the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 13 Mar 2025
Quantifying CO emissions from boreal wildfires by assimilating TROPOMI and TCCON observations
Abstract. We perform an inverse modelling analysis to quantify biomass burning emissions of carbon monoxide (CO) from the extreme wildfires in Canada between May and September 2023. Using the GEOS-Chem model, we assimilated observations from the Tropospheric Monitoring Instrument (TROPOMI) separately and then jointly with Total Carbon Column Observing Network (TCCON) measurements. We also evaluated prior emissions from the Quick Fire Emissions Dataset (QFED), Blended Global Biomass Burning Emissions Product eXtended (GBBEPx), Global Fire Assimilation System (GFAS), and Canadian Forest Fire Emissions Prediction System (CFFEPS). The assimilation of TROPOMI-only measurements estimated posterior North America emissions for QFED, GBBEPx, GFAS, and CFFEPS of 110.4±20, 112.8±20, 127.2±17, and 125.6±18 Tg CO compared to prior estimates of 37.1, 42.7, 91.0, and 90.2 Tg CO, respectively. The joint assimilation of TROPOMI+TCCON reduced the uncertainty on the North American emission estimates by up to about 30 %, while showing only a modest impact (< 5 %) on the magnitude of the inferred emissions. An evaluation against independent measurements reveals that adding TCCON data increases the correlations and slightly lowers the biases and standard deviations. Additionally, including an experimental TCCON product at East Trout Lake with higher surface sensitivity, we find better agreement of assimilation results with nearby in situ tall tower and aircraft measurements. This highlights the potential importance of vertical sensitivity in these experimental data for constraining local surface emissions. Our results demonstrate the complementarity of the greater temporal coverage provided by TCCON with the spatial coverage of TROPOMI when these data are jointly assimilated.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(5213 KB)
-
Supplement
(446 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(5213 KB) - Metadata XML
-
Supplement
(446 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
- RC1: 'Comment on egusphere-2025-858', Anonymous Referee #1, 18 Apr 2025
-
RC2: 'Comment on egusphere-2025-858', Anonymous Referee #2, 12 May 2025
The paper by Voshtani et al. discusses the estimate of wildfire CO emissions on a global scale by assimilating both TROPOMI and TCCON observations. The paper is very well written, with high-quality graphics and a very comprehensive set of references. The comparison of the results for 3 (4) independent CO fire inventories provides insightful new results. The sparse TCCON network, in combination with TROPOMI XCO, is convincingly shown to provide a large impact on the emission estimates and a-posteriori emission uncertainty. As a consequence I am in support of publishing these results. I only have relatively few comments, ranked as minor, for which I would ask the authors to include their responses in the manuscript.
Minor comments:
l 82: Is "bottom-up" the right term? FRP is retrieved from satellite observations, and not estimated from "activity data". For instance, GFAS stands for "Global Fire Assimilation System".
l 100: A major benefit of TROPOMI is it's high sensitivity for CO close to the surface, in contrast to e.g. MOPITT or IASI. The total column measured has a strong relation to the emission, which can be estimated using basically only wind information. Please emphasise this point.
l 141: Why is XCO used as acronym of the "Total column abundances of carbon monoxide" ? Where does X come from (why not TCO or TCCO)?
l 149: "quality flag equal to and greater than 0.7 to ensure high-quality data obtained under cloud-free or low cloud conditions." Fires produce smoke, which may look similar to clouds. I was wondering if part of the measurements close to the fires may not be included in the quality flag > 0.7 dataset. Does this filtering impact the inversions?
l 154: The GEOS-Chem model resolution used in this study (2 x 2.5) is relatively coarse. Does this resolution influence the BB emission estimate?
l 157: super-observations "we average the observations, weighted based on their reported retrieval errors".
l 158: "The super-observation errors in each grid are also obtained by averaging the reported retrieval errors combined with the error correlation between measurements following Pendergrass et al. (2023)"
The way the construction of the super-observations is described here seems to contradict Pendergrass 2023, section 3.3. If I understand correctly observations are averaged, and not weighted, in this paper. Please provide the details on how the weights are constructed (the correlation factor is mentioned later). Is the weight inversely proportional to the error^2? This is important to appreciate the overfitting and gamma factor discussed later.Fig. 1, panel a: What is the time period for collecting the 1.7 million TROPOMI super-observations? Is it also May-September 2023?
p12, eq 1: What does the gamma factor represent? (p13 mentions the gamma used) Why is it needed?
l 314: Is this an OSSE? For me the term OSSE refers an assessment of the impact of a future (satellite) observational dataset on an existing analysis system. It is typically using two independent models/systems where one is generating the Nature run and synthetic observations and the other performs the analysis.
l 338: "We set a localization radius of 500 km" Given the model resolution of 2 x 2.5 degree, this is only two grid cells. Also, for CO with a long lifetime, the plumes (emission and concentration location spatial distances) remain visible over thousands of km. Could you please discuss this in some more detail. Were tests performed with longer correlation lengths?
l 340: The correlation factor of 0.28 was reported for CH4, while the present study is for CO. Please comment.
l 387: "the assimilation did not ingest TROPOMI observations poleward of 60°N". Since there is a clear focus on Boreal fires, and since this choice impacts the results, I was wondering if the sharp cut-off at 60 degree could be justified a bit more. It is mentioned that snow-covered land should be avoided. But most of the fires will occur after the snow has melted. Is snow-cover provided in the TROPOMI product? If so, could this be used to refine the filtering and provide an option to use observations north of 60 degree?
l 402: "higher accuracy provided by TCCON". The surface column observations are point-like while the model represents 2x2.5 degree box averages. There may be a considerable representativity uncertainty, bigger than the measurement uncertainty, especially for fire plumes passing over the station. Was such a representativity term considered/included, and is there a way to estimate it? Please add a discussion.
Figure 5: I was wondering if the extra reduction due to TCCON occurs close to the TCCON stations? I would find it interesting to have extra plots of the difference between the right and left panels, with the TCCON locations also indicated.
l 521: "This likely implies that the difference of the perturbed and unperturbed forecast of the state vector, which approximates their covariances, correlates better with the actual emissions, such that the greater variations in GFAS emissions result in higher DOFS." I do not understand this argument. What is the relation between the presence of variations and DOFS?
Fig.8: In the text on p 24 the R^2 is reported as "correlation". This is normally called the "Coefficient of determination", or explained variance. Does figure 8 report the correlation, or the square of the correlation?
Fig.8: Are there stations shown in this plot which are equipped with both NDACC and TCCON instruments? If so, please indicate those, or note this is not the case.
l 568: "sites close to or downwind of TCCON sites" It would be useful to mention the distance between the sites. Is it more or less than a grid box?
l 764: "there is stronger agreement at the NDACC and in situ sites that are located in close proximity to the
TCCON measurements used in the inversion". This is not really a surprise.l 853: "Our analysis yields the following optimal values: gamma_tropomi = 0.2 and gamma_tccon = 5, delta = 0.08, r = 500 km, with a minimum of three months for spin-up, one month for burn-in, and a minimum of 24 ensemble members - each chosen to minimize OmF statistics." This is quite a number of parameters to obtain from one set of OmFs, which are also influencing each other. Please provide a plot to document this analysis. In particular it is interesting to learn how the two gamma values are obtained.
Citation: https://doi.org/10.5194/egusphere-2025-858-RC2 -
RC3: 'Comment on egusphere-2025-858', Anonymous Referee #3, 16 May 2025
The manuscript titled “Quantifying CO emissions from boreal wildfires by assimilating TROPOMI and TCCON observations” by Voshtani et al. with reference egusphere-2025-858 is a highly valuable scientific contribution in atmospheric air quality modeling and emission inversion for wildfires. Overall, it is very well written and provides a complete investigation of the benefit of large-scale CO emission inversion using TCCON total column retrievals. Overall, I suggest accepting the manuscript with the following minor corrections in the attached document.
- AC1: 'Comment on egusphere-2025-858', Sina Voshtani, 17 Jul 2025
- AC2: 'Comment on egusphere-2025-858', Sina Voshtani, 29 Aug 2025
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-858', Anonymous Referee #1, 18 Apr 2025
This study optimized the CO emissions between May and September 2023 with the CHEEREIO toolkit. Various inversion experiments were conducted with global and regional priori emission data, TropOMI and TCCON observations and different vertical profiles of TCCON data. This study is well designed and the manuscript is in great organization. The findings could benefit future emission inversions. My comments are listed below:
Fig 1(a) the TCCON observation is at hourly scale and the TROPOMI observation is at daily scale, thus please use separate color bar for plotting number of observations.
Fig 3 The posteriors from different priori showed observable differences in temporal variations. Please explain and evaluate these differences.
There are figures showing the differences between priori and posteriors, and between observations and priori/posteriors. There lacks posteriors emission maps.
Citation: https://doi.org/10.5194/egusphere-2025-858-RC1 -
RC2: 'Comment on egusphere-2025-858', Anonymous Referee #2, 12 May 2025
The paper by Voshtani et al. discusses the estimate of wildfire CO emissions on a global scale by assimilating both TROPOMI and TCCON observations. The paper is very well written, with high-quality graphics and a very comprehensive set of references. The comparison of the results for 3 (4) independent CO fire inventories provides insightful new results. The sparse TCCON network, in combination with TROPOMI XCO, is convincingly shown to provide a large impact on the emission estimates and a-posteriori emission uncertainty. As a consequence I am in support of publishing these results. I only have relatively few comments, ranked as minor, for which I would ask the authors to include their responses in the manuscript.
Minor comments:
l 82: Is "bottom-up" the right term? FRP is retrieved from satellite observations, and not estimated from "activity data". For instance, GFAS stands for "Global Fire Assimilation System".
l 100: A major benefit of TROPOMI is it's high sensitivity for CO close to the surface, in contrast to e.g. MOPITT or IASI. The total column measured has a strong relation to the emission, which can be estimated using basically only wind information. Please emphasise this point.
l 141: Why is XCO used as acronym of the "Total column abundances of carbon monoxide" ? Where does X come from (why not TCO or TCCO)?
l 149: "quality flag equal to and greater than 0.7 to ensure high-quality data obtained under cloud-free or low cloud conditions." Fires produce smoke, which may look similar to clouds. I was wondering if part of the measurements close to the fires may not be included in the quality flag > 0.7 dataset. Does this filtering impact the inversions?
l 154: The GEOS-Chem model resolution used in this study (2 x 2.5) is relatively coarse. Does this resolution influence the BB emission estimate?
l 157: super-observations "we average the observations, weighted based on their reported retrieval errors".
l 158: "The super-observation errors in each grid are also obtained by averaging the reported retrieval errors combined with the error correlation between measurements following Pendergrass et al. (2023)"
The way the construction of the super-observations is described here seems to contradict Pendergrass 2023, section 3.3. If I understand correctly observations are averaged, and not weighted, in this paper. Please provide the details on how the weights are constructed (the correlation factor is mentioned later). Is the weight inversely proportional to the error^2? This is important to appreciate the overfitting and gamma factor discussed later.Fig. 1, panel a: What is the time period for collecting the 1.7 million TROPOMI super-observations? Is it also May-September 2023?
p12, eq 1: What does the gamma factor represent? (p13 mentions the gamma used) Why is it needed?
l 314: Is this an OSSE? For me the term OSSE refers an assessment of the impact of a future (satellite) observational dataset on an existing analysis system. It is typically using two independent models/systems where one is generating the Nature run and synthetic observations and the other performs the analysis.
l 338: "We set a localization radius of 500 km" Given the model resolution of 2 x 2.5 degree, this is only two grid cells. Also, for CO with a long lifetime, the plumes (emission and concentration location spatial distances) remain visible over thousands of km. Could you please discuss this in some more detail. Were tests performed with longer correlation lengths?
l 340: The correlation factor of 0.28 was reported for CH4, while the present study is for CO. Please comment.
l 387: "the assimilation did not ingest TROPOMI observations poleward of 60°N". Since there is a clear focus on Boreal fires, and since this choice impacts the results, I was wondering if the sharp cut-off at 60 degree could be justified a bit more. It is mentioned that snow-covered land should be avoided. But most of the fires will occur after the snow has melted. Is snow-cover provided in the TROPOMI product? If so, could this be used to refine the filtering and provide an option to use observations north of 60 degree?
l 402: "higher accuracy provided by TCCON". The surface column observations are point-like while the model represents 2x2.5 degree box averages. There may be a considerable representativity uncertainty, bigger than the measurement uncertainty, especially for fire plumes passing over the station. Was such a representativity term considered/included, and is there a way to estimate it? Please add a discussion.
Figure 5: I was wondering if the extra reduction due to TCCON occurs close to the TCCON stations? I would find it interesting to have extra plots of the difference between the right and left panels, with the TCCON locations also indicated.
l 521: "This likely implies that the difference of the perturbed and unperturbed forecast of the state vector, which approximates their covariances, correlates better with the actual emissions, such that the greater variations in GFAS emissions result in higher DOFS." I do not understand this argument. What is the relation between the presence of variations and DOFS?
Fig.8: In the text on p 24 the R^2 is reported as "correlation". This is normally called the "Coefficient of determination", or explained variance. Does figure 8 report the correlation, or the square of the correlation?
Fig.8: Are there stations shown in this plot which are equipped with both NDACC and TCCON instruments? If so, please indicate those, or note this is not the case.
l 568: "sites close to or downwind of TCCON sites" It would be useful to mention the distance between the sites. Is it more or less than a grid box?
l 764: "there is stronger agreement at the NDACC and in situ sites that are located in close proximity to the
TCCON measurements used in the inversion". This is not really a surprise.l 853: "Our analysis yields the following optimal values: gamma_tropomi = 0.2 and gamma_tccon = 5, delta = 0.08, r = 500 km, with a minimum of three months for spin-up, one month for burn-in, and a minimum of 24 ensemble members - each chosen to minimize OmF statistics." This is quite a number of parameters to obtain from one set of OmFs, which are also influencing each other. Please provide a plot to document this analysis. In particular it is interesting to learn how the two gamma values are obtained.
Citation: https://doi.org/10.5194/egusphere-2025-858-RC2 -
RC3: 'Comment on egusphere-2025-858', Anonymous Referee #3, 16 May 2025
The manuscript titled “Quantifying CO emissions from boreal wildfires by assimilating TROPOMI and TCCON observations” by Voshtani et al. with reference egusphere-2025-858 is a highly valuable scientific contribution in atmospheric air quality modeling and emission inversion for wildfires. Overall, it is very well written and provides a complete investigation of the benefit of large-scale CO emission inversion using TCCON total column retrievals. Overall, I suggest accepting the manuscript with the following minor corrections in the attached document.
- AC1: 'Comment on egusphere-2025-858', Sina Voshtani, 17 Jul 2025
- AC2: 'Comment on egusphere-2025-858', Sina Voshtani, 29 Aug 2025
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 1,122 | 70 | 32 | 1,224 | 44 | 22 | 41 |
- HTML: 1,122
- PDF: 70
- XML: 32
- Total: 1,224
- Supplement: 44
- BibTeX: 22
- EndNote: 41
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Sina Voshtani
Dylan B. A. Jones
Debra Wunch
Drew C. Pendergrass
Paul O. Wennberg
David F. Pollard
Isamu Morino
Hirofumi Ohyama
Nicholas M. Deutscher
Frank Hase
Ralf Sussmann
Damien Weidmann
Rigel Kivi
Omaira García
Jack Chen
Kerry Anderson
Robin Stevens
Shobha Kondragunta
Aihua Zhu
Douglas Worthy
Senen Racki
Kathryn McKain
Maria V. Makarova
Nicholas Jones
Emmanuel Mahieu
Andrea Cadena-Caicedo
Paolo Cristofanelli
Casper Labuschagne
Elena Kozlova
Thomas Seitz
Martin Steinbacher
Reza Mahdi
Isao Murata
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(5213 KB) - Metadata XML
-
Supplement
(446 KB) - BibTeX
- EndNote
- Final revised paper
This study optimized the CO emissions between May and September 2023 with the CHEEREIO toolkit. Various inversion experiments were conducted with global and regional priori emission data, TropOMI and TCCON observations and different vertical profiles of TCCON data. This study is well designed and the manuscript is in great organization. The findings could benefit future emission inversions. My comments are listed below:
Fig 1(a) the TCCON observation is at hourly scale and the TROPOMI observation is at daily scale, thus please use separate color bar for plotting number of observations.
Fig 3 The posteriors from different priori showed observable differences in temporal variations. Please explain and evaluate these differences.
There are figures showing the differences between priori and posteriors, and between observations and priori/posteriors. There lacks posteriors emission maps.