the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 20 Dec 2023
A survey of methane point source emissions from coal mines in Shanxi province of China using AHSI on board Gaofen-5B
Abstract. Satellite-based detection of methane (CH4) point sources is crucial in identifying and mitigating anthropogenic emissions of CH4, a potent greenhouse gas. Previous studies have indicated the presence of CH4 point source emissions from coal mines in Shanxi, China, an important source region with large CH4 emissions, but a comprehensive survey has remained elusive. This study aims to conduct a survey of CH4 point sources over Shanxi's coal mines based on observations of the Advanced HyperSpectral Imager (AHSI) on board the Gaofen-5B satellite (GF-5B/AHSI) between 2021 and 2023. The spectral shift in center wavelength and change in full-width-half-maximum (FWHM) are estimated for all spectra channels, which are used as inputs for retrieving the enhancement of column-averaged dry-air mole fraction of CH4 (ΔXCH4) using a matched-filter based algorithm. Our results show that the spectral calibration on GF-5B/AHSI reduced estimation biases of emission flux rate by up to 5.0 %. We applied the flood-fill algorithm to automatically extract emission plumes from ΔXCH4 maps. We adopted the integrated mass enhancement (IME) model to estimate the emission flux rate values from each CH4 point source. Consequently, we detected CH4 point sources in 32 coal mines with 93 plume events in Shanxi province. The estimated emission flux rate ranges from 857.67 ± 207.34 kg·h-1 to 14333.02 ± 5249.32 kg·h-1. The total emission flux rate reaches 13.26 t·h-1 in Shanxi, assuming all point sources emit simultaneously. Our results show that wind speed is the dominant source of uncertainty contributing about 84.84 % to the total uncertainty in emission flux rate estimation. Interestingly, we found a number of false positive detections due to solar panels that are widely spread in Shanxi. This study also evaluates the accuracy of wind fields in ECMWF ERA5 reanalysis by comparing with ground-based meteorological station. We found large discrepancy, especially in wind direction, suggesting incorporating local meteorological measurements into the study CH4 point source are important to achieve high accuracy. The study demonstrates that GF-5B/AHSI possesses capabilities for monitoring large CH4 point sources over complex surface characteristics in Shanxi.
- Preprint
(8705 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2023-3047', Anonymous Referee #1, 24 Jan 2024
The authors report a survey of methane plumes from underground coal mines in the Shanxi province of China using the GF-5B/AHSI hyperspectral satellite instrument. Their study builds on a growing body of literature on satellite remote sensing of methane point sources from diverse regions and industrial sectors. To the best of my knowledge, it documents the largest dataset of Shanxi coal mine methane plumes to date, including 93 plumes from 32 mines. The study is interesting and well-designed. The authors document each step of the data processing pipeline from instrument spectral calibration to characterization of wind speed errors in source rate estimation, employing the latest plume mapping/quantification techniques to uncover new emissions in a challenging study area. The study’s main weakness is a lack of clarity in the description of some the methods and materials, which complicates evaluation of the conclusions. I recommend the study be accepted for publication after these issues are addressed.
Comments
- L. 13-14: Please clarify what shift in center wavelength and change in FWHM are being referred to here. I believe they are departures from the nominal design values, but this could also be interpreted as temporal drift over mission operating period.
- L. 21: It’s not clear what is meant by “simultaneously”. Later in the manuscript it is explained that the emissions are assumed to be simultaneous and continuous at the median observed rate. Does that include null detections? Please clarify so the reader understands from the outset.
- L. 56: “Canada’s GHGSat” is a bit odd here since GHGSat is a private company whereas the other platforms in the list (belonging to Italy, China, Germany, etc.) are national/public missions.
- L. 63-65: Please specify that those 37 sources were seen in just a handful of satellite passes.
- L. 77: It’s not clear how wind uncertainty would influence detection. I believe it’s through the flood-fill algorithm but that comes later in the manuscript. Please clarify.
- L. 84: Again, please clarify the nature of this shift/change.
- L. 97-100: Is the suggestion that the coal mine plumes in Shanxi are coming primarily from abandoned or illegal mines? That would be an interesting claim, but the satellite observations don’t seem to suggest it. Have you researched the operators/practices of the 32 mines you detected? If not and to avoid this analysis I would suggest revising this passage to include other explanations (normal mine venting, for example). All coal mines are known to emit methane, including in China; I don’t really see a reason to invoke abandonment or crime to explain emissions.
- Figure 1: What is the source of the landcover imagery in panel 1a?
- L. 140: I don’t think you defined “ILS” yet.
- Section 3.3.3: To clarify the IME/measurement uncertainty part of this section, please describe your implementation of the flood-fill algorithm in more detail. It wasn’t clear to me, for example, that it depended on defining a background. (And what is that background? See comments below.) Can you please lay this out in a few sentences?
- L. 226: A flat 50% wind error would underestimate uncertainty for slow winds and overestimate uncertainty for fast winds. Why not use a fixed absolute wind uncertainty? You have local measurements to compute such an error and that would seem more defensible than an arbitrary 50%.
- L. 227: “uncertainty of the used wind uncertainty”. I don’t understand this. Typo?
- Figure 3: It would be helpful to note the lat/lon of the source somewhere in the caption and main text. What do the black ellipses signify? I assume the inset times are local?
- L. 235: This is kind of an obvious statement – maybe not worth saying.
- Section 4.1: The discussion of Figure 3 would be easier to follow if the role of the “background” in the retrieval was better explained – see above comment on Sect 3.3.3. At this point it’s unclear to me whether this background is a 2D retrieval image background or a 1D spectral background. Please clarify.
- It’s also not clear how GF-5B/AHSI can image the same scene twice in under 10 seconds. Can you please explain the measurement strategy? Section 2.2 would be a good place for this.
- L. 255-257: The plumes sampled 8 seconds apart look the same, as expected – it’s just the retrieval noise that varies between passes. Why is that? Again, please explain the observing configuration.
- Figure 6: Please mark the source locations on the plot – in some cases I can’t tell where the plume starts/ends.
- L. 287-292: Assuming GF-5B/AHSI observes around 11:30 local time (please specify in Section 2.2), I wonder if the difference in orbit, with TROPOMI passing ~2 hours later, is another possible explanation. It also wasn’t clear to me why solar panels (which may be small compared to a TROPOMI pixel) would affect TROPOMI retrievals. Later in the manuscript, Fig. 10 shows that the solar installations are quite large, comparable in size to the TROPOMI footprint. This may be obvious to some readers, but it would be helpful to reference Fig. 10 here for others.
- L. 294-295: I don’t get what point is being made here. Also, I wonder if “background spectra” is the background relevant to the flood-fill algorithm?
- L. 301-302: I don’t follow this. Are you saying that the IME range is bigger than the Q range, and that’s because of variable wind speed?
- L. 304: This follows directly from the assumption of 50% wind error, so it’s a foregone conclusion. I suggest estimating a typical wind error in m/s for the Shanxi area.
- L. 310-311: This argument can only be made if you include null detections in the median/mean to obtain persistence-weighted mean/median emissions for each source. Is that being done?
- Section 4.3.1: I didn’t quite follow how the correction works. Does it simply substitute the estimated wavelength/FWHM values for the nominal values in the retrieval?
- Section 4.3.3: Earlier in the manuscript (L. 186-186), it is argued that wind direction data may be too uncertain to be useful for plume identification. But then how can it be useful for the flood-fill plume detection? This is inconsistent. Also, why not use the wind error statistics you find here instead of the flat 50% error for source rates?
Technical corrections
- 14: spectra → “spectral”
- 36: contribute to → “contribute”
- 55: EOS-1 → “EO-1”
- 78: haven → “have”
Citation: https://doi.org/10.5194/egusphere-2023-3047-RC1 - AC1: 'Reply on RC1', Zhonghua He, 25 Feb 2024
-
RC2: 'Comment on egusphere-2023-3047', Anonymous Referee #2, 12 Feb 2024
He et al. used the observations of Advanced HyperSpectral Imager (AHSI) on board the Gaofen-5B satellite (GF-5B/AHSI) to estimate methane emissions from coal mines in Shanxi province in China. The spectral shift in center wavelength and change in full-width-half-maximum (FWHM) was characterized to improve the accuracy of the spectra. Based on the improved dataset, the matched filter method was applied to calculate the enhancement ΔXCH4, which is followed by the use of the integrated mass enhancement (IME) model to estimate the methane emissions. Besides these, an automated plume segmentation method was adopted to reduce the dependence of subjective judgement during the data processing phase, and the major factors that affect the uncertainties of the estimates are discussed.
General comments:
- It is apparent that the estimates are sensitive to the selection of background, which is acknowledged in the manuscript, and is also demonstrated with a good example using the same background region in Fig.5a&b. The question is how the background regions were selected in practice? i.e., where is "a background region in square (with length of 600 pixels, which is 18 km)" located?
- Are the estimates of methane emissions correlated with wind speed? As the wind speed is low, the dispersion of the methane plumes may be very uncertain, and the estimates may be biased.
- I understand that the overall emission flux rate of 13.26 ton h-1 in Shanxi refers to the 32 coal mines between 2021 and 2023. How does it relate to the total coal mine emissions? How does it compare with the estimates from TROPOMI? What's the detection limit of GF-5B/AHSI?
detailed comments:
L138-139: the matched filter method and the IME model are actually combined to estimate the CH4 emission rates. Therefore, they should not be considered two methods.
L184: change "be very differ" to "be very different", what are the possible reasons of the mismatch?
L204-205: k may be affected by a few factors, such as surface pressure, temperature, and water vapor. How would these affect the estimate?
L227: change "evaluations" to "evaluation"
L235: not "emissions" but "the direction of plumes"
L241-243: "as the plumes appears at different locations of the imaging scene. The plumes appear at the bottom of the scene in Figure (f) and at the top in Figure (g)", why does the position of the plumes in the scenes matter?
L310: 13.26 t*24*365 /yr
L356: consider combining or to combine
Citation: https://doi.org/10.5194/egusphere-2023-3047-RC2 - AC2: 'Reply on RC2', Zhonghua He, 25 Feb 2024
Status: closed
-
RC1: 'Comment on egusphere-2023-3047', Anonymous Referee #1, 24 Jan 2024
The authors report a survey of methane plumes from underground coal mines in the Shanxi province of China using the GF-5B/AHSI hyperspectral satellite instrument. Their study builds on a growing body of literature on satellite remote sensing of methane point sources from diverse regions and industrial sectors. To the best of my knowledge, it documents the largest dataset of Shanxi coal mine methane plumes to date, including 93 plumes from 32 mines. The study is interesting and well-designed. The authors document each step of the data processing pipeline from instrument spectral calibration to characterization of wind speed errors in source rate estimation, employing the latest plume mapping/quantification techniques to uncover new emissions in a challenging study area. The study’s main weakness is a lack of clarity in the description of some the methods and materials, which complicates evaluation of the conclusions. I recommend the study be accepted for publication after these issues are addressed.
Comments
- L. 13-14: Please clarify what shift in center wavelength and change in FWHM are being referred to here. I believe they are departures from the nominal design values, but this could also be interpreted as temporal drift over mission operating period.
- L. 21: It’s not clear what is meant by “simultaneously”. Later in the manuscript it is explained that the emissions are assumed to be simultaneous and continuous at the median observed rate. Does that include null detections? Please clarify so the reader understands from the outset.
- L. 56: “Canada’s GHGSat” is a bit odd here since GHGSat is a private company whereas the other platforms in the list (belonging to Italy, China, Germany, etc.) are national/public missions.
- L. 63-65: Please specify that those 37 sources were seen in just a handful of satellite passes.
- L. 77: It’s not clear how wind uncertainty would influence detection. I believe it’s through the flood-fill algorithm but that comes later in the manuscript. Please clarify.
- L. 84: Again, please clarify the nature of this shift/change.
- L. 97-100: Is the suggestion that the coal mine plumes in Shanxi are coming primarily from abandoned or illegal mines? That would be an interesting claim, but the satellite observations don’t seem to suggest it. Have you researched the operators/practices of the 32 mines you detected? If not and to avoid this analysis I would suggest revising this passage to include other explanations (normal mine venting, for example). All coal mines are known to emit methane, including in China; I don’t really see a reason to invoke abandonment or crime to explain emissions.
- Figure 1: What is the source of the landcover imagery in panel 1a?
- L. 140: I don’t think you defined “ILS” yet.
- Section 3.3.3: To clarify the IME/measurement uncertainty part of this section, please describe your implementation of the flood-fill algorithm in more detail. It wasn’t clear to me, for example, that it depended on defining a background. (And what is that background? See comments below.) Can you please lay this out in a few sentences?
- L. 226: A flat 50% wind error would underestimate uncertainty for slow winds and overestimate uncertainty for fast winds. Why not use a fixed absolute wind uncertainty? You have local measurements to compute such an error and that would seem more defensible than an arbitrary 50%.
- L. 227: “uncertainty of the used wind uncertainty”. I don’t understand this. Typo?
- Figure 3: It would be helpful to note the lat/lon of the source somewhere in the caption and main text. What do the black ellipses signify? I assume the inset times are local?
- L. 235: This is kind of an obvious statement – maybe not worth saying.
- Section 4.1: The discussion of Figure 3 would be easier to follow if the role of the “background” in the retrieval was better explained – see above comment on Sect 3.3.3. At this point it’s unclear to me whether this background is a 2D retrieval image background or a 1D spectral background. Please clarify.
- It’s also not clear how GF-5B/AHSI can image the same scene twice in under 10 seconds. Can you please explain the measurement strategy? Section 2.2 would be a good place for this.
- L. 255-257: The plumes sampled 8 seconds apart look the same, as expected – it’s just the retrieval noise that varies between passes. Why is that? Again, please explain the observing configuration.
- Figure 6: Please mark the source locations on the plot – in some cases I can’t tell where the plume starts/ends.
- L. 287-292: Assuming GF-5B/AHSI observes around 11:30 local time (please specify in Section 2.2), I wonder if the difference in orbit, with TROPOMI passing ~2 hours later, is another possible explanation. It also wasn’t clear to me why solar panels (which may be small compared to a TROPOMI pixel) would affect TROPOMI retrievals. Later in the manuscript, Fig. 10 shows that the solar installations are quite large, comparable in size to the TROPOMI footprint. This may be obvious to some readers, but it would be helpful to reference Fig. 10 here for others.
- L. 294-295: I don’t get what point is being made here. Also, I wonder if “background spectra” is the background relevant to the flood-fill algorithm?
- L. 301-302: I don’t follow this. Are you saying that the IME range is bigger than the Q range, and that’s because of variable wind speed?
- L. 304: This follows directly from the assumption of 50% wind error, so it’s a foregone conclusion. I suggest estimating a typical wind error in m/s for the Shanxi area.
- L. 310-311: This argument can only be made if you include null detections in the median/mean to obtain persistence-weighted mean/median emissions for each source. Is that being done?
- Section 4.3.1: I didn’t quite follow how the correction works. Does it simply substitute the estimated wavelength/FWHM values for the nominal values in the retrieval?
- Section 4.3.3: Earlier in the manuscript (L. 186-186), it is argued that wind direction data may be too uncertain to be useful for plume identification. But then how can it be useful for the flood-fill plume detection? This is inconsistent. Also, why not use the wind error statistics you find here instead of the flat 50% error for source rates?
Technical corrections
- 14: spectra → “spectral”
- 36: contribute to → “contribute”
- 55: EOS-1 → “EO-1”
- 78: haven → “have”
Citation: https://doi.org/10.5194/egusphere-2023-3047-RC1 - AC1: 'Reply on RC1', Zhonghua He, 25 Feb 2024
-
RC2: 'Comment on egusphere-2023-3047', Anonymous Referee #2, 12 Feb 2024
He et al. used the observations of Advanced HyperSpectral Imager (AHSI) on board the Gaofen-5B satellite (GF-5B/AHSI) to estimate methane emissions from coal mines in Shanxi province in China. The spectral shift in center wavelength and change in full-width-half-maximum (FWHM) was characterized to improve the accuracy of the spectra. Based on the improved dataset, the matched filter method was applied to calculate the enhancement ΔXCH4, which is followed by the use of the integrated mass enhancement (IME) model to estimate the methane emissions. Besides these, an automated plume segmentation method was adopted to reduce the dependence of subjective judgement during the data processing phase, and the major factors that affect the uncertainties of the estimates are discussed.
General comments:
- It is apparent that the estimates are sensitive to the selection of background, which is acknowledged in the manuscript, and is also demonstrated with a good example using the same background region in Fig.5a&b. The question is how the background regions were selected in practice? i.e., where is "a background region in square (with length of 600 pixels, which is 18 km)" located?
- Are the estimates of methane emissions correlated with wind speed? As the wind speed is low, the dispersion of the methane plumes may be very uncertain, and the estimates may be biased.
- I understand that the overall emission flux rate of 13.26 ton h-1 in Shanxi refers to the 32 coal mines between 2021 and 2023. How does it relate to the total coal mine emissions? How does it compare with the estimates from TROPOMI? What's the detection limit of GF-5B/AHSI?
detailed comments:
L138-139: the matched filter method and the IME model are actually combined to estimate the CH4 emission rates. Therefore, they should not be considered two methods.
L184: change "be very differ" to "be very different", what are the possible reasons of the mismatch?
L204-205: k may be affected by a few factors, such as surface pressure, temperature, and water vapor. How would these affect the estimate?
L227: change "evaluations" to "evaluation"
L235: not "emissions" but "the direction of plumes"
L241-243: "as the plumes appears at different locations of the imaging scene. The plumes appear at the bottom of the scene in Figure (f) and at the top in Figure (g)", why does the position of the plumes in the scenes matter?
L310: 13.26 t*24*365 /yr
L356: consider combining or to combine
Citation: https://doi.org/10.5194/egusphere-2023-3047-RC2 - AC2: 'Reply on RC2', Zhonghua He, 25 Feb 2024
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 281 | 99 | 24 | 404 | 9 | 7 |
- HTML: 281
- PDF: 99
- XML: 24
- Total: 404
- BibTeX: 9
- EndNote: 7
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1