0°-90° for NE sector and 180°-270° for SW sector) and different wind segmentation (<90° or >270° for N sector and 90°-270° for S sector) introduce minor uncertainties of +4.2 % and -2.1 %, respectively. The agreements for these sensitivity tests 355 between the CAMS ΔXCH4 and wind-assigned model ΔXCH4 are as good as that using previous NE and SW wind fields. The results suggest that our method is robust since it is insensitive to the separation of the wind regimes. It is also suitable for estimating CH4 and CO2 emissions in other regions. Schneider, Benjamin Ertl and Christopher J. Diekmann provided the combined (MUSICA IASI + TROPOMI) data and supported technically for the analysis of these data. Jarosław Necki supported in consultation of the local situation and CoMet inventory. All authors discussed the results and contributed to the final manuscript.

375
Competing interests. The authors declare that they have no conflict of interest.
Acknowledgements. The CAMS results were generated using Copernicus Atmosphere Monitoring Service (2017)(2018)(2019)(2020) information. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains. We also thank Michela Giusti and Kevin Marsh in the Data Support Team at ECMWF for 380 retrieving and providing comments about the CAMS data. This research has largely benefit from funds of the Deutsche