the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 19 Dec 2023
Strength of TROPOMI satellite observations in retrieving hourly resolved sources of volcanic sulfur dioxide by inverse modeling
Abstract. Volcanic eruptions release sulfur dioxide (SO2), impacting air quality, ecosystems, and aviation. To comprehensively assess these effects, high-temporal-resolution SO2 emission data is crucial. In this study, we use an inverse modeling procedure, assimilating SO2 column measurements from TROPOMI and OMPS low-Earth orbit satellites into an Eulerian chemistry-transport model. This procedure allows us to derive precise hourly SO2 mass flux and injection heights. TROPOMI, with its exceptional spatial resolution, excels at detecting short-lived, concentrated SO2 plumes near the source shortly before satellite overpasses. This high-resolution data enables more robust identification and precise characterization of strong SO2 emissions, surpassing the capabilities of lower-resolution OMPS measurements, which may overlook or underestimate vigorous degassing periods. Notably, this high-resolution data also facilitates the detection of pre-eruptive SO2 emissions. Cloud cover can obscure SO2 plumes from satellite observations, but our inverse modeling procedure effectively distinguishes and tracks them by assimilating successive satellite overpass data. Furthermore, this procedure proves less susceptible to ash emissions compared to geostationary Himawari-8/AHI observations. We apply our methodology to study the 2018 Ambrym eruption, a former major volcanic SO2 emitter. This eruption marked the end of long-lived lava lake activity and initiated a submarine eruption through a massive magma intrusion. Our detailed SO2 flux time series unveils the evolution of the eruption and identifies distinct SO2 sources, including lava flows and shallow magma intrusions. In summary, the assimilation of TROPOMI data into inverse modeling procedures offers significant potential for enhancing our understanding of magma transport and environmental impacts during volcanic eruptions.
- Preprint
(43925 KB) - Metadata XML
-
Supplement
(143278 KB) - BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2023-2545', Ben Esse, 19 Jan 2024
General comments
---------------------------------------------------------------------------------------
In this manuscript the authors present hourly resolved SO2 emissions from the 2018 eruption of Ambrym volcano using an inverse modelling procedure based on data assimilation of multiple low earth orbit satellite images over a series of days. In particular, the authors highlight that the high spatial resolution and sensitivity of TROPOMI provide significant advantages over lower spatial resolution instruments (here OMPS). These are:
- Capturing short-lived SO2 emissions
- Less impact from the presence of volcanic ash in the plume
- Heightened sensitivity to pre-eruption emissions
- Capturing SO2 emissions obscured by meteorological cloud
They use the presented SO2 emissions to argue two sources of SO2 in this eruption, firstly from the emitted lava flows and a second paroxysmal emission from the emplacement of a dyke towards the end of the eruption. The topic of this eruption is very interesting: Ambrym was the largest single emitter of SO2 globally until 2018, with no significant activity since this eruption. The process of this transition is very worthy of investigation, and the results of the sharp peak in SO2 emission seen later in the eruption after the lava flows is very interesting.
However, I am not convinced that the results shown in the paper fully support the conclusions presented. For this reason, I would suggest a major revision to present additional data to fully address the issues detailed below.
Specific comments
---------------------------------------------------------------------------------------
I have split my specific comments into major and minor comments, the first addressing the overall conclusions and the second addressing specific points in the manuscript that I believe require further clarification.
Major comments
---------------------------------------------------------------------------------------
The first advantage presented for TROPOMI is that it is more sensitive to short-lived SO2 emissions. There are two main pieces of evidence presented for this: the fact that OMPS misses completely the “paroxysmal degassing” seen by TROPOMI and Himawari-8 around midnight of 16th December, and that by smoothing the TROPOMI time series they can approach the OMPS time series (figure 7). The authors suggest that this is because the lower spatial resolution of OMPS means that the high vertical column densities (VCDs) seen by TROPOMI close to the volcano are smoothed out. However, SO2 mass should not be lost in this smoothing process, so I would expect to see a lower but broader peak in the emissions from OMPS compared to the high but narrow peak from TROPOMI, such as seen when smoothing the TROPOMI time series in figure 7. Instead, the OMPS SO2 emission has a trough at this time. Additionally, I would expect that this peak in emissions would be preserved in subsequent images even if it is not resolved in the overpass just after the emission, which is one of the benefits of this data assimilation approach. I think that this requires further investigation. Perhaps the authors could re-run their analysis on the spatially smoothed TROPOMI data shown in figure 4 to see if the OMPS time series is recreated when at a similar resolution?
The second advantage of being less effected by volcanic ash is an excellent strength of this inversion approach, allowing the ash to diffuse and gravitationally settle compared to the observations in the very near field.
The third advantage highlighted was that TROPOMI is sensitive to pre-eruptive SO2 emissions, discussed in section 3.1.4. Here, the pre-eruptive SO2 emission is constrained by a few pixels detected in the edge of the swath on both 14th and 15th December. However, what is not clear to me is why the main SO2 plumes seen closer to the volcano on 13th and 14th December do not result in any SO2 emission in the reconstructed time series, when these few scattered pixels on the swath edge do? These emissions are also injected at very high altitude in the reconstructed emissions. The authors suggest that this is due to the model having to smear out dense emissions to match the observation, however changing wind speed and direction with altitude would mean that if this was injected lower in altitude then the emission time would likely have to be very different to transport plume to the location shown, which is not discussed.
The fourth advantage presented is that the method can reconstruct SO2 plumes below meteorological cloud. In figure 6, the authors compare the observed and modeled plumes for each sensor on three days, marking areas of cloud cover. However, it is not clear how these areas of cloud are defined. Plotting the cloud fraction parameter within the TROPOMI product on top of the SO2 VCD values shows that there is some cloud coverage that appears to block portions of the plume, however the cloudy regions do not align with those drawn. I have plotted one example attached for 16th December, (SO2 VCD coloured red/blue, cloud in grey, higher cloud is a darker shade) which shows significant area of clear sky between the plume and the cloud to the north. This is not reflected in the region shown in figure 6, and I would expect to see SO2 to the north of the plume if it was where the model suggests.
Another major comment is on the interpretation of the paroxysmal phase of degassing. The authors state that the source of this emission is from SO2 degassed from the dyke intruded on 15th – 17th December (as discussed in Shreve et al. (2019)). What is not clear to me is physically how this SO2 is transported from the dyke to the surface, and where it is emitted? This section requires further explanation on the mechanisms of this release.
My final major comment is on a lack of discussion of uncertainty. There are uncertainty estimates for the total masses given (L265 and L506) but there is no discussion on how these errors are derived and no errors given in any other values. I feel the manuscript requires these to be added, both to values in the text and to the figures of the emission time series.
Minor comments
---------------------------------------------------------------------------------------
Below is a list of minor comments in order of appearance in the manuscript:
- L31: “SO2 is an unambiguous indicator of volcanic plume” I wouldn’t necessarily agree with this, as SO2 plumes from anthropogenic sources are often visible in satellite imagery.
- L49-50: Assuming the inputs are daily data from LEO sensors, the delta-M method does not provide sub-daily emission rates, I would suggest rewording this.
- L54: The stacking method does not have to be monthly; it can be whatever time resolution desired.
- L56: The delta-M method does not require wind field data to my knowledge.
- L62: Back-trajectory analysis only has issues with re-circulating plumes in some situations with large eruption plumes, and this issue is not unique to this method. In particular, in the paper by Queißer et al. (2019) recirculation is not highlighted as a drawback with this approach, so I would suggest rewording this.
- L66: Plume injection altitude can also be used as an a priori.
- L99-103: The discussion of conversion of slant columns to vertical columns needs more information. Specifically, the TM5 profile is not usually used in volcanic applications, this is typically the 1, 7, and 15 km box profiles, which are not mentioned here. Also, the air mass factors computed and contained within the TROPOMI product files include information on the scattering weighting function and reflecting surface, these are not separate parameters. If discussing them separately, I would highlight that it is the geometrical air mass factor that is combined with these other factors.
- L120: Here you say that you use the 7 km VCD product, however the plumes you measure are injected at a wide range of altitudes. How do you account for the changing sensitivity to SO2 with altitude? This could dramatically impact the reconstruction results, especially with respect to the issue with weak emissions tending to result in abnormally high injection altitudes (section 4.2.2).
- L122-125: The discussion of the thresholding at the swath edges is not fully clear. You remove any VCD > 1 DU, but then “set a specific threshold for the SO2 column values of pixels at the swath edge” and manually tune this threshold to 1.1 – 1.4 DU. Does this mean that you discard any pixels that are 1 < VCD < 1.1 (if using a 1.1 DU threshold)? Also, how did tuning this threshold impact the detection of plume in the swath edge attributed to pre-eruptive degassing?
- L153-154: Why limit the analysis to above 2 km? There appear to be emissions lower than this, so would it not be better to include these instead of disregarding them?
- L236-237: You state “a clear correlation emerges with high values of the lava flow indices”. Although there is some similarity (peak at 00:00 on 15th December, both decay from 12:00), the peak in the lava flow proxy coincides with a trough in emissions, so I do not think this sentence is necessarily valid.
- L297: “TROPOMI’s hyperspectral…” – do you mean here that TROPOMI has a higher spectral resolution (~0.5 vs ~ 1 nm)? I am not sure how much difference the spectral resolution makes, for example TROPOMI has a similar spectral resolution to OMI. The higher sensitivity is driven primarily by the spatial resolution.
- L299-309: You show here that by spatially smoothing the TROPOMI data to the spatial resolution of OMPs you can recreate similar VCDs, but is it not the total mass in view that is more important? Smoothing the data out should conserve the total mass in the view, and so the emission rate reconstructed should be the same when integrated with time. Have you tried reconstructing the emission rates with the smoothed TROPOMI data to see if it matches the OMPS time series?
- L332: Can you expand on the phrase “In contrast, ash remained at lower altitudes due to wind shear”? Wind shear would explain the separation of ash and SO2 if injected at different altitudes, but it would not cause the ash to be at lower altitudes.
- L361: It is not clear here how the high spatial resolution of TROPOMI helps to detect gas below cloud. If I understand correctly, this is achieved by using multiple images over several days such that any blocked plume is visible in other scenes. So OMPS should also show this behaviour (and indeed the model results look broadly similar for TROPOMI and OMPS in terms of spatial extent).
- L366: “yet faint SO2 signals are visible in TROPOMI observations”. I am not sure I agree here from the figures shown. By eye, the SO2 in the red regions looks of a similar level to elsewhere in the image outside the plume. Can you show that the level of the SO2 VCDs in this region are above the background noise?
- L366: How are the red contoured areas defined and what is the source of the cloud data? As shown above, the cloud product within the TROPOMI data, which appears to map well with seen gaps in the plume, do not show significant cloud cover for all regions marked.
- L371: If OMPS is not able to detect the SO2 below the cloud, then why does the model create emissions in this region? What information is it using to place the plume here that the model with TROPOMI is not?
- L392: Why is the 7–8-hour solution picked? By eye I would argue that the 4-hour solution is a better fit, I suspect that the better correlation in the 8-hour smoothing is driven primarily by the paroxysmal peak missing in the OMPS time series. Also, the smoothed results appear to be shifted in time (the peak in the green data shifts later the longer the smoothing window applied). Is this an artefact of the smoothing?
- L403: Many of the plots shown in figure S15 show quite an asymmetric distribution, so fitting a Gaussian function does not seem to work well (e.g. panels a, c, d). I would instead suggest fitting an asymmetric function to better capture the difference in the positive and negative values.
- L449-450: In addition to spreading out a plume to obtain lower VCDs, injecting the plume at higher altitude will have an impact on its location due to wind shear. Is it feasible for winds at lower altitudes to have transported the pre-eruptive plume that you attribute to the pixels in the swath edge, or is this only possible at higher altitudes? If so, then this would suggest that this is not pre-eruptive emissions.
- L452: When you say “larger wind fields”, do you mean higher velocity?
- L453-458: Is it possible that using the 7 km VCD product is artificially pushing altitudes higher due to this effect? The retrieved VCD in the 7 km product will be lower than for the plume measured if the reported injection altitudes are similar to the plume altitudes at the time of measurement, so a higher altitude/faster wind may have been selected by the model to account for this.
- L460: “greatly aids in accurately capturing emission timing” – on this point, if the altitude is incorrect then the emission timing will also be off due to the difference in wind speed. Have you considered the magnitude of this error?
- L462-464: It is worth noting that the plume height retrievals only work for strong plumes, so may not help for constraining weaker emissions.
- L470-471: No mention is made of the trough in SO2 emissions at the peak in lava flow index.
- L485-486: Can you expand on this further? No comment on the physical mechanism of transporting this SO2 from depth to the surface, nor if it came from the lava lake or the lava flow region (or somewhere else?). I would also note that the dyke intrusion lasted ~3 days, so why is the SO2 emission such a sharp peak? Are there any ground-based observations to support this? Shreve et al. (2019) also note that the total magnitude of SO2 emissions matches that expected from the volume of the lava flow, so that contributions from the dyke were not significant. I agree that the timing of this peak in SO2 emission after the lava flow is interesting and may provide insights into the magmatic processes, but this needs further explanation.
- Figure 1: in panel bii the #1 arrow is pointing to an empty region near the plume, should this be pointing to the pre-eruptive plume in the swath edge?
Technical comments
---------------------------------------------------------------------------------------
- L82: Consider replacing “emissions” with “emission”
- L212: You refer to points in figures 1 and 2 as “T2” and “O2”, for example, but in the figures these are just numbers (e.g. #1, #2). It would be clearer if these were consistent (i.e. use #T1 in figure 1, #O1 in figure 2).
- L299: “SO2” is missing a subscript here.
- Many figures use non-perceptually uniform colourmaps, which can be misleading or difficult to interpret, especially for colourblind people or when printed in black and white (Crameri et al., 2020). I would suggest replacing the colourmaps used throughout with colourblind-friendly versions.
- Satellite figures: The figures of satellite data all have elliptical pixels for the data. Why is this used instead of a continuous grid of rectangular pixels? This would avoid the overlapping in the OMPS data.
- Figures S8-11: How have the sub-frames been ordered? They do not appear to be in terms of number, time, altitude or strength of emission and interpreting these plots was difficult. Would it be possible to reorder these, or have I missed the ordering?
References
---------------------------------------------------------------------------------------
Crameri, F., Shephard, G.E., Heron, P.J., 2020. The misuse of colour in science communication. Nat. Commun. 11, 5444. https://doi.org/10.1038/s41467-020-19160-7
-
RC2: 'Comment on egusphere-2023-2545', Anonymous Referee #2, 01 Feb 2024
Overview
The authors compare using either TROPOMI or OMPS total column SO2 retrievals in an inverse modelling approach (CHIMERE) to infer the SO2 emissions flux and plume height data of the Ambrym volcano during the eruption in December 2018. They compare their results with SO2 flux proxy data derived from Himawari-8/AHI. The find a better performance when the TROPOMI data, which have a higher spatial resolution.
General comments
The paper seems to lack a scientifically sound method to quantify and compare the inversion result using either TROPOMI or OMPS. It remains often unclear, why they can conclude the strong superiority of TROPOMI data. The recommended approach would be to use the derived emissions parameters (flux and injection height) in a forward simulation and to develop a metric to quantify to what extend the simulates plumes agree with observations. It is not satisfactory to only juxtapose maps of plumes with HIMAWARI proxy SO2 data.
While the authors come to the plausible conclusion that TROPOMI data are much better, it remains unclear how this conclusion is derived especially considering this lack of rigor mentioned above. It should also be emphasized more that they study only one specific episode using their specific modeling framework. I recommend using a more neutral language when describing the results.
The inversion approach should be explained in more detail from a practical perspective. In particular the robustness of the derived plume heights remains unclear. It is also not clear, what prior information was used and how the ensemble of tracer plumes injected at different height is part of the framework.
The paper should provide in a numerical way the derived time series of flux and injection height to allow the scientific community to use the data. If applicable, the result should be compared to the results of other authors.
The abstract needs to be strongly revised to report in more detail the findings and to reduce the number of the more general or introductory statements.
The paper should get a more precise title such as “Inversion of SO2 emissions from Ambrym 2018 using TROPOMI or OMPS retrievals.”
The Figure captions are too long because they interpret the shown results. This should be done in the body text. On the other hand, to little explanation of the meaning of the different lines of the time series is provided. I recommend showing the emission parameter time series as separate plots, not mixed with the maps.
Specific Comments
L10 Please mention, how this was achieved.
L13 Please provide some actual numbers of the emissions here.
L 129 It needs to be properly discussed that the subject comparison with HIMAWARI is the main reference for your study.
152 A model top of 200 hPa seems too low to simulate the fate of many volcanic SO2 plumes. Are you sure the plume is always located below that height ?
L 154 Please explain in more detail how the tracer ensemble is used in the inversion framework.
L163 Does the state vector include the emissions, or is the state only the atmospheric concentration.
L 170 Please introduce the emissions in your framework here.
L 160 Please add more information about the time stepping and assimilation window length in the section. What method do you use – a ensemble (KF) or variational approach.
L245 This is the first time you mention an indication that TROPOMI agrees better with Himawari than OMPS.
L 299 This sensitivity study should be given more attention, perhaps its own section. Was it done only for the outgassing or also for the eruptive events.
L311 Discuss the relation to the model resolution.
L322 How do you know that TROPOMI does not overestimate this value ?
L333 Is this bimodal injection height something that has been observed before - or is it an artefact of the assimilation, which does not get enough information about the vertical distribution of the plume.
L336 Please provide more details. HYSPLIT has not been introduced before to be part of the study. Did you run the model yourself?
L343 If HIMAWARI is you reference, why is the better agreement between OMPS and HIMAWARI not a good result for OMPS?
L375 This is a very controversial statement that needs more backing up with evidence. Given the high variability of the plumes, using the observations from the previous day to substitute cloud-contaminated observations seems risky.
L416 Why initialized? The emissions should be injected during the model run.
L487 This conclusion is not clear in this context.
L495 Provide more reasons.
L496 It is not clear why a high spatial resolution is helpful for capturing short - lived (short?) emissions. The overpass intervals is the same for TROPOMI and OMPS.
L501 That is a very general statement, please provide evidence from the study.
L 503 Again, I am not convinced that has been proven universally in that paper. Please explain if you mean the approach to use yesterday’s observations to gap fill.
L 508 Compare that to values from the literature.
Figure 1,2 Consider putting the time series in a separate picture. Please explain in more detail all 4 lines in the time series plot.
Citation: https://doi.org/10.5194/egusphere-2023-2545-RC2 -
AC1: 'Comment on egusphere-2023-2545', Abhinna Behera, 20 Apr 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-2545/egusphere-2023-2545-AC1-supplement.pdf
-
AC1: 'Comment on egusphere-2023-2545', Abhinna Behera, 20 Apr 2024
-
AC1: 'Comment on egusphere-2023-2545', Abhinna Behera, 20 Apr 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-2545/egusphere-2023-2545-AC1-supplement.pdf
Status: closed
-
RC1: 'Comment on egusphere-2023-2545', Ben Esse, 19 Jan 2024
General comments
---------------------------------------------------------------------------------------
In this manuscript the authors present hourly resolved SO2 emissions from the 2018 eruption of Ambrym volcano using an inverse modelling procedure based on data assimilation of multiple low earth orbit satellite images over a series of days. In particular, the authors highlight that the high spatial resolution and sensitivity of TROPOMI provide significant advantages over lower spatial resolution instruments (here OMPS). These are:
- Capturing short-lived SO2 emissions
- Less impact from the presence of volcanic ash in the plume
- Heightened sensitivity to pre-eruption emissions
- Capturing SO2 emissions obscured by meteorological cloud
They use the presented SO2 emissions to argue two sources of SO2 in this eruption, firstly from the emitted lava flows and a second paroxysmal emission from the emplacement of a dyke towards the end of the eruption. The topic of this eruption is very interesting: Ambrym was the largest single emitter of SO2 globally until 2018, with no significant activity since this eruption. The process of this transition is very worthy of investigation, and the results of the sharp peak in SO2 emission seen later in the eruption after the lava flows is very interesting.
However, I am not convinced that the results shown in the paper fully support the conclusions presented. For this reason, I would suggest a major revision to present additional data to fully address the issues detailed below.
Specific comments
---------------------------------------------------------------------------------------
I have split my specific comments into major and minor comments, the first addressing the overall conclusions and the second addressing specific points in the manuscript that I believe require further clarification.
Major comments
---------------------------------------------------------------------------------------
The first advantage presented for TROPOMI is that it is more sensitive to short-lived SO2 emissions. There are two main pieces of evidence presented for this: the fact that OMPS misses completely the “paroxysmal degassing” seen by TROPOMI and Himawari-8 around midnight of 16th December, and that by smoothing the TROPOMI time series they can approach the OMPS time series (figure 7). The authors suggest that this is because the lower spatial resolution of OMPS means that the high vertical column densities (VCDs) seen by TROPOMI close to the volcano are smoothed out. However, SO2 mass should not be lost in this smoothing process, so I would expect to see a lower but broader peak in the emissions from OMPS compared to the high but narrow peak from TROPOMI, such as seen when smoothing the TROPOMI time series in figure 7. Instead, the OMPS SO2 emission has a trough at this time. Additionally, I would expect that this peak in emissions would be preserved in subsequent images even if it is not resolved in the overpass just after the emission, which is one of the benefits of this data assimilation approach. I think that this requires further investigation. Perhaps the authors could re-run their analysis on the spatially smoothed TROPOMI data shown in figure 4 to see if the OMPS time series is recreated when at a similar resolution?
The second advantage of being less effected by volcanic ash is an excellent strength of this inversion approach, allowing the ash to diffuse and gravitationally settle compared to the observations in the very near field.
The third advantage highlighted was that TROPOMI is sensitive to pre-eruptive SO2 emissions, discussed in section 3.1.4. Here, the pre-eruptive SO2 emission is constrained by a few pixels detected in the edge of the swath on both 14th and 15th December. However, what is not clear to me is why the main SO2 plumes seen closer to the volcano on 13th and 14th December do not result in any SO2 emission in the reconstructed time series, when these few scattered pixels on the swath edge do? These emissions are also injected at very high altitude in the reconstructed emissions. The authors suggest that this is due to the model having to smear out dense emissions to match the observation, however changing wind speed and direction with altitude would mean that if this was injected lower in altitude then the emission time would likely have to be very different to transport plume to the location shown, which is not discussed.
The fourth advantage presented is that the method can reconstruct SO2 plumes below meteorological cloud. In figure 6, the authors compare the observed and modeled plumes for each sensor on three days, marking areas of cloud cover. However, it is not clear how these areas of cloud are defined. Plotting the cloud fraction parameter within the TROPOMI product on top of the SO2 VCD values shows that there is some cloud coverage that appears to block portions of the plume, however the cloudy regions do not align with those drawn. I have plotted one example attached for 16th December, (SO2 VCD coloured red/blue, cloud in grey, higher cloud is a darker shade) which shows significant area of clear sky between the plume and the cloud to the north. This is not reflected in the region shown in figure 6, and I would expect to see SO2 to the north of the plume if it was where the model suggests.
Another major comment is on the interpretation of the paroxysmal phase of degassing. The authors state that the source of this emission is from SO2 degassed from the dyke intruded on 15th – 17th December (as discussed in Shreve et al. (2019)). What is not clear to me is physically how this SO2 is transported from the dyke to the surface, and where it is emitted? This section requires further explanation on the mechanisms of this release.
My final major comment is on a lack of discussion of uncertainty. There are uncertainty estimates for the total masses given (L265 and L506) but there is no discussion on how these errors are derived and no errors given in any other values. I feel the manuscript requires these to be added, both to values in the text and to the figures of the emission time series.
Minor comments
---------------------------------------------------------------------------------------
Below is a list of minor comments in order of appearance in the manuscript:
- L31: “SO2 is an unambiguous indicator of volcanic plume” I wouldn’t necessarily agree with this, as SO2 plumes from anthropogenic sources are often visible in satellite imagery.
- L49-50: Assuming the inputs are daily data from LEO sensors, the delta-M method does not provide sub-daily emission rates, I would suggest rewording this.
- L54: The stacking method does not have to be monthly; it can be whatever time resolution desired.
- L56: The delta-M method does not require wind field data to my knowledge.
- L62: Back-trajectory analysis only has issues with re-circulating plumes in some situations with large eruption plumes, and this issue is not unique to this method. In particular, in the paper by Queißer et al. (2019) recirculation is not highlighted as a drawback with this approach, so I would suggest rewording this.
- L66: Plume injection altitude can also be used as an a priori.
- L99-103: The discussion of conversion of slant columns to vertical columns needs more information. Specifically, the TM5 profile is not usually used in volcanic applications, this is typically the 1, 7, and 15 km box profiles, which are not mentioned here. Also, the air mass factors computed and contained within the TROPOMI product files include information on the scattering weighting function and reflecting surface, these are not separate parameters. If discussing them separately, I would highlight that it is the geometrical air mass factor that is combined with these other factors.
- L120: Here you say that you use the 7 km VCD product, however the plumes you measure are injected at a wide range of altitudes. How do you account for the changing sensitivity to SO2 with altitude? This could dramatically impact the reconstruction results, especially with respect to the issue with weak emissions tending to result in abnormally high injection altitudes (section 4.2.2).
- L122-125: The discussion of the thresholding at the swath edges is not fully clear. You remove any VCD > 1 DU, but then “set a specific threshold for the SO2 column values of pixels at the swath edge” and manually tune this threshold to 1.1 – 1.4 DU. Does this mean that you discard any pixels that are 1 < VCD < 1.1 (if using a 1.1 DU threshold)? Also, how did tuning this threshold impact the detection of plume in the swath edge attributed to pre-eruptive degassing?
- L153-154: Why limit the analysis to above 2 km? There appear to be emissions lower than this, so would it not be better to include these instead of disregarding them?
- L236-237: You state “a clear correlation emerges with high values of the lava flow indices”. Although there is some similarity (peak at 00:00 on 15th December, both decay from 12:00), the peak in the lava flow proxy coincides with a trough in emissions, so I do not think this sentence is necessarily valid.
- L297: “TROPOMI’s hyperspectral…” – do you mean here that TROPOMI has a higher spectral resolution (~0.5 vs ~ 1 nm)? I am not sure how much difference the spectral resolution makes, for example TROPOMI has a similar spectral resolution to OMI. The higher sensitivity is driven primarily by the spatial resolution.
- L299-309: You show here that by spatially smoothing the TROPOMI data to the spatial resolution of OMPs you can recreate similar VCDs, but is it not the total mass in view that is more important? Smoothing the data out should conserve the total mass in the view, and so the emission rate reconstructed should be the same when integrated with time. Have you tried reconstructing the emission rates with the smoothed TROPOMI data to see if it matches the OMPS time series?
- L332: Can you expand on the phrase “In contrast, ash remained at lower altitudes due to wind shear”? Wind shear would explain the separation of ash and SO2 if injected at different altitudes, but it would not cause the ash to be at lower altitudes.
- L361: It is not clear here how the high spatial resolution of TROPOMI helps to detect gas below cloud. If I understand correctly, this is achieved by using multiple images over several days such that any blocked plume is visible in other scenes. So OMPS should also show this behaviour (and indeed the model results look broadly similar for TROPOMI and OMPS in terms of spatial extent).
- L366: “yet faint SO2 signals are visible in TROPOMI observations”. I am not sure I agree here from the figures shown. By eye, the SO2 in the red regions looks of a similar level to elsewhere in the image outside the plume. Can you show that the level of the SO2 VCDs in this region are above the background noise?
- L366: How are the red contoured areas defined and what is the source of the cloud data? As shown above, the cloud product within the TROPOMI data, which appears to map well with seen gaps in the plume, do not show significant cloud cover for all regions marked.
- L371: If OMPS is not able to detect the SO2 below the cloud, then why does the model create emissions in this region? What information is it using to place the plume here that the model with TROPOMI is not?
- L392: Why is the 7–8-hour solution picked? By eye I would argue that the 4-hour solution is a better fit, I suspect that the better correlation in the 8-hour smoothing is driven primarily by the paroxysmal peak missing in the OMPS time series. Also, the smoothed results appear to be shifted in time (the peak in the green data shifts later the longer the smoothing window applied). Is this an artefact of the smoothing?
- L403: Many of the plots shown in figure S15 show quite an asymmetric distribution, so fitting a Gaussian function does not seem to work well (e.g. panels a, c, d). I would instead suggest fitting an asymmetric function to better capture the difference in the positive and negative values.
- L449-450: In addition to spreading out a plume to obtain lower VCDs, injecting the plume at higher altitude will have an impact on its location due to wind shear. Is it feasible for winds at lower altitudes to have transported the pre-eruptive plume that you attribute to the pixels in the swath edge, or is this only possible at higher altitudes? If so, then this would suggest that this is not pre-eruptive emissions.
- L452: When you say “larger wind fields”, do you mean higher velocity?
- L453-458: Is it possible that using the 7 km VCD product is artificially pushing altitudes higher due to this effect? The retrieved VCD in the 7 km product will be lower than for the plume measured if the reported injection altitudes are similar to the plume altitudes at the time of measurement, so a higher altitude/faster wind may have been selected by the model to account for this.
- L460: “greatly aids in accurately capturing emission timing” – on this point, if the altitude is incorrect then the emission timing will also be off due to the difference in wind speed. Have you considered the magnitude of this error?
- L462-464: It is worth noting that the plume height retrievals only work for strong plumes, so may not help for constraining weaker emissions.
- L470-471: No mention is made of the trough in SO2 emissions at the peak in lava flow index.
- L485-486: Can you expand on this further? No comment on the physical mechanism of transporting this SO2 from depth to the surface, nor if it came from the lava lake or the lava flow region (or somewhere else?). I would also note that the dyke intrusion lasted ~3 days, so why is the SO2 emission such a sharp peak? Are there any ground-based observations to support this? Shreve et al. (2019) also note that the total magnitude of SO2 emissions matches that expected from the volume of the lava flow, so that contributions from the dyke were not significant. I agree that the timing of this peak in SO2 emission after the lava flow is interesting and may provide insights into the magmatic processes, but this needs further explanation.
- Figure 1: in panel bii the #1 arrow is pointing to an empty region near the plume, should this be pointing to the pre-eruptive plume in the swath edge?
Technical comments
---------------------------------------------------------------------------------------
- L82: Consider replacing “emissions” with “emission”
- L212: You refer to points in figures 1 and 2 as “T2” and “O2”, for example, but in the figures these are just numbers (e.g. #1, #2). It would be clearer if these were consistent (i.e. use #T1 in figure 1, #O1 in figure 2).
- L299: “SO2” is missing a subscript here.
- Many figures use non-perceptually uniform colourmaps, which can be misleading or difficult to interpret, especially for colourblind people or when printed in black and white (Crameri et al., 2020). I would suggest replacing the colourmaps used throughout with colourblind-friendly versions.
- Satellite figures: The figures of satellite data all have elliptical pixels for the data. Why is this used instead of a continuous grid of rectangular pixels? This would avoid the overlapping in the OMPS data.
- Figures S8-11: How have the sub-frames been ordered? They do not appear to be in terms of number, time, altitude or strength of emission and interpreting these plots was difficult. Would it be possible to reorder these, or have I missed the ordering?
References
---------------------------------------------------------------------------------------
Crameri, F., Shephard, G.E., Heron, P.J., 2020. The misuse of colour in science communication. Nat. Commun. 11, 5444. https://doi.org/10.1038/s41467-020-19160-7
-
RC2: 'Comment on egusphere-2023-2545', Anonymous Referee #2, 01 Feb 2024
Overview
The authors compare using either TROPOMI or OMPS total column SO2 retrievals in an inverse modelling approach (CHIMERE) to infer the SO2 emissions flux and plume height data of the Ambrym volcano during the eruption in December 2018. They compare their results with SO2 flux proxy data derived from Himawari-8/AHI. The find a better performance when the TROPOMI data, which have a higher spatial resolution.
General comments
The paper seems to lack a scientifically sound method to quantify and compare the inversion result using either TROPOMI or OMPS. It remains often unclear, why they can conclude the strong superiority of TROPOMI data. The recommended approach would be to use the derived emissions parameters (flux and injection height) in a forward simulation and to develop a metric to quantify to what extend the simulates plumes agree with observations. It is not satisfactory to only juxtapose maps of plumes with HIMAWARI proxy SO2 data.
While the authors come to the plausible conclusion that TROPOMI data are much better, it remains unclear how this conclusion is derived especially considering this lack of rigor mentioned above. It should also be emphasized more that they study only one specific episode using their specific modeling framework. I recommend using a more neutral language when describing the results.
The inversion approach should be explained in more detail from a practical perspective. In particular the robustness of the derived plume heights remains unclear. It is also not clear, what prior information was used and how the ensemble of tracer plumes injected at different height is part of the framework.
The paper should provide in a numerical way the derived time series of flux and injection height to allow the scientific community to use the data. If applicable, the result should be compared to the results of other authors.
The abstract needs to be strongly revised to report in more detail the findings and to reduce the number of the more general or introductory statements.
The paper should get a more precise title such as “Inversion of SO2 emissions from Ambrym 2018 using TROPOMI or OMPS retrievals.”
The Figure captions are too long because they interpret the shown results. This should be done in the body text. On the other hand, to little explanation of the meaning of the different lines of the time series is provided. I recommend showing the emission parameter time series as separate plots, not mixed with the maps.
Specific Comments
L10 Please mention, how this was achieved.
L13 Please provide some actual numbers of the emissions here.
L 129 It needs to be properly discussed that the subject comparison with HIMAWARI is the main reference for your study.
152 A model top of 200 hPa seems too low to simulate the fate of many volcanic SO2 plumes. Are you sure the plume is always located below that height ?
L 154 Please explain in more detail how the tracer ensemble is used in the inversion framework.
L163 Does the state vector include the emissions, or is the state only the atmospheric concentration.
L 170 Please introduce the emissions in your framework here.
L 160 Please add more information about the time stepping and assimilation window length in the section. What method do you use – a ensemble (KF) or variational approach.
L245 This is the first time you mention an indication that TROPOMI agrees better with Himawari than OMPS.
L 299 This sensitivity study should be given more attention, perhaps its own section. Was it done only for the outgassing or also for the eruptive events.
L311 Discuss the relation to the model resolution.
L322 How do you know that TROPOMI does not overestimate this value ?
L333 Is this bimodal injection height something that has been observed before - or is it an artefact of the assimilation, which does not get enough information about the vertical distribution of the plume.
L336 Please provide more details. HYSPLIT has not been introduced before to be part of the study. Did you run the model yourself?
L343 If HIMAWARI is you reference, why is the better agreement between OMPS and HIMAWARI not a good result for OMPS?
L375 This is a very controversial statement that needs more backing up with evidence. Given the high variability of the plumes, using the observations from the previous day to substitute cloud-contaminated observations seems risky.
L416 Why initialized? The emissions should be injected during the model run.
L487 This conclusion is not clear in this context.
L495 Provide more reasons.
L496 It is not clear why a high spatial resolution is helpful for capturing short - lived (short?) emissions. The overpass intervals is the same for TROPOMI and OMPS.
L501 That is a very general statement, please provide evidence from the study.
L 503 Again, I am not convinced that has been proven universally in that paper. Please explain if you mean the approach to use yesterday’s observations to gap fill.
L 508 Compare that to values from the literature.
Figure 1,2 Consider putting the time series in a separate picture. Please explain in more detail all 4 lines in the time series plot.
Citation: https://doi.org/10.5194/egusphere-2023-2545-RC2 -
AC1: 'Comment on egusphere-2023-2545', Abhinna Behera, 20 Apr 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-2545/egusphere-2023-2545-AC1-supplement.pdf
-
AC1: 'Comment on egusphere-2023-2545', Abhinna Behera, 20 Apr 2024
-
AC1: 'Comment on egusphere-2023-2545', Abhinna Behera, 20 Apr 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-2545/egusphere-2023-2545-AC1-supplement.pdf
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 386 | 108 | 28 | 522 | 21 | 21 | 20 |
- HTML: 386
- PDF: 108
- XML: 28
- Total: 522
- Supplement: 21
- BibTeX: 21
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1