the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 May 2023
Towards an improved understanding of wildfire CO emissions: a satellite remote-sensing perspective
Abstract. Emissions from wildfires are a significant source of air pollution, which can adversely impact air quality and ecosystems thousands of kilometers downwind. These emissions can be estimated by a bottom-up approach, using inputs such fuel type, burned area, and standardized emission factors. Emissions are also commonly derived with a top-down approach, using satellite observed fire radiative power (FRP) as proxy for fuel consumption. More recently, wildfire emissions have been demonstrated to be estimated directly from satellite observations, including carbon monoxide (CO). Here, we explore the potential of satellite-derived CO emission rates from wildfires and provide new insights into the understanding of satellite-derived fire CO emissions globally, with respect to differences in regions and vegetation type. Specifically, we use the TROPOMI (Tropospheric Monitoring Instrument) high spatial-resolution satellite datasets to create a global inventory database of burning emissions CO emissions between 2019 and 2021. Our retrieval methodology includes an analysis of conditions under which emission estimates may be inaccurate and filters these accordingly. Additionally, we determine biome specific emission coefficients (emissions relative to FRP) and show how combining the satellite derived CO emissions with satellite observed FRP from the Moderate Resolution Imaging Spectrometer (MODIS) establishes an annual CO emission budget from wildfires. The resulting emissions totals are compared to other top-down and bottom-up emission inventories over the past two decades. In general, the satellite-derived emissions inventory values and bottom-up emissions inventories have similar CO emissions totals across different global regions, though the discrepancies may be large for some regions (Southern Hemisphere South America, Southern Hemisphere Africa, Southeast Asia) and for some bottom-up inventories (e.g. FINN2.5, where CO emissions are a factor of 2 to 5 higher than other inventories). Overall, these estimates can help to validate emission inventories and predictive air quality models, and help to identify limitations present in existing bottom-up emissions inventory estimates.
-
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
(2077 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2077 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-649', Anonymous Referee #1, 01 Jun 2023
The manuscript entitled "Towards an improved understanding of wildfires CO emissions: a satellite remote-sensing perspective” by Griffin et al aims to derive CO emission coefficients by correlating TROPOMI CO with MODIS FRP and produce a new global CO emission inventory using the emission coefficients and GFAS FRP. The authors first estimate CO emissions directly for forest fires in an hour near the overpass time of S5P and evaluate temporally redistributed fire emissions from a Canada emission model – CFFEPS. By correlating TROPOMI CO flux to MODIS FRP, biome-specific CO emission coefficients are derived over different numbers of fire events. The annual budget of CO emissions is estimated finally by applying the derived CO emission coefficients to GFAS FRP, and further compared with several other inventories.
Overall, the topic of this study fits the scope of ACP well and it has very meaningful goals. As S5P TROPOMI provides CO observations of global fires at the highest spatial resolution yet, it provides a good opportunity to explore CO emission coefficient for global fires, which potentially improves the estimation of biomass burning emissions. Unfortunately, I think the study fails to achieve these goals due to the seriously flawed method for deriving emission coefficients and the failure of assessing the accuracy of CO budgets. First, I acknowledge that using TROPOMI CO observations to directly estimate CO emissions from fires during a specific short period of time is sound, and direct CO estimates are valuable independent emission data for evaluating other emission estimates. Yet, both the idea and CO estimation method are not new as TROPOMI CO has been successfully applied to assess fire emissions in several recently published papers. Now, let me lay out reasons why I think the method to derive emission coefficients is flawed.
Theoretically, emission coefficient (g/J), which represents the mass of emissions per Joule radiative energy emitted from fire, can be derived if continuous, accurate rate of emission and FRP (or emission mass and FRE) are known. One MODIS instrument provides daily up to two observations of fires at the same location at low-mid latitudes. If only daytime Aqua MODIS FRP is used, it only provides one observation as with TROPOMI CO observation. In the method of deriving CO emission coefficient by correlating TROPOMI CO to Aqua MODIS FRP, the underline assumptions are that emission flux based TROPOMI CO and Aqua MODIS FRP are able to represent mean CO flux and mean FRP for a given fire sample during a specific period of time (±30min or several hours?). I would not think these simplified assumptions hold in most cases. For MODIS FRP, it has a strong dependency on MODIS scan angle. In other words, FRP value can be largely different if the instrument observes the same fire at nadir and in large scan angles. In that case, emission coefficient likely changes largely when the scan angle varies. Moreover, the observation gaps between S5P and Aqua can be up to about 60 minutes although they are thought to be in similar orbits. I think this explains the very scattering distribution of samples in Fig 6, not to mention the very pool correlations in evergreen needle leaf dominated by forest wildfires. It looks like the authors are not aware of the characteristics of MODIS FRP except for listing the incapability of detecting very small fires and cloud/smoke contamination. A scientifically sound way would be deriving coefficients based on TROPOMI CO and FRP from the new-generation geostationary satellites, which has been done in several published papers that are never mentioned as background in Introduction nor discussed in Discussion. The accuracy of CO flux also relies on wind directions and speed, which are also a concern.
The accuracy of the new CO inventories depends on the accuracy of the derived CO emission coefficients and that of GFAS FRE. FRE calculation requires continuous FRP observations. Diurnal FRP varies very largely from day to day even for the same fire, especially large forest wildfires, which have been reported in several JGR and RSE papers. I would not expect reliable FRE to be calculated from daily mean GFAS FRP that is averaged using merely up to four daily MODIS FRP observations, although GFAS emissions are used in ECMWF forecast models. Furthermore, simply comparing it with a few other inventories doesn’t tell any information about the accuracy of the new CO inventory. There are more than 10 BB emission inventories for different purposes, and they can differ from each other by a factor of up to 30 in individual fire events although the difference in their annual budget could be much smaller. I don’t see any meaningful contributions of a new inventory to the BB community without knowing its accuracy.
To sum up, I don’t see sound contributions from this study, thus I would not recommend it for publication in ACP.Citation: https://doi.org/10.5194/egusphere-2023-649-RC1 -
AC2: 'Reply on RC1', Debora Griffin, 29 Sep 2023
We would like to thank Reviewer 1 for their review. In response to the feedback from both reviewers, we have implemented significant revisions to the manuscript. I believe there might be some misunderstanding by the reviewer and much of the criticisms raised by the first reviewer do not provide valid grounds for rejection. The specific points mentioned in the review are addressed, see attached pdf for details.
-
AC2: 'Reply on RC1', Debora Griffin, 29 Sep 2023
-
RC2: 'Comment on egusphere-2023-649', Anonymous Referee #2, 06 Jul 2023
This manuscript focuses on carbon monoxide (CO) emissions from wildfires; its goal is to produce a ~20 year global budget based on emission coefficients derived from TROPOMI and MODIS data:
- First, TROPOMI-derived emissions are calculated for the 2019-2021 period with an automated version of the flux method. To verify the robustness and uncertainties of the automated method, emissions are derived from synthetic CO column values generated with the GEM-MACH model and CFFEPS for the May-September 2019 period over North America. From this test, several filters (dealing with background CO, wind, plume geometry, etc.) are defined.
- Then, global TROPOMI-derived emissions are calculated for the 2019-2021 period, the filters defined in the previous step are applied, and CO emission coefficient values (EC) are calculated separately for 15 biomes: EC=E/FRP=emissions/fire radiative power. FRP values are from MODIS Aqua. Somehow
the GFAS FRP dataset (which results from assimilating both MODIS Terra and Aqua FRP values in GFAS) is used to provide “a guidance on total daily FRP that can then be combined with the derived ratio between TROPOMI CO emissions and MODIS FRP”; meaning unclear from the manuscript.
- Following, a global budget of CO emissions from fires is calculated for the 2003-2021 period based on TROPOMI-MODIS Aqua EC values and GFAS FRP values. Results are analyzed by region and by biome, and compared with respect to several emission inventories.
The manuscript should be improved to avoid repetitions, typos, correct wording and sentence integrity, provide explanations for acronyms/abbreviations when first used, etc. The text is hard to follow at times because of these issues. Some of the Appendices are never mentioned in the text. Expressions such as “many of the retrieved emissions”, “many outliers”, “aligns very well”, “the model agrees pretty well” should be avoided. Quantitative statements should be used instead.
Some figures would benefit from more consistent axis ranges (e.g., Fig. 1 b and c), axis labels with both axis title and units (Fig. 5). Panels labels should be properly referred to in text (e.g., Fig. 5). For clarity, please add a map showing all the regions discussed in the manuscript (CEAM, NHSA, EURO, …). Consider including a figure illustrating the areal extent of biomes.
The manuscript should justify why the TROPOMI averaging kernels are not applied to retrievals with qa_value <1. From the TROPOMI CO readme document: “We recommend using only data with a qa_value = 1 in case the averaging kernel is not applied. Data with a qa_value = 0.7 are of similar quality provided the averaging kernel is used to account for the vertical retrieval sensitivity in the presence of mid-level clouds. Quality assurance values of qa_value = 0.4 represent experimental data to be used with caution.” Table 3 in that same document provides additional information regarding qa_values, cloud heights, and tau_aer values.
While the masking effect of smoke and clouds on FRP observations is discussed several times, the effect of smoke and clouds on TROPOMI retrievals is never mentioned. This is a very important issue, given the focus on TROPOMI observations acquired over active fires, with very smoky and (potentially) very cloudy conditions. Please discuss.
In lines 319-321, to explain discrepancies between emissions from TROPOMI and GFFEPS, it is suggested that GFFEPS values may be low because of missed fires due to thick smoke. Could TROPOMI values be high due to smoke? clouds? other factors?
Figure 6 shows that the correlation between CO and FRP may not be very robust, despite R being 0.70. How would the slope (i.e., EC) change if the most extreme outliers were removed one by one?
Line 379: “The emission coefficients vary between 120 and 39 g/MJ” Please clarify that EC=120 value corresponds to (11): shrub cover, closed-open, evergreen; EC=39 to (12): shrub cover, closed-open, deciduous. The most extreme EC values correspond to rather similar biomes, please discuss.
Lines 379-380: “where the largest CO emissions relative to FRP are from broadleaved evergreen tree cover (1) and the lowest are from cultivated managed areas (16)”. How could readers see CO emissions relative to FRP? Similar issue in lines 474-475.
Line 401: “Most CO emissions are from evergreen forests (biome type 1), which also has the largest EC for CO”. Table 2 shows that biome 11 has the largest EC, please correct.
Please justify why is biome 15 (“Regularly flooded shrub and/or herbaceous cover”) included in the analysis. Biomes 7 and 8 are excluded because “fires were not observed, namely: regularly flooded tree cover (7 and 8)”
Biome # 15 (regularly flooded shrub and/or herbaceous cover; should this biome be included in the analysis, given that it is regularly flooded?) has the second largest EC (105 g/MJ). Please explain.
The manuscript assumes that EC values “do not change drastically over the years”. However, tables C1, C2, and C3 in Appendix C (never mentioned in the manuscript; please correct) show otherwise. EC values change between -50% and +150%, depending on the period analyzed. EC values change strongly among biomes; changes do not seem to follow recognizable patterns. How would the global budget change if a different set of EC values was used? Please quantify.
What is the areal extent of each biome? That information could help readers understand what biomes (and what biome’s EC) may have a stronger effect on the global budget.
Figure 9 shows emissions decreasing with time. An evaluation of the temporal trend in the global budget with respect to actual measurements (not inventories) is missing in the manuscript. Have wildfire emissions really decreased in the last two decades? While measurements of global emissions for all biomes may be unavailable, a literature search may provide useful information for specific regions/biomes.
Lines 446-447: “certain regions see increased emissions (e.g. TENA, AUST)” (text refers to Fig. 9). Hard to know for sure, but it looks like neither TENA nor AUST show increasing emissions. Either provide a better figure to illustrate the statement or remove statement. Similar issue in line 498.
Lines 476-477: “for forests we determined ECs between 64 and 120 g/MJ” Please note that EC=120 is not for a forest but for biome 11: Shrub Cover, closed-open, evergreen.
The manuscript refers to the need for multiple observations in a day (from geostationary instruments) in order to better understand fire evolution; TEMPO is mentioned but GEMS (which has been operating for a couple of years now) is not. Please discuss both.
Citation: https://doi.org/10.5194/egusphere-2023-649-RC2 -
AC1: 'Reply on RC2', Debora Griffin, 29 Sep 2023
We extend our gratitude to Reviewer 2 for their detailed review. In response to the feedback from both reviewers, we have implemented significant revisions to the manuscript. We offer detailed insights into these changes that can be found in the pdf attached.
-
AC1: 'Reply on RC2', Debora Griffin, 29 Sep 2023
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-649', Anonymous Referee #1, 01 Jun 2023
The manuscript entitled "Towards an improved understanding of wildfires CO emissions: a satellite remote-sensing perspective” by Griffin et al aims to derive CO emission coefficients by correlating TROPOMI CO with MODIS FRP and produce a new global CO emission inventory using the emission coefficients and GFAS FRP. The authors first estimate CO emissions directly for forest fires in an hour near the overpass time of S5P and evaluate temporally redistributed fire emissions from a Canada emission model – CFFEPS. By correlating TROPOMI CO flux to MODIS FRP, biome-specific CO emission coefficients are derived over different numbers of fire events. The annual budget of CO emissions is estimated finally by applying the derived CO emission coefficients to GFAS FRP, and further compared with several other inventories.
Overall, the topic of this study fits the scope of ACP well and it has very meaningful goals. As S5P TROPOMI provides CO observations of global fires at the highest spatial resolution yet, it provides a good opportunity to explore CO emission coefficient for global fires, which potentially improves the estimation of biomass burning emissions. Unfortunately, I think the study fails to achieve these goals due to the seriously flawed method for deriving emission coefficients and the failure of assessing the accuracy of CO budgets. First, I acknowledge that using TROPOMI CO observations to directly estimate CO emissions from fires during a specific short period of time is sound, and direct CO estimates are valuable independent emission data for evaluating other emission estimates. Yet, both the idea and CO estimation method are not new as TROPOMI CO has been successfully applied to assess fire emissions in several recently published papers. Now, let me lay out reasons why I think the method to derive emission coefficients is flawed.
Theoretically, emission coefficient (g/J), which represents the mass of emissions per Joule radiative energy emitted from fire, can be derived if continuous, accurate rate of emission and FRP (or emission mass and FRE) are known. One MODIS instrument provides daily up to two observations of fires at the same location at low-mid latitudes. If only daytime Aqua MODIS FRP is used, it only provides one observation as with TROPOMI CO observation. In the method of deriving CO emission coefficient by correlating TROPOMI CO to Aqua MODIS FRP, the underline assumptions are that emission flux based TROPOMI CO and Aqua MODIS FRP are able to represent mean CO flux and mean FRP for a given fire sample during a specific period of time (±30min or several hours?). I would not think these simplified assumptions hold in most cases. For MODIS FRP, it has a strong dependency on MODIS scan angle. In other words, FRP value can be largely different if the instrument observes the same fire at nadir and in large scan angles. In that case, emission coefficient likely changes largely when the scan angle varies. Moreover, the observation gaps between S5P and Aqua can be up to about 60 minutes although they are thought to be in similar orbits. I think this explains the very scattering distribution of samples in Fig 6, not to mention the very pool correlations in evergreen needle leaf dominated by forest wildfires. It looks like the authors are not aware of the characteristics of MODIS FRP except for listing the incapability of detecting very small fires and cloud/smoke contamination. A scientifically sound way would be deriving coefficients based on TROPOMI CO and FRP from the new-generation geostationary satellites, which has been done in several published papers that are never mentioned as background in Introduction nor discussed in Discussion. The accuracy of CO flux also relies on wind directions and speed, which are also a concern.
The accuracy of the new CO inventories depends on the accuracy of the derived CO emission coefficients and that of GFAS FRE. FRE calculation requires continuous FRP observations. Diurnal FRP varies very largely from day to day even for the same fire, especially large forest wildfires, which have been reported in several JGR and RSE papers. I would not expect reliable FRE to be calculated from daily mean GFAS FRP that is averaged using merely up to four daily MODIS FRP observations, although GFAS emissions are used in ECMWF forecast models. Furthermore, simply comparing it with a few other inventories doesn’t tell any information about the accuracy of the new CO inventory. There are more than 10 BB emission inventories for different purposes, and they can differ from each other by a factor of up to 30 in individual fire events although the difference in their annual budget could be much smaller. I don’t see any meaningful contributions of a new inventory to the BB community without knowing its accuracy.
To sum up, I don’t see sound contributions from this study, thus I would not recommend it for publication in ACP.Citation: https://doi.org/10.5194/egusphere-2023-649-RC1 -
AC2: 'Reply on RC1', Debora Griffin, 29 Sep 2023
We would like to thank Reviewer 1 for their review. In response to the feedback from both reviewers, we have implemented significant revisions to the manuscript. I believe there might be some misunderstanding by the reviewer and much of the criticisms raised by the first reviewer do not provide valid grounds for rejection. The specific points mentioned in the review are addressed, see attached pdf for details.
-
AC2: 'Reply on RC1', Debora Griffin, 29 Sep 2023
-
RC2: 'Comment on egusphere-2023-649', Anonymous Referee #2, 06 Jul 2023
This manuscript focuses on carbon monoxide (CO) emissions from wildfires; its goal is to produce a ~20 year global budget based on emission coefficients derived from TROPOMI and MODIS data:
- First, TROPOMI-derived emissions are calculated for the 2019-2021 period with an automated version of the flux method. To verify the robustness and uncertainties of the automated method, emissions are derived from synthetic CO column values generated with the GEM-MACH model and CFFEPS for the May-September 2019 period over North America. From this test, several filters (dealing with background CO, wind, plume geometry, etc.) are defined.
- Then, global TROPOMI-derived emissions are calculated for the 2019-2021 period, the filters defined in the previous step are applied, and CO emission coefficient values (EC) are calculated separately for 15 biomes: EC=E/FRP=emissions/fire radiative power. FRP values are from MODIS Aqua. Somehow
the GFAS FRP dataset (which results from assimilating both MODIS Terra and Aqua FRP values in GFAS) is used to provide “a guidance on total daily FRP that can then be combined with the derived ratio between TROPOMI CO emissions and MODIS FRP”; meaning unclear from the manuscript.
- Following, a global budget of CO emissions from fires is calculated for the 2003-2021 period based on TROPOMI-MODIS Aqua EC values and GFAS FRP values. Results are analyzed by region and by biome, and compared with respect to several emission inventories.
The manuscript should be improved to avoid repetitions, typos, correct wording and sentence integrity, provide explanations for acronyms/abbreviations when first used, etc. The text is hard to follow at times because of these issues. Some of the Appendices are never mentioned in the text. Expressions such as “many of the retrieved emissions”, “many outliers”, “aligns very well”, “the model agrees pretty well” should be avoided. Quantitative statements should be used instead.
Some figures would benefit from more consistent axis ranges (e.g., Fig. 1 b and c), axis labels with both axis title and units (Fig. 5). Panels labels should be properly referred to in text (e.g., Fig. 5). For clarity, please add a map showing all the regions discussed in the manuscript (CEAM, NHSA, EURO, …). Consider including a figure illustrating the areal extent of biomes.
The manuscript should justify why the TROPOMI averaging kernels are not applied to retrievals with qa_value <1. From the TROPOMI CO readme document: “We recommend using only data with a qa_value = 1 in case the averaging kernel is not applied. Data with a qa_value = 0.7 are of similar quality provided the averaging kernel is used to account for the vertical retrieval sensitivity in the presence of mid-level clouds. Quality assurance values of qa_value = 0.4 represent experimental data to be used with caution.” Table 3 in that same document provides additional information regarding qa_values, cloud heights, and tau_aer values.
While the masking effect of smoke and clouds on FRP observations is discussed several times, the effect of smoke and clouds on TROPOMI retrievals is never mentioned. This is a very important issue, given the focus on TROPOMI observations acquired over active fires, with very smoky and (potentially) very cloudy conditions. Please discuss.
In lines 319-321, to explain discrepancies between emissions from TROPOMI and GFFEPS, it is suggested that GFFEPS values may be low because of missed fires due to thick smoke. Could TROPOMI values be high due to smoke? clouds? other factors?
Figure 6 shows that the correlation between CO and FRP may not be very robust, despite R being 0.70. How would the slope (i.e., EC) change if the most extreme outliers were removed one by one?
Line 379: “The emission coefficients vary between 120 and 39 g/MJ” Please clarify that EC=120 value corresponds to (11): shrub cover, closed-open, evergreen; EC=39 to (12): shrub cover, closed-open, deciduous. The most extreme EC values correspond to rather similar biomes, please discuss.
Lines 379-380: “where the largest CO emissions relative to FRP are from broadleaved evergreen tree cover (1) and the lowest are from cultivated managed areas (16)”. How could readers see CO emissions relative to FRP? Similar issue in lines 474-475.
Line 401: “Most CO emissions are from evergreen forests (biome type 1), which also has the largest EC for CO”. Table 2 shows that biome 11 has the largest EC, please correct.
Please justify why is biome 15 (“Regularly flooded shrub and/or herbaceous cover”) included in the analysis. Biomes 7 and 8 are excluded because “fires were not observed, namely: regularly flooded tree cover (7 and 8)”
Biome # 15 (regularly flooded shrub and/or herbaceous cover; should this biome be included in the analysis, given that it is regularly flooded?) has the second largest EC (105 g/MJ). Please explain.
The manuscript assumes that EC values “do not change drastically over the years”. However, tables C1, C2, and C3 in Appendix C (never mentioned in the manuscript; please correct) show otherwise. EC values change between -50% and +150%, depending on the period analyzed. EC values change strongly among biomes; changes do not seem to follow recognizable patterns. How would the global budget change if a different set of EC values was used? Please quantify.
What is the areal extent of each biome? That information could help readers understand what biomes (and what biome’s EC) may have a stronger effect on the global budget.
Figure 9 shows emissions decreasing with time. An evaluation of the temporal trend in the global budget with respect to actual measurements (not inventories) is missing in the manuscript. Have wildfire emissions really decreased in the last two decades? While measurements of global emissions for all biomes may be unavailable, a literature search may provide useful information for specific regions/biomes.
Lines 446-447: “certain regions see increased emissions (e.g. TENA, AUST)” (text refers to Fig. 9). Hard to know for sure, but it looks like neither TENA nor AUST show increasing emissions. Either provide a better figure to illustrate the statement or remove statement. Similar issue in line 498.
Lines 476-477: “for forests we determined ECs between 64 and 120 g/MJ” Please note that EC=120 is not for a forest but for biome 11: Shrub Cover, closed-open, evergreen.
The manuscript refers to the need for multiple observations in a day (from geostationary instruments) in order to better understand fire evolution; TEMPO is mentioned but GEMS (which has been operating for a couple of years now) is not. Please discuss both.
Citation: https://doi.org/10.5194/egusphere-2023-649-RC2 -
AC1: 'Reply on RC2', Debora Griffin, 29 Sep 2023
We extend our gratitude to Reviewer 2 for their detailed review. In response to the feedback from both reviewers, we have implemented significant revisions to the manuscript. We offer detailed insights into these changes that can be found in the pdf attached.
-
AC1: 'Reply on RC2', Debora Griffin, 29 Sep 2023
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 659 | 373 | 55 | 1,087 | 38 | 50 |
- HTML: 659
- PDF: 373
- XML: 55
- Total: 1,087
- BibTeX: 38
- EndNote: 50
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Cited
Jack Chen
Kerry Anderson
Paul Makar
Chris A. McLinden
Enrico Dammers
Andre Fogal
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2077 KB) - Metadata XML