Spectrometric imaging of sub-hourly methane emission dynamics from coal mine ventilation

Anthropogenic methane (CH4) emissions contribute significantly to the current radiative forcing driving climate change. Localized CH4 sources such as occurring in the fossil fuel industry contribute a substantial share to the anthropogenic emission total. The temporal dynamics of such emissions is largely unresolved and unaccounted for when using atmospheric measurements by satellites, aircraft, and ground-based instruments to monitor emission rates and verify reported numbers. Here, we demonstrate the usage of a ground-based imaging spectrometer for quantifying the CH4 emission dynamics of a ventilation facility of a coal mine in the Upper Silesian Coal Basin, Poland. To this end, we deployed the imaging spectrometer at roughly 1 km distance from the facility and collected plume images of CH4 column enhancements during the sunlit hours of four consecutive days in June 2022. Together with wind information from a co-deployed wind-lidar, we inferred CH4 emission rates with roughly 1 min resolution. Daily average emission rates ranged between 1.39 ± 0.19 and 4.44 ± 0.76 tCH4 h−1, 10 min averages ranged between (min) 0.82 and (max) 5.83 tCH4 h−1, and puff-like events caused large variability on time scales below 15 min. Thus, to monitor CH4 emissions from such sources, it requires measurement techniques such as the imaging spectrometer evaluated here that can capture emission dynamics on short time scales.


Introduction
Atmospheric methane (CH 4 ) is, after carbon dioxide, the second most important driver of anthropogenic climate change. The global warming potential of CH 4 is more than 80 times higher than that of CO 2 on a 20 year time scale (Masson-Delmotte et al 2021). Due to its comparatively short lifetime of roughly 9 years, CH 4 recently has received particular attention as a near-term climate change mitigation target since reducing its emissions would show a climate effect on decadal time scales (e.g. Shindell et al 2012, Stohl et al 2015, Nisbet et al 2020, Ocko et al 2021. Spectroscopic remote sensing can help monitor CH 4 emissions and their changes in the context of the Paris climate accord (Jacobs 2022) and the derived monitoring, reporting, and verification requirements. The fossil fuel producing sector appears particularly interesting since it accounts for more than 30% of the global anthropogenic CH 4 emissions (Saunois et al 2020), the avoidance of leakages can be of both economic and environmental interest, and the sectoral emission total is dominated by large point sources , Balcombe et al 2018.
Imaging spectrometers deployed on aircraft (e.g. Thompson et al 2015, Duren et al 2019, Ayasse 2022) and satellites (e.g. Thompson et al 2016, Varon et al 2018, Cusworth et al 2019, Guanter et al 2021, Maasakkers et al 2022 have shown to be able to detect CH 4 plumes. The plume images have been used to discover previously unknown, accidental and catastrophic CH 4 leakages and to quantify the instantaneous emission rates of individual facilities and groups of facilities in the fossil fuel and waste sectors. Repeated observations of prominent CH 4 source regions have enabled regional emission estimates and comparison to reported inventories (e.g. Frankenberg et al 2016, Rafiq et al 2020, Varon et al 2020, Zhang et al 2020, Sadavarte et al 2021. However, Bhardwaj et al (2022) point out that plume snapshots and the inferred emission estimates are blind to the dynamics of the plumes and sources. These temporal dynamics might bias the inferred emission rates for variable sources in particular.
Here, we report on the first study to directly measure the dynamics of CH 4 plumes and emissions from coal mine venting on time scales of minutes using a ground-based imaging spectrometer. The method quantifies CH 4 column enhancements by means of absorption spectroscopy in the shortwave-infrared (SWIR) spectral range, and, thus, it is similar to the technique employed for aircraft and satellites. But, upward-looking, the spectrometer collects skylight instead of surface-reflected sunlight, and its stationary deployment enables repeated imaging of a single target with a high temporal resolution instead of area mapping. In contrast to optical gas imaging (OGI) designed for natural gas leak detection on distances of meters (Zeng and Morris 2019, Zimmerle et al 2020), our method allows for remote quantification of CH 4 emission rates similar to spectroscopic cameras operating in the thermal infrared (Gålfalk et al 2016(Gålfalk et al , 2022. In contrast to the latter, our method does not require thermal contrast between the CH 4 plume and the background, and it can be deployed further away in kilometer-scale distance to the target.
The targeted coal mine is located in the Upper Silesian coal basin (USCB), Poland. The USCB is Poland's industrial heartland, and it is a hotspot of CH 4 emissions in Europe, with an emission total of more than 500 ktCH 4 year −1 (Fiehn et al 2020) emitted from tens of underground hard-coal mines, each of them typically disposing of several ventilation facilities. CH 4 accumulates in the coal bed during coal formation through the carbonification of plants. Due to the excavation activities, the pressure on the CH 4 -bearing rock reduces and CH 4 , together with other hydrocarbons, carbon dioxide, and water vapor, degases into the air in the mine. To avoid builtup of hazardous and explosive concentrations in the mine, mining companies vent CH 4 through shafts into the atmosphere, although a minor fraction is sometimes captured through drainage systems and used for on-site energy production. The CH 4 content of the hard coal seams in the USCB generally increases with depth but is highly variable across the USCB depending on the permeability of the rock and its saturation with CH 4 . Therefore, how much CH 4 is released into the atmosphere depends on the location and the particular mining activity underway (Swolkień et al 2022). The area has been subject of intense scientific study during several measurement campaigns in recent years (Luther et al 2019, Fiehn et al 2020, Kostinek et al 2021, Krautwurst et al 2021, Andersen et al 2022, Swolkień et al 2022. Individual ventilation shafts were found emitting up to a few tCH 4 h −1 .
Our demonstrator study covers four days in June 2022 for which we collected day-time images of CH 4 plumes above a ventilation shaft in the USCB and determined emission rates with roughly a minute temporal resolution. Thus, we demonstrate the performance of the imaging spectrometer setup, and we quantify the CH 4 emission dynamics to be expected from coal mine ventilation.

Hyperspectral camera and wind lidar in the USCB
The field instrumentation consists of a HySpex SWIR-384 hyperspectral camera by Norsk Elektro Optikk ® and a Windranger 200 wind-lidar by METEK ® . Both devices are commercially available.
The HySpex SWIR-384 collects incoming light on a 2D-detector array where one dimension consists of 384 pixels that contain vertical (elevation) samples of the scene and the other dimension consists of 288 pixels that contain the spectral samples (figure 1). For our ground-based operations, the camera is mounted on a rotation stage which allows for scanning of the target scene in horizontal (azimuth) direction. The camera is pointed into the sky along a shallow elevation angle, collecting incoming skylight from a vertical opening angle of 16 • . Each vertical pixel covers approximately a solid angle of 7.3 mrad × 7.3 mrad. In the spectral dimension, the HySpex records the SWIR spectrum between 950 nm and 2500 nm with a spectral sampling distance of 5.45 nm. Further information on the camera can be found in Baumgartner et al (2012) and Lenhard et al (2015).
The Windranger 200 is a compact and portable wind-lidar of approximately 50 kg in weight and dimensions of 840 mm × 540 mm × 580 mm. Its working principle is based on the frequency modulated continuous wave technology (Peters 2018) operating with a 1545 nm laser. It measures wind direction and velocity at six heights between 10 m and 200 m, taking approximately 9 s per profile. The Windranger is typically used for studying boundary layer turbulence (Adler et al 2021).
In June 2022, we deployed both instruments, the HySpex and the Windranger, in the vicinity of a coal mine ventilation shaft, termed Pniowek V (49.9754 • N, 18.7345 • E), of the Pniowek mine in the USCB. The instruments were positioned at approximately 1 km distance from the shaft, which is equipped with three ventilators, but only one of them was active (left in figure 1). At 1 km distance, the collected HySpex images have a spatial resolution of approximately 0.8 m × 0.8 m. They typically cover an azimuth range of 10 • -15 • in the horizontal, such that a single scan of the scene took roughly 60 s. Both instruments, the HySpex and the Windranger, were geometrically referenced with respect to the lineof-sight toward the ventilation shaft, i.e. an azimuth angle of 0 • corresponds to the direction of the CH 4 ventilation facility.
For the four days (17-20 June 2022) with decently fair weather conditions reported here, the camera continuously collected azimuth scans with roughly 1 min resolution from the morning into the afternoon. The field dataset of a typical day comprises of several hundred HySpex images of the target scene, collocated Windranger measurements of vertically resolved wind velocity and direction, and measurements of the ambient temperature and pressure via a meteorological data logger.

Matched-filter CH 4 retrieval
We retrieve CH 4 column enhancements with the albedo-corrected reweighted-ℓ 1 matched filter described in detail in Foote et al (2020). This algorithm improves on the classic matched filter by introducing a sparsity constraint on the CH 4 enhancements and a pixel-wise albedo correction. Since we observe skylight, the albedo, in our case, accounts for variable sky brightness, e.g. caused by clouds and aerosols.
Matched filter algorithms identify a spectral target signature ⃗ t i (dimension: number of spectral samples M) in each spatial pixel i = 1 . . . K of a hyperspectral image. The target signature is constructed from a unit absorption spectrum ⃗ s i (dim: M) that, in our case, is the relative change of the absorption spectrum per unit change (1 ppm m) in CH 4 column, and from a mean background spectrum ⃗ µ (dim: M) via elementwise multiplication (⊙), Following the iterative approach suggested by Foote et al (2020), we start out with ⃗ µ 0 at iteration n = 0 being the mean spectrum of the entire image and C 0 (dim: M × M) being its covariance across the image. Then, for the iteration updates n > 0, we iteratively remove the found CH 4 signature from ⃗ µ n via and recalculate the image covariance C n accordingly. Thereby, ⃗ L i is the measured radiance of spatial pixel i and α n,i is the CH 4 column enhancement for iteration n in spatial pixel i. The latter calculates via where r i is the brightness of spatial pixel i and w n,i is the pixel's regularization weight of the sparsity prior of iteration step n, and κ is a small number for numerical robustness. Note that the max-condition in equation (3) enforces non-negative enhancements and that the classic matched-filter without brightness correction and sparsity constraint can be recovered by setting r i = 1 and w n = 0. After convergence of the iteration, the above algorithm delivers a CH 4 column enhancement α i for each spatial pixel i, thus an image of CH 4 column enhancements across the scene.
The key ingredient for the CH 4 retrieval is the unit absorption spectrum ⃗ s i . For its calculation, we use a radiative transfer model under single-scattering assumption to simulate the radiances ⃗ L i,plume and ⃗ L i, bkg with and without CH 4 plume enhancement. For the plume enhancement, we assume absorption by a typical CH 4 plume in the lowest atmospheric layer represented by Beer-Lambert's law. The unit absorption spectrum⃗ s i then calculates according to Foote et al (2021) emphasize the importance of scene-specific unit absorption spectra ⃗ s i , i.e. the fact that they actually depend on spatial pixel i. Therefore, we generate a look-up table of unit absorption spectra for solar zenith angles (SZAs) between 10 • and 70 • and camera viewing elevation angles (VEAs) between 1 • and 20 • . The retrieval of the column enhancements interpolates a specific⃗ s i from the look-up table for the VEA and SZA at time and position of the observation. For more details on the retrieval procedure we refer to supplement S1.

Estimation of emission rates
For estimating emission rates from the images of CH 4 column enhancements, we use the integrated mass enhancement (IME) method which has been evaluated extensively for aircraft and satellite based studies , Varon et al 2018, Duren et al 2019, Ayasse 2022. The method relates the total CH 4 mass enhancement M CH4 and the effective residence time τ in the observed plume to the influx into the plume and thus, the emission rate E. It is essentially a one-box model with in-and outflux under steady-state assumption.
First, we identify the CH 4 plume in the image as the largest contiguous area of non-zero enhancements. The sparsity constraint of equation (3) effectively forces most enhancements to zero if there is no clear CH 4 enhancement detectable. Therefore, the largest contiguous patch of enhancements is most likely the emitted plume. We also take into account the second-largest patch in the image if it is at least 70% the size of the largest one and both patches correspond in the vertical. Once the K P plume pixel are identified, the total plume mass M CH4 is given by where α i (ppm·m) is the CH 4 column enhancement delivered by the matched filter and A i (m 2 ) is the geometrical area of pixel i, and k CH4 ≈ 7.2 × 10 −4 (g ppm −1 m −3 ) is a conversion factor assuming normal conditions. The pixel area is known from the distance between camera position and shaft position. We use the collocated measurements of wind velocity u and wind direction ϕ to calculate the residence time τ in the observed plume. In our images, we define the projected length d of a plume as the apparent distance between two vertical cross-sections through the plume in the image. The projected length d is given by d =d · sin(ϕ), whered is the true plume length and ϕ the angle between the wind direction and the camera pointing. Thus, the residence time is and the emission rate in the IME approach calculates to where M CH4 and d are inferred from the HySpex measurements and u and ϕ from the Windranger measurements at the mass-weighted mean height of the identified plume.
To quantify the errors for the emission rate estimates, we use an empirical approach. We estimate the error of the first factor ρ = M CH4 /d in equation (7) by calculating it for ten different distances along each plume, i.e. by assuming ten different plume box sizes (Duren et al 2019). The standard deviation ∆ρ among these ten slices provides an error estimate that, in particular, includes errors due to the assumption of a constant emission rate during the measurement scan (roughly 1 min), fluctuations due to turbulence, and mass loss due to cutting out enhancements below the detection limit. To estimate errors of the wind measurements, we take the standard errors of the mean for u and ϕ, ∆u and ∆ϕ, over 10 min. Based on these error contributions, we calculate the emission rate error ∆E via Gaussian error propagation

Quality filters
We exclude observations from the emission estimation if they either were taken under ill-suited conditions or do not meet the assumptions of the IME method. Firstly, we remove trivial cases of viewing obstructions, e.g. sporadically appearing corn plants in the image due to wind on June 19. A dark or heterogeneous image background, e.g. under cloudy conditions, decreases the retrieval sensitivity and thus, the matched filter misses plume pixels. Therefore, we remove images where the number of plume pixels is smaller than 900. Furthermore, plumes which do not start out close to the coal mine shaft (radius ≈ 7 m) cannot be attributed unambiguously to the source and are excluded. The IME method assumes that the lateral movement of the plume is dominated by advection with the background wind. Turbulent wind conditions and low background wind speeds can lead to situations where the turbulent spread dominates over the advective effect. Therefore, we exclude images if less than 85% of the enhanced pixels are downwind of the emitting shaft. Lastly, we exclude plumes carried away under slant angles (| sin(ϕ)| < 0.45) relative to the camera line-of-sight, since under such conditions small direction errors propagate into large emission errors and the plume length d cannot be derived reliably from the image.

Results
We report on the emission dynamics of a ventilation shaft of the Pniowek mine in the USCB for four days in June 2022. An illustrative snapshot image of a CH 4 enhancement plume is shown in figure 2. Figure 3 shows an event during which a puff-like plume emanates from the ventilation shaft causing a jump in the emissions. Figure 4 shows the time series of the estimated emission rates, and table 1 lists the mean emission rates, error contributions, and peak-to-peak variation of the daily emission dynamics of the shaft. We recommend watching the videos provided in the supplementary material to observe the full dynamics of the entire dataset. The CH 4 plume image in figure 2 was taken on 19 June 2022, under stable, clear-sky conditions. Throughout that day wind speeds were reasonably large ranging between 4.0 and 7.6 m s −1 and wind directions were reasonably close to being perpendicular to the lines-of-sight with deviations ranging between 5 and 54 • . Note that, for the duration of a horizontal scan (roughly 1 min), the wind typically carries the emitted CH 4 farther than the horizontal extend of the image. Therefore, we observe a new plume every image. From the CH 4 plume enhancements such as shown in figure 2, we calculate the emission rate according to equation (7) and the respective uncertainty estimate according to equation (8). The latter calculation is illustrated in panel (b) of figure 2. We slice the plume into ten boxes that include more and more parts of the plume, i.e. both the IME M CH4 and the length of the plume d grow from box to box. Note that the boxes start 5 m downwind of the shaft to exclude the near-field of the source where the buoyant rise dominates over advective transport. If the emission rate was constant during scan duration, if no mass was lost to the calculation (e.g. because of being below the detection limit or because of viewing obstructions), and if turbulence fluctuations were to average out, the emission estimates for the various pairs of M CH4 and d should be equal. Thus, we take the standard deviation of the ten estimates as a measure for non-compliance with the listed assumptions and we use it to estimate the uncertainty of the factor M CH4 /d as explained in section 2.3. For the particular image of figure 2, we find an emission rate of 2.28 ± 0.19 tCH 4 h −1 with uncertainty contributions of 0.16, 0.08, 0.06 tCH 4 h −1 coming from the factor M CH4 /d, from wind speed and direction, respectively.
To illustrate the plume and emission dynamics observable with our camera, figure 3 shows two plume images on June 18, 2022, which were observed 11 min apart. The first plume image was taken at 08:04 UTC and carried 7.00 kg of CH 4 according to our analysis, while the plume at 08:15 UTC carried 19.7 kg. This rapid increase in plume mass increases the emission rate estimate from 1.5 ± 0.2 tCH 4 h −1 to 4.6 ± 1.0 tCH 4 h −1 . The period of elevated emissions lasted over approximately 10 min (figure 3(c)), but returned to previous levels afterwards. Similar events have been observed at 13:15 UTC and 15:05 UTC on June 18, 2022, and at 09:15 UTC and 09:28 UTC on June 20, 2022, which are clearly visible in the emission rate time series (figure 4) and in the supplementary videos. This suggests that, indeed, there is significant source dynamics exceeding 1 tCH 4 h −1 for a single facility on time scales of minutes and thus, source dynamics needs to be considered as a systematic uncertainty for emission estimates from snapshot observations (Bhardwaj et al 2022). Figure 4 shows the emission time series from June 17 to 20, 2022. Each day covers between 149 and 372 images which pass the quality filters, with image scan times between 57 s and 72 s. Conditions were most favorable for June 19 with persistent clear-sky and stable wind conditions while background cloud cover (June 17) and wind conditions (June 18 and 20) caused filtering out some data (see supplement for details on daily measurement conditions).
We perform an emission estimate with uncertainty analysis on every image that passes the quality filters. The daily average uncertainty contributions (see equation (8)) from wind direction and wind velocity vary between 2.92% and 7.24%, and 3.41% and 4.95%, respectively, while the uncertainty of the plume mass factor ρ = M CH4 /d contributes between 10.48% and 17.41% (see table 1). Notably, the uncertainty budget is not dominated by the wind contribution due to on-site measurements by the co-located wind-lidar. The daily average emission rates vary by a factor of more than 3, growing from 1.39 ± 0.19 tCH 4 h −1 on June 17 (Friday) to 4.44 ± 0.76 tCH 4 h −1 on June 20 (Monday). Previous studies of the single coal mine shafts in the USCB show single shaft emission rates from 0.13 tCH 4 h −1 to 1.03 tCH 4 h −1 (Andersen et al 2021) and 0.68 tCH 4 h −1 to 1.14 tCH 4 h −1 (Luther et al 2019). Using AirCore samples collected by a UAV, Andersen et al (2022) report emissions of 0.1 to 1.7 tCH 4 h −1 for Pniowek V in May 2018. The E-PRTR (European Pollutant Release and Transfer Register) 2018 emission inventory lists 2.08 tCH 4 h −1 as the annual mean of the Pniowek V coal mine shaft. Swolkień et al (2022) found Pniowek V exhibited the largest emission variations during two months in 2018 with hourly emission snapshots from in-situ sensors between 0.00 tCH 4 h −1 and 2.45 tCH 4 h −1 .
To provide a quantitative estimate of source dynamics, we calculate the 10 min rolling mean of the time series. The rolling mean kernel size must be chosen small enough to capture the source dynamics and long enough to reduce the errors that are inherent to individual measurements. Observations under favorable conditions (19 June 2022) suggest that a 10 min rolling mean is a reasonable choice. Table 1 lists the peak-to-peak dynamics of the 10 min rolling mean, indicating that emission rates averaged over 10 min can be up to a factor 1.5 to 2 lesser or greater than the daily mean. Furthermore, we observe significant emission dynamics within less than 30 min. On 19 June 2022, the emissions rise and fall by approximately 1.1 tCH 4 h −1 between 7:40 UTC and 8:20 UTC and by 0.8 tCH 4 h −1 between 13:50 UTC and 14:20 UTC.
To illustrate limitations of the imaging technique, supplementary figure S2 shows cases which make either one of the filters respond or which induce substantial error propagation. Cloudy skies (such as visible in the background of figure S2(a)) cause the matched filter to occasionally miss the plume or to only detect a fraction of the plume. The heterogeneous reflectivity of partially cloudy skies induces structures in the background covariance (equation (3)) and light path effects of clouds can cause interfering spectral patterns in the spectral background distribution. This makes it harder to identify CH 4 column enhancements, and images are prone to be filtered because of too small plume sizes. Wind directions driving the plume away or toward the observer (such as panel (b) in figure S2) are unfavorable for the emission estimates since small direction errors propagate into large emission rate errors and since the advection driven plume size is difficult to determine. Therefore, we filter such cases. Likewise, low wind speeds are problematic (such as panel (c) in figure S2) since the CH 4 plumes are less streamlined and more unstable with respect to lateral and vertical dispersion within the image than for higher wind speeds. This implies that we miss a large fraction of the plume mass since it is below the detection limit. We find that the method works well for wind-speeds down to 2 m s −1 . Furthermore, we find a small spurious correlation of emission estimates with effective plume velocity u eff = |u · sin(ϕ)|. Figure S3 shows the time series and the coefficient of determination R 2 for each day, which ranges from 0.01 to 0.28 implying that a corresponding fraction of the observed variance in the emission rates correlates with wind variability. While this needs further investigations, a possible explanation might be that, for   (8) from wind speed u (⟨eu⟩) and direction ϕ ⟨e ϕ ⟩ as well as the plume mass factor ρ = MCH 4 /d (⟨eρ⟩). The peak-to-peak variation of the 10 min rolling mean E 10 min is listed with min(E 10 min ) and max(E 10 min ). Date

Discussion and conclusions
Our study demonstrates the usage of a ground-based HySpex SWIR-384 imaging spectrometer for the quantification of CH 4 emission dynamics from a coal mine ventilation facility in the USCB. For the demonstration campaign in June 2022, we deployed the spectrometer at 1 km distance from the Pniowek V ventilation shaft and collected hyperspectral images in the SWIR spectral range above the facility. Submitting the images to a matched-filter algorithm, we inferred plumes of CH 4 column enhancements above the facility with a temporal resolution of roughly 1 min. Together with wind information from a windlidar, the IME method yielded emission rate estimates. Typical errors are in the range of a few hundred kgCH 4 h −1 for wind speeds down to 2 m s −1 .
Our data shows substantial variability on all covered timescales from minutes to days. The daily average emission rates grow from 1.39 ± 0.19 to 4.44 ± 0.76 tCH 4 h −1 , thus by a factor of 3, over the four observation days. The 10 min average emission rates are a factor of 1.5 to 2 smaller or greater than the daily average, and there is considerable variability on time scales below 15 min, as illustrated by puff events caught by the camera images. The observed emission rates fit the range of the ones found by Swolkień et al (2022), who report on the CH 4 mole fractions inside ventilation shafts in the USCB ranging between 0.0% and 0.4% for the Pniowek V shaft during their study period in 2018, which they translated into emission rates between 0.00 and 2.5 tCH 4 h −1 . The high variability of the CH 4 emissions is due to the complex interplay between the structure and composition of the CH 4 -bearing rock under excavation, the locations and operational phases of the mining processes, the ventilation intensity and the turbulent ventilation currents inside the mine (Swolkień 2020, Swolkień et al 2022.
Monitoring CH 4 emissions from such sources requires tools that can quantitatively resolve and attribute the temporal variability. The particular advantages of the technique presented here are (a) its high specificity to CH 4 facilitated by spectroscopically resolving CH 4 absorption in skylight spectra, (b) its imaging capabilities supporting direct source attribution, plume identification and, together with wind information, emission rate estimation, and (c) its high temporal resolution enabling quantification of emission dynamics on the time scale of minutes. The method relies on sunlight scattered in the sky, which limits its deployment to daytime and reasonably fair weather conditions. On the other hand, using spectra of sunlight in the SWIR instead of thermally emitted radiation such as imaging spectrometers in the mid-infrared (Gålfalk et al 2016(Gålfalk et al , 2022 enables kilometer-scale distances to the source, and it makes the method independent of thermal contrast between the plume and the background. Cloudy conditions, variable wind directions, and low wind speeds pose challenges to the detection of the CH 4 plumes and the quality of the emission estimates but, in many cases, such conditions can be filtered reliably. Thus, the technique complements observations from aircraft and satellites that cannot observe source variability on such short time scales. It also complements qualitative OGI methods for local leak detection at the facility and thermal infrared spectrometers that operate closer to the source and require thermal contrast in the scene. Thus, our setup adds to state-of-the-art emission monitoring methods (Bastviken et al 2022) by bringing in the capability to remotely measure the emission dynamics of strong localized sources with a temporal resolution of minutes. While CH 4 emissions from coal mines might be particularly variable, other CH 4 source types, such as in the oil and gas industry (Bhardwaj et al 2022, Foulds et al 2022 or waste treatment, likely also show substantial variability on timescales of minutes, hours and days.

Data availability statement
The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors.