2.1 Study area

Shanxi province is the most extensively mined region in China, harbouring nearly half of the nation's suspected point sources based on TROPOMI observations (Schuit et al., 2023). It stands as a typical area for CH₄ point source emissions in China and has been a focal point in prior comparative studies on point source emissions (Sánchez-Garcia et al., 2022; Guanter et al., 2021). Shanxi province (Fig. 1), situated in northern China, experiences a temperate continental monsoon climate characterised by cold, dry winters and hot, humid summers. The region boasts diverse topography, comprising mountains, plateaus, and basins. Consequently, the stable atmospheric conditions during winter can lead to the accumulation of pollutants closer to the ground, impacting the detection of CH₄ emissions. Although the region has strict rules in regulating the process of CH₄, a byproduct of coal mining, underground coal mines in Shanxi release CH₄ to the atmosphere from mine venting. Therefore, the identification of these plumes will help mitigate CH₄ emissions over this region.

Figure 1(a) The study area in Shanxi enclosed by the blue boundary, with the background image from © Google Maps. (b) Gaofen-5B observed scene images used for the CH₄ plume survey. The black dots represent the potential point sources detected by TROPOMI (Schuit et al., 2023). The red dots represent the three national weather stations for monitoring meteorological variables in Yangquan, Changzhi, and Jincheng that are used for the wind field comparison with ERA5 reanalysis (Sect. 4.3).

2.2 GF-5B/AHSI dataset

The GF-5B satellite is the second satellite of the Gaofen-5 series and was launched on 7 September 2021. It has accumulated over 2 years' worth of global observational data to date. Equipped with the Advanced Hyperspectral Imager (AHSI), it can capture spectral information spanning 400 to 2500 nm, with a spatial resolution of 30 m over a 60 km swath, encompassing 330 spectral channels with spectral resolutions of 5 and 10 nm in the visible near-infrared (VNIR) and short-wave infrared (SWIR) spectroscopy, respectively (Y.-N. Liu et al., 2019). Its relatively high signal-to-noise ratio (around 500 in the SWIR) presents notable advantages in detecting CH₄ point source emissions (Irakulis-Loitxate et al., 2021). The retrieval of the enhancement of column-averaged dry-air mole fraction of CH₄ (ΔXCH₄) relies primarily on strong CH₄ absorption features near 2300 nm, while the 2100 to 2450 nm spectral window of the GF-5B/AHSI demonstrates higher sensitivity to XCH₄ variations, thereby possessing enhanced capabilities for precise CH₄ concentration inversion. This study focuses on Shanxi province, using images from 111 GF-5B/AHSI scenes covering suspected point sources from September 2021 to September 2023, with a cloud cover of less than 10 %, employed for ΔXCH₄ inversion and point source identification (Fig. 1b). These images cover the major emission hotspots, as identified by TROPOMI data (Schuit et al., 2023). Noted that, in Shanxi, the overpassing time of GF-5/AHSI is around 11:00–12:00 Beijing time (BJT).

The SWIR imagery in the AHSI band employs a strategic arrangement of four strips. Each SWIR strip corresponds to a 15 km ground swath, resulting in a continuous 60 km swath width across the satellite orbit with four images combined. This configuration yields a total of 2012 px (including 36 overlapped pixels) along the spatial dimension of the SWIR detectors (Y.-N. Liu et al., 2019). Therefore, the target inside the overlapped pixels could be observed twice in 8 s.

2.3 Auxiliary data

Methane point source detection and emission estimation involve various auxiliary datasets, mainly including (1) ultra-high-resolution surface imagery for checking false positive in point source detection, (2) wind fields information for estimate emissions from point source plumes, and (3) digital elevation model (DEM) data for the geometric correction of AHSI imagery. High-resolution surface imagery is an indispensable dataset in point source identification and serves as direct evidence for distinguishing interference signals. The high-resolution imagery used in this study primarily comes from Google Earth. Wind speed data are a critical parameter for calculating emission flux rates. The study utilised U₁₀ hourly wind speed reanalysis products from ECMWF ERA5, with a spatial resolution of 0.25° × 0.25° (Muñoz-Sabater et al., 2021). Terrain data are crucial for the geometric correction of AHSI imagery, directly impacting the positioning and identification of ΔXCH₄ plume signals. The study used DEM data from SRTM (Farr et al., 2007) with a spatial resolution of 30 m. Additionally, the study obtained hourly meteorology monitoring data, including wind speed and wind direction, from January 2021 to September 2023 from three national meteorological stations in Yangquan, Changzhi, and Jincheng (Fig. 1), obtained from the China Meteorological Administration Data Centre. These data were compared with ERA5's U₁₀ hourly wind speed reanalysis products to investigate the uncertainty in the ERA5's wind field.

3.1 ΔXCH₄ retrieval using matched-filter method

3.1.2 Spectral matched filter for retrieving ΔXCH₄

Spectral matched-filter method derives the ΔXCH₄ by calculating the difference between the “polluted” spectra over a source region with background spectra of the ambient atmosphere and expressing the difference by the number of target absorption spectrum from one unit of XCH₄ (e.g. 1 ppm of XCH₄; Guanter et al., 2021). The retrieval using matched filter is depicted in Eq. (1):

$$\begin{array}{}\text{(1)}& \mathrm{\Delta }{\mathrm{XCH}}_{\mathrm{4}}=\left((\mathit{x}-\mathit{\mu }{)}^{\mathrm{T}}{\sum }^{-\mathrm{1}}\mathit{t}\right)/\left({\mathit{t}}^{\mathrm{T}}{\sum }^{-\mathrm{1}}\mathit{t}\right),\end{array}$$

where x denotes a vector of the observed SWIR hyperspectral spectra from a target pixel. In this study, the CH₄ strong absorption band (2110–2455 nm) is used; μ and Σ represent the mean and covariance of the SWIR hyperspectral observation over background regions, respectively. t is the target spectrum, representing the disturbance vector of SWIR hyperspectral due to enhanced XCH₄ relative to the background. It can be derived from an element-wise multiplication of μ and the unit XCH₄ absorption spectrum κ, which is generated from GFIT3 (Zeng et al., 2021), as shown in Fig. 2, assuming a perturbation of 1 ppm XCH₄.

Figure 2Example of unit XCH₄ absorption spectrum used as target signature in the matched-filter retrieval method. The high-resolution target signature (in grey) represents absorptivity induced by 1 ppm XCH₄ enhancement, which is calculated using GFIT3 (Zeng et al., 2021). The high-resolution absorptions are then convolved with a Gaussian ILS with nominal FWHM from GF-5B/AHSI to derive the spectra (in red) that can be compared with that observed from AHSI.

Download

3.2 Identifying point source plumes from ΔXCH₄ maps

After data pre-processing, including spectral re-calibration and ΔXCH₄ retrieval, we implemented a geometric localisation to change the GF-5B/AHSI imagery index for row and column pixels to latitude and longitude under WGS84 projection. The detailed description of this geometric localisation is in Appendix A. Then, this study compares the ΔXCH₄ maps with high-resolution Google Earth imagery to visually inspect and preliminarily identify the CH₄ point source plumes. The identification criteria include: (1) high ΔXCH₄ values displaying plume characteristics, (2) the presence of ground facilities such as factories or pipelines representing potential CH₄ emission sources, and (3) plume distribution characteristics not caused by terrain features that may impact short-wave infrared strong absorption in surface features. Although wind conditions directly affect plume features, however, the reanalysis data (e.g. ERA5) of the wind direction may be very different from the plume structure. Several factors could contribute to this mismatch, including the temporal and spatial resolution of the reanalysis data, local topographical features, and microscale meteorological phenomena that are not fully captured by the reanalysis data. Therefore, this study temporarily refrains from utilising wind direction from ERA5 reanalysis as a direct criterion for point source identification.

3.3 Estimation of emission flux rates

4.1 Detection and estimation of emission flux rate for single CH₄ point source using GF-5B/AHSI

Figure 3 demonstrates the retrieval results of point sources ΔXCH₄ based on multiple capturing of the same point source using GF-5B/AHSI from January 2022 to August 2023. Under different emission flux rates and wind conditions, the emission plumes exhibited various characteristics. Six observations occurred during the winter–spring seasons (Fig. 3b–g), showing ΔXCH₄ plumes spreading northeastward, while the observation in summer (Fig. 3h) displayed a plume drifting northwestward. An intriguing aspect is the occurrence of two repeated observations of the same point source within an 8 s interval (Fig. 3d, e and f, g). Theoretically, CH₄ emissions from the same point source within an 8 s interval should exhibit very similar patterns. However, using the full-scene image as the background region for the background spectra calculation for each plume, similar to previous studies, the ΔXCH₄ of the plumes from the same point source showed large differences, especially for Fig. 3f and g. The notable difference primarily arises from the different background used, suggesting the importance of selecting appropriate background regions. Note that the difference may also be slightly caused by the different signal-to-noise ratio (SNR), as the plumes appear at different locations in the imaging scene. The plumes appear at the bottom of the scene in Fig. D1f and at the top in Fig. D1g. They may be observed by different detectors with different SNR of the instrument that affect the detection accuracy of the plumes.

Figure 3Example of ΔXCH₄ retrievals from one typical point source with multiple overpasses by GF-5B/AHSI, with its origin (lat 37°57^′36^′′, long 113°16^′04^′′) marked with a red/black star. The detected plumes from the seven overpasses shown in panels (b)–(h). The observation times (in UTC + 8, standard Beijing time) are shown for each plume event, which are close to local time. The background image in panel (a) is adopted from © Google Maps.

Based on the ΔXCH₄ retrieval and the flood-fill plume segmentation method, we obtained the plume characteristics and emission flux rate of the seven detections, as shown in Fig. 4. The results indicate the following: (1) differences exist between the extracted plume features and visual segmentation. For instance, in Fig. 4c, the elaborate plume automatically extracted using flood-fill would be challenged for manually drawing; (2) the point source emission flux rate varies between 2834.79 ± 1330.98 kg h⁻¹ in Fig. 4e and 6678.66 ± 2316.49 kg h⁻¹ in Fig. 4g (excluding incomplete observations in Fig. 4d). Among these observations, four fall within a similar range between 2834.79 and 3247.22 kg h⁻¹; (3) the uncertainty in the IME ranges from 5.07 % to 66.38 %, with the majority being below 30 %, which is lower than the uncertainty caused by wind speed (∼ 50 %) in the emission flux rate calculation. (4) Significant differences are evident in the plumes from adjacent detections of the same point source (e.g. Fig. 4f and g), indicating the different backgrounds chosen for different imagery scenes are not optimal to monitoring the same emission plumes.

Figure 4Examples of the extracted CH₄ point source, marked with red/black star (lat 37°57^′36^′′, long 113°16^′04^′′), with the plume using flood-fill method based on the retrieved ΔXCH₄ maps, as shown in Fig. 3, from a single point source with multiple overpasses by GF-5B/AHSI. The plume mass from IME model and the estimated emission flux rates are also indicated at the bottom of each map. The observation time in Beijing time is shown for each plume event. All background images in panels (a)–(h) are from © Google Maps.

In order to eliminate the impact of background selection on estimating emission flux rate from the same point source, this study conducted a ΔXCH₄ retrieval experiment using overlapping area in the imagery maps of Fig. 3f and g as the new background. The results based on the new backgrounds shown in Fig. 5 demonstrate highly similar ΔXCH₄ plume features between the two observations that are 8 s apart (Fig. 5a and b). The extracted plume distribution and emission flux rate calculations shown in Fig. 5c and d are almost identical. The integrated enhanced masses were 1001.59 ± 13.98 and 1046.86 ± 15.20 kg, respectively, with emission flux rates of 5477.44 ± 1839.08 and 5730.52 ± 1925.58 kg h⁻¹. This reduced estimation discrepancy between the two by 485.11 kg h⁻¹, which is about 8.5 % of the emission flux rate. Therefore, for the calculation of emissions from all plumes in Shanxi, we adopted a two-step approach to identify CH₄ plumes and estimate emissions. In step 1, the whole image is used to calculate ΔXCH₄ and identify plumes; in step 2, when implementing the flood-fill method with the strategy of selecting background regions as described in Sect. 3.3.3, the ΔXCH₄ is re-calculated using the same background regions for the flood-fill method. In other words, the chosen background regions are used for calculating ΔXCH₄ using Eq. (1), segmenting plumes using flood-fill method, and estimating IME using Eq. (2).

Figure 5ΔXCH₄ retrievals from GF-5B/AHSI observations that are 8 s apart in panels (a) and (b) over the same point source and marked with red/black star (lat 37°57^′36^′′, long 113°16^′04^′′). The retrievals are carried out using the same background region. The corresponding IME values and emission flux rates (Q) based on the extracted ΔXCH₄ maps are shown in panels (c) and (d), respectively. The difference between the two IME values is shown in panel (e). All background images in panels (c)–(e) are from © Google Maps.

4.2 Spatial distribution of point sources and their emission rates in Shanxi

Based on the methods described above for estimating the CH₄ emission flux rate of point sources, we conducted a survey of all detectable point source emissions using all available imagery of GF-5B/AHSI from 2021 to 2023. In total, 93 point source plumes were identified. After averaging repetitive observations over the same point sources, a total of 32 point sources were identified, and their spatial distribution is depicted in Fig. 6. Figure 6a–i exhibit typical plume extraction results around three typical cities of Yangquan, Changzhi, and Jincheng. The emission flux rates range from 2147.08 ± 427.42 to 9198.03 ± 2059.18 kg h⁻¹. This result demonstrates a reasonably good consistency between the spatial locations of the actual CH₄ emission point sources identified in this study (red dots in Fig. 6) and those extracted based on TROPOMI data (black dots in Fig. 6), which are primarily concentrated around the three cities of Yangquan, Changzhi, and Jincheng. Given its high spatial resolution, the spatial locations derived from GF-5B/AHSI are expected to be more accurate. We found that the number of identified point sources is much fewer than those extracted from TROPOMI. This is primarily attributed to the much denser observations with daily global coverage and its different overpass time from TROPOMI (∼ 13:30 LT, local time). In addition, the high-resolution of the ΔXCH₄ retrieval results helped eliminate false positive signals due to surface interference elements like photovoltaic panels (further details are discussed in Sect. 4.3.2) and greenhouse cultivation structures that are ubiquitous in Shanxi. Driven by wind speed and topography, different plumes from various point sources show distinctly varying dispersion distances, ranging from less than 1.0 km (e.g. Fig. 6h) to 5.0 km (e.g. Fig. 6d). In the calculation of emission flux rate, the plume length plays an important role in the background region selection, which affects both the inversion of ΔXCH₄ (background spectra) and plume segmentation using the flood-fill algorithm (statistical parameters).

Figure 6The spatial distribution of the identified CH₄ plumes (in red dots; in total 93) in Shanxi using GF-5B/AHSI observations, as shown in the centre panel. The black dots represent the potential point sources detected by TROPOMI (Schuit et al., 2023). CH₄ plumes (a–i) are examples of the identified ΔXCH₄ plumes in Shanxi, and the yellow arrow points to the origins of the identified point sources. All background images in panels (a)–(i) are adopted from © Google Maps.

We further conducted IME calculations and emission flux rate estimations for the 93 plumes extracted from GF-5B/AHSI (Fig. 7a and c). Additionally, based on multiple observations (from two to eight times) of the same point source, we provided the highest and lowest emission flux rates and IME for the same point source (Fig. 7b and d). The survey results revealed a diverse range of point source emission flux rates, varying from 761.78 ± 185.00 (minimum) to 12 729.12 ± 4658.13 kg h⁻¹ (maximum), with an average of approximately 4040.30 kg h⁻¹. The range of IME from point source emissions spans from 33.58 ± 6.27 (minimum) to 6587.50 ± 1925.31 kg (maximum). This discrepancy in the range of IME and the emission flux rates is indeed notable and can be attributed to the variability in wind conditions. Moreover, assuming a 50 % uncertainty in U₁₀ (wind speed at 10 m), in the uncertainty calculation of Q (emission flux rate), the impact of wind speed and IME uncertainties accounts for approximately 84.84 % and 15.16 %, respectively. While it is evident that wind speed is the predominant factor contributing to the uncertainty in estimating CH₄ point source emissions, it is crucial to acknowledge the significance of IME-related uncertainties. Because the wind uncertainty could be decreased significantly with accurate weather observations from high-density sites from China Meteorological Administration in theory. In addition, we should acknowledge that low wind speeds may also lead to uncertain dispersion of methane plumes, potentially biasing emission estimates, which need deeper analysis in future studies.

Figure 7(a) CH₄ emission flux rate from point source plumes nos. 1–93 in descending order of emissions, with the error bars representing the estimation uncertainty. (b) The maximum and minimum emission flux rates for each point source with more than two observations. (c) The corresponding IME estimates for plume nos. 1–93, following the order in panel (a). (d) The maximum of the minimum emission flux rates for each point source with more than two observations.

Download

Furthermore, multiple observations of the same point source indicate significant variations in CH₄ emissions over time. The difference is as large as 10 204.71 kg h⁻¹, which is about 7 times the amount between the maximum and the minimum, as shown in Fig. 7d. This difference suggests that a single observation does not adequately represent the overall or averaged emission scenario for any point source. Because the specific emission law of each emission point is unclear, more coal mine emission time series detection experiments (Qin et al., 2023) are needed for the overall emission rate evaluation. Although, Y. Chen et al. (2022) used high-density (26 292 active wells) and highly repeated (115 flight days) measurements from an aerial instrument to quantify methane emissions from the whole regional study area of the New Mexican Permian Basin with the persistence-averaged method. The persistent emission rate from a single point source was calculated with the emission detection probability derived from highly repeated observations. In this study, this may not be feasible because the observations are too few to calculate the possibility of emission detection. As the emission rates of the detected plume in this study (Fig. 7), the detection limit based on GF-5B AHSI and our current method could be 761.78 ± 185.00 kg h⁻¹. However, there might be some smaller point source emissions in Shanxi, which could be detected with the inversion method improved.

4.3 Improvements on ΔXCH₄ retrieval and emission flux rate estimation

4.3.1 Spectral calibration of GF-5B/AHSI observations

The impact of the wavelength shift and changes in FWHM of the spectral observations from GF-5B/AHSI on deriving ΔXCH₄ is demonstrated in Figs. 8 and 9. Figure 8 illustrates an example of the cross-track pixel variations in the estimated centre wavelength in Fig. 8a and FWHM in Fig. 8b in a single-scene image collected on 29 January 2022. The results reveal the distinct deviations from the nominal centre wavelength and FWHM among different track pixels during satellite imaging. Figure 8c displays the ΔXCH₄ of the corrected image, capturing plumes seen in Figs. 3b and 6c, d, among others. Figure 8d and e show the evident striping differences and spectral calibration's impact on calculating ΔXCH₄ of individual plume. The difference can reach up to 100 ppb. To further assess the spectral calibration's influence on CH₄ point source estimation, this study analysed the shift in centre wavelength and changes in FWHM in 111 representative scenes with potential point source emissions using GF-5B/AHSI, as shown in Fig. 9a and b. The results indicate that the average shift in centre wavelength of GF-5B/AHSI is approximately −0.05 nm, mostly ranging between −0.2 and 0.1 nm. The ratio of change in FWHM averages around 1.1, predominantly falling between 1.0 and 1.25 times (between 0 and 2.13 nm). Furthermore, the study evaluated the impact of spectral shift and FWHM change on the estimation of point source emission flux rate, as shown in Fig. 9c and d. The results indicate that the caused difference in the point source emission flux rate ranges from 0.43 to 500.96 kg h⁻¹. The average percentage of change is (1.78 ± 1.39)%. The maximum difference reaches up to about 5.0 %. By considering the shift in central wavelength and change in FWHM in the spectral observations, it exhibits a potential to reduce the uncertainty in the XCH₄ emission rate estimation using GF-5B/AHSI.

Figure 8Example of the shift in the centre wavelength and FWHM change for across-track pixels of channel no. 142 from GF-5B/AHSI SWIR imagery and their impacts for ΔXCH₄ retrieval. Panel (a) shows the shift in the centre wavelength for across-track pixels. Panel (b) shows the FWHM variation ratio for across-track pixels. Panel (c) shows the ΔXCH₄ retrieval of a single image with inputs of updated spectral calibration parameters. Panels (d) and (e) are the comparison of zoomed-in plumes with and without the inputs of updated spectral calibration parameters, in which panels (d-1) and (e-1) are results without calibration, panels (d-2) and (e-2) are results with calibration, and panels (d-3) and (e-3) are the corresponding difference in ΔXCH₄ retrieval.

Download

Figure 9(a) Statistics of the shift in centre wavelength and (b) FWHM variation ratio of all 111 GF-5B/AHSI SWIR images with potential CH₄ point sources. (c) The difference in the estimations of emission flux rates and (d) the corresponding difference in percentages for all detected CH₄ plumes shown in Figs. 6 and 7.

Download

4.3.2 Impact of heterogeneous surface features

Complex surface features significantly affect the identification of suspected point sources based on ΔXCH₄ maps and the derivation of point source emissions. In this study, we originally observed 219 instances of 113 suspicious point sources. In a more refined identification of these sources, we cross-checked and confirmed their positions using ΔXCH₄ retrievals from GF-5B/AHSI against high-resolution Google Earth imagery. Our findings revealed that the identification of point sources was significantly affected by the complex surfaces that exhibit strong SWIR absorption similar to CH₄ and therefore result in false positive signal. Notably, the array of solar panels that is widespread in Shanxi is the primary disruptor of the spectral matched-filter retrieval method. An example of solar panel arrays is shown in Fig. 10. Moreover, we found that surface features such as greenhouse structures, certain buildings, waterbodies with plume-like distributions, and moist cultivated lands (like paddy fields) also generated noticeable high-value ΔXCH₄ interference signals. Therefore, in CH₄ point source detection using GF-5B image, it is essential to consider combining with high-resolution images to filter out false positive signals.

Figure 10Examples of the impact of array of solar panels, which generates false positive signals, on the ΔXCH₄ retrieval in Shanxi. ΔXCH₄ retrievals with high values are similar to plume shapes in panels (a) and (d). The false positive signals are caused by the similar patterns of solar panel arrays, which can be seen from high resolution of the ©Google maps in panels (b) and (e). Zoomed-in details of solar panels in the red boxes in panels (b) and (e) can be found in panels (c) and (f), respectively.

Download

4.3.3 Evaluating wind fields from ERA5 reanalysis using observations from meteorological stations in Shanxi

Wind fields, including wind speed and direction, are the primary drivers of uncertainty in estimating point source emissions, especially in plume segmentation and flux rate calculations. For plume segmentation, instead of visual interpretation, this study introduces the flood-fill method. Instead of searching in eight directions in the current study, accurate wind direction information could enable us to precisely narrow down the flood-fill search directions, thereby removing abnormal signals from non-point source emissions, enhancing the reliability of plume segmentation. In the emission flux rate estimation, aligning with previous studies, this study defined an uncertainty in ERA5 wind speed as 50 %, thus leading to a significant uncertainty in the estimated emission rate. To evaluate the uncertainty in the wind fields from ERA5 reanalysis, which is widely used in many previous studies, this study compared them with data from three ground-based meteorological sites in Shanxi over the concentrated point source areas (Fig. 11). The comparison results indicate that from 2021 to 2023, the overall bias in the ERA5 wind speed was approximately 1.30 m s⁻¹, which is close to 100 % of bias on average. It has been recognised that the wind speed should be in a moderate range to allow detectable plumes from space. A wind speed that is too small may hamper the development of the plume, while a wind speed that is too large may dilute the plume. It is observed in our cases that the wind speeds roughly fall within 0.5 to 2.5 m s⁻¹ for most days with detectable point source plumes. If we assume this is the suitable wind speed range for satellite detection, as shown in black dots in the upper panel of Fig. 11, the deviation is about 0.45–0.54 m s⁻¹, which is close to about 50 % of the wind speed from ERA5. This uncertainty is consistent with the assumption of wind speed uncertainty (50 %) in this study. In terms of wind direction, there are significant differences between ERA5 and the observations from meteorological sites. While ERA5 reanalysis data (at a height of 10 m) show relatively constant wind direction, the measurements of wind direction from meteorological stations show a much larger range. This discrepancy indicates significant deviations between ERA5 reanalysis wind fields and actual wind conditions, challenging their direct application in point source plume identification and emission estimation. Consequently, leveraging high-density and high-precision meteorological observations from automatic meteorological monitoring stations, especially over regions with complex surface properties, could reduce the uncertainty and enhance the accuracy of satellite-based detection and estimation of CH₄ point source emissions.

Figure 11Comparison of wind speeds (a–c) and directions (d–f) between ERA5 reanalysis and meteorological stations located in Yangquan, Changzhi, and Jincheng cities in Shanxi, as indicated in Fig. 1. Wind field data from 2021 to 2023 are extracted in the daytime and correspond to the GF-5B overpass time. Black dot data are selected according to the wind speed range of 0.5 to 2.5 m s⁻¹.

Download

A flat 50 % wind error could underestimate uncertainty for slow winds and overestimate uncertainty for fast winds. Therefore, we carried out an evaluation of the plume emission uncertainty using the absolute wind error (1.297 m s⁻¹ on average) estimated by comparing wind speeds from EAR5 and local meteorological stations in Shanxi. The results of CH₄ flux rates and their uncertainty are shown in Fig. E1. As we expected, the uncertainty in the flux decreased/increased at high/low wind speed, respectively. In addition, the impact of wind speed uncertainties accounts for approximately 86.31 %, which is similar to the previous result based on a flat 50 % wind error. This result supports the fact that wind speed remains the dominant factor contributing to the uncertainty in estimating CH₄ point source emissions.