Direct Estimation of Global Anthropogenic CO<sub>2</sub> Emissions Using Satellite Data

He, Jia; Huang, Bo

doi:https://doi.org/10.5194/egusphere-2023-1347

Preprints

https://doi.org/10.5194/egusphere-2023-1347

Preprints

30 Aug 2023

| 30 Aug 2023

Direct Estimation of Global Anthropogenic CO₂ Emissions Using Satellite Data

Jia He and Bo Huang

Abstract. Reliable statistics on anthropogenic CO₂ emissions are fundamental for carbon cycle and climate change research. Satellite observations offer a potential objective and efficient alternative to the current self-reporting mechanism. However, the current satellite projects provide only column-averaged CO₂ amount (XCO₂) data. This paper proposes a direct estimation method based on satellite-based CO₂ column amount, different from the conventional “top-down” approaches, which usually adopt satellite-observed data as an indicator to disaggregate consumption statistics. Here, the monthly CO₂ emissions from 2010 to 2019 are estimated globally using CO₂ data retrieved from the Greenhouse Gases Observing Satellite. The geographically and temporally weighted regression model is adopted to account for local spatial and temporal variability. The enhanced XCO₂ data and the local wind speed, vertical velocity, air temperature, water vapor concentration, and fire emissions are included in the estimation process. The validation results of the newly derived CO₂ emissions strongly agree with the Open-source Data Inventory for Anthropogenic CO₂ data (R² = 0.929). This high global consistency demonstrates the great potential of direct estimation from satellites, with improved frequency and a broader coverage range.

Received: 20 Jun 2023 – Discussion started: 30 Aug 2023

Download & links

Jia He and Bo Huang

Status: closed

RC1:
'Comment on egusphere-2023-1347', Anonymous Referee #1, 20 Sep 2023

General comments
This manuscript attempts to estimate anthropogenic CO2 emissions using satellite observations of column-averaged CO2 amount (XCO2) data and the geographically and temporally weighted regression (GTWR) model. This study's topic is significant and timely, given the increasing interest in emission mitigation monitoring from various sectors. This study trains the regression model using the ΔXCO2 data derived from GOSAT and other ancillary variables (i.e., Wind speed, Air temperature, Total column water vapor, Fire emissions) as regressors and ODIAC’s CO2 emission estimates as a response variable. However, this study does not describe any physical basis that explains the relationship between the response variable (y) and regressors (x), making it difficult to justify scientifically. The authors evaluate the regression model by comparing the model estimates against the ODIAC CO2 emissions data. Therefore, even with the reasonable agreement between the regression model estimates and the ODIAC emission estimates, there is no supporting evidence that the emission estimate presented in this study is close to “true” emissions nor any practical advantages for policy-relevant applications. The manuscript's presentation quality is well constructed, allowing readers to follow the model development and evaluation process easily. Providing a more “Physical” basis for the regression model and demonstrating the advantage of the satellite-based regression model approach over other methods (i.e., “bottom-up” emission inventories or “top-down” satellite inversion studies) will improve the overall quality of the manuscript. The following are specific comments to the authors.
Specific comments
Line 17: In this study, the regression model (i.e., GTWR) is developed using the ODIAC fossil fuel CO2 emissions data as inputs for the training and for model validation. In the big picture, how does this regression model contribute to our current capability of tracking fossil fuel CO2 emissions? Why should one rely on GTWR model emission estimates rather than ODIAC? From this manuscript, I don’t see any evidence showing that the GTWR model estimates are closer to “true” emissions than ODIAC estimates, as section 4.1 “Validation of CO2 emissions,” only shows how close the GTWR model estimates are to the ODIAC estimates.
Lines 94-95: By definition, the XCO2 anomaly represents the column enhancement in atmospheric CO2 due to anthropogenic CO2 emissions. Correspondingly, background XCO2 is the XCO2 that would have been measured if there were no anthropogenic emissions. In this regard, how could the monthly median for each sub-region be used as the background XCO2?
In a physical sense, the monthly median XCO2 is more like a representative XCO2 value that reflects both monthly fossil-fuel emissions and biogenic CO2 fluxes within the region. Therefore, the difference between the monthly mean XCO2 from each grid cell against the monthly median XCO2 from the region will represent the spatial differences in anthropogenic (i.e., fossil fuel) and biogenic CO2 fluxes between each grid cell vs. the whole sub-region. Such “spatial difference” cannot be interpreted as the anthropogenic CO2 emissions for the region.
Also, the authors argue that using the monthly median value will de-trend the XCO2 data. However, using the monthly median, unlike the daily median used by Hakkarainen et al., 2016, is more likely to be affected by seasonality in XCO2, especially when there is a rapid shift and biospheric CO2 flux (i.e., growing season). Sensitivity analysis could be considered to test the impacts of the size of the temporal windows for averaging the satellite data (i.e., 1 day, 15 days, 1 month, etc.).
Lines 172-174: What is the physical basis of the found correlation between the anthropogenic CO2 emissions and other ancillary variables, such as air temperature and tcwv, wind speed? I could see how the air surface temperature is (non-linearly) correlated to the fossil fuel CO2 emissions: Higher energy demand for spatial heating/cooling during the winter/summer periods leads to higher fossil fuel combustion and corresponding CO2 emissions. However, I do not see any physical relationship between CO2 emissions and other variables such as wind speed and tcwv. I do not think the larger amount of water vapor in the atmospheric column is related to the surface CO2 emissions. When there is no physical relationship between the regressor and predictors, the regression model is hard to justify, and the result (even when it’s showing good R2) is likely to be overfitting of the input parameters.
Line 175: How can the strong subsiding motion be associated with significant CO2 emissions? I could understand how these ancillary variables are physically related to XCO2 data, not emissions. If that’s what the author originally intended, then the regression model should be constructed around XCO2, not emissions.
Line 187: AE is the ODIAC CO2 emissions for the regression model training, correct? If so, AE is not the “anthropogenic” emissions but CO2 emissions from fossil fuel combustion and cement production. Therefore, having fire CO2 emissions as a regressor in the model doesn’t have a physical basis.
Lines 235-237: Because the regression model presented in this study relies heavily on ODIAC emission data, spatiotemporally disaggregated bottom-up emission inventory, the emission estimates from this study are not entirely independent from the conventional “bottom-up” method.
Lines 244-246: How does the MB value of 0.05 gC/m²/month translate into a policy-relevant scale? For example, what are the MB values when the grid pixels are aggregated for annual emissions from large emitting countries (i.e., United States, China, India)? Also, what are the correlation coefficients at each sub-region?
Technical corrections
Lines 11-12: Within the atmospheric science/satellite observation research community, the “top-down” approach usually indicates the emission estimation method using the direct atmospheric observations of trace gas (i.e., Inversion analysis, mass balance analysis).
Line 16: The term enhanced XCO2 seems vague. Please use more specific terms (i.e., enhanced XCO2 relative to the regional background).
Line 27: The words “and removal” seem unnecessary in this sentence.
Figure 9: Map is too small to see some regional features in the mean bias. Increasing the map size or zooming in to -60 to 60 degrees will make the map legible.

Citation: https://doi.org/10.5194/egusphere-2023-1347-RC1
- AC1: 'Reply on RC1', Jia He, 16 Dec 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-1347/egusphere-2023-1347-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-1347-AC1
RC2:
'Comment on egusphere-2023-1347', Tomohiro Oda, 14 Oct 2023

I second what Anonymous Referee #1 discussed. As discussed by Referee #1, this study developed a method to reasonably replicate ODIAC estimates, but the study did not demonstrate its scientific significance. As presented in many previous studies, ODIAC should/does have its own errors like other emission products. For example, Oda et al. (2019) characterized modeling errors in ODIAC. There are many other papers that studied potential errors in ODIAC. ODIAC has a data policy. While users can freely use the product for research purposes, users should read papers suggested before using it mainly to reasonably understand the limitation of the data product. In the manuscript, the potential errors in ODIAC were not discussed at all. Why? Does it not matter?

I also do not see support for the policy relevance of this study. As clearly stated in our papers, ODIAC is primarily designed and developed for accurately prescribing atmospheric CO2 simulations, rather than informing policy. What climate policy do you mean? Sector? How can you possibly inform sectoral emission differences independently just using CO2? I failed to find the relevance of this study to climate mitigation policy.

I also don't think the correlation coefficient is a great metric for emission estimation. You need to know the accuracy of emission estimates if you truly wanted to do emission monitoring.

This study has a fundamental design flaw, and the conclusion is not supported. While Referee #1 kindly provided their detailed feedback, I do not believe the review process is for correcting fundamental errors or addressing knowledge gaps in the authors to make the manuscript something publishable. Our role as a referee is only to evaluate what is presented. I suggest Editor to reject the manuscript. I failed to see any chance for this study to be published in ACP and contribute to the high level science of ACP.
Lastly, the authors should respectfully use the data provided. Where possible, the authors should show original data sources and acknowledge the data properly (e.g. GOSAT). For TCCON data, did you follow the data license policy? https://tccon-wiki.caltech.edu/Main/DataLicense
Sincerely,

Tomohiro Oda (toda@usra.edu)
Reference

Oda et al. (2019) https://link.springer.com/article/10.1007/s11027-019-09877-2

Citation: https://doi.org/10.5194/egusphere-2023-1347-RC2
- AC1: 'Reply on RC1', Jia He, 16 Dec 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-1347/egusphere-2023-1347-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-1347-AC1

Status: closed

RC1:
'Comment on egusphere-2023-1347', Anonymous Referee #1, 20 Sep 2023

General comments
This manuscript attempts to estimate anthropogenic CO2 emissions using satellite observations of column-averaged CO2 amount (XCO2) data and the geographically and temporally weighted regression (GTWR) model. This study's topic is significant and timely, given the increasing interest in emission mitigation monitoring from various sectors. This study trains the regression model using the ΔXCO2 data derived from GOSAT and other ancillary variables (i.e., Wind speed, Air temperature, Total column water vapor, Fire emissions) as regressors and ODIAC’s CO2 emission estimates as a response variable. However, this study does not describe any physical basis that explains the relationship between the response variable (y) and regressors (x), making it difficult to justify scientifically. The authors evaluate the regression model by comparing the model estimates against the ODIAC CO2 emissions data. Therefore, even with the reasonable agreement between the regression model estimates and the ODIAC emission estimates, there is no supporting evidence that the emission estimate presented in this study is close to “true” emissions nor any practical advantages for policy-relevant applications. The manuscript's presentation quality is well constructed, allowing readers to follow the model development and evaluation process easily. Providing a more “Physical” basis for the regression model and demonstrating the advantage of the satellite-based regression model approach over other methods (i.e., “bottom-up” emission inventories or “top-down” satellite inversion studies) will improve the overall quality of the manuscript. The following are specific comments to the authors.
Specific comments
Line 17: In this study, the regression model (i.e., GTWR) is developed using the ODIAC fossil fuel CO2 emissions data as inputs for the training and for model validation. In the big picture, how does this regression model contribute to our current capability of tracking fossil fuel CO2 emissions? Why should one rely on GTWR model emission estimates rather than ODIAC? From this manuscript, I don’t see any evidence showing that the GTWR model estimates are closer to “true” emissions than ODIAC estimates, as section 4.1 “Validation of CO2 emissions,” only shows how close the GTWR model estimates are to the ODIAC estimates.
Lines 94-95: By definition, the XCO2 anomaly represents the column enhancement in atmospheric CO2 due to anthropogenic CO2 emissions. Correspondingly, background XCO2 is the XCO2 that would have been measured if there were no anthropogenic emissions. In this regard, how could the monthly median for each sub-region be used as the background XCO2?
In a physical sense, the monthly median XCO2 is more like a representative XCO2 value that reflects both monthly fossil-fuel emissions and biogenic CO2 fluxes within the region. Therefore, the difference between the monthly mean XCO2 from each grid cell against the monthly median XCO2 from the region will represent the spatial differences in anthropogenic (i.e., fossil fuel) and biogenic CO2 fluxes between each grid cell vs. the whole sub-region. Such “spatial difference” cannot be interpreted as the anthropogenic CO2 emissions for the region.
Also, the authors argue that using the monthly median value will de-trend the XCO2 data. However, using the monthly median, unlike the daily median used by Hakkarainen et al., 2016, is more likely to be affected by seasonality in XCO2, especially when there is a rapid shift and biospheric CO2 flux (i.e., growing season). Sensitivity analysis could be considered to test the impacts of the size of the temporal windows for averaging the satellite data (i.e., 1 day, 15 days, 1 month, etc.).
Lines 172-174: What is the physical basis of the found correlation between the anthropogenic CO2 emissions and other ancillary variables, such as air temperature and tcwv, wind speed? I could see how the air surface temperature is (non-linearly) correlated to the fossil fuel CO2 emissions: Higher energy demand for spatial heating/cooling during the winter/summer periods leads to higher fossil fuel combustion and corresponding CO2 emissions. However, I do not see any physical relationship between CO2 emissions and other variables such as wind speed and tcwv. I do not think the larger amount of water vapor in the atmospheric column is related to the surface CO2 emissions. When there is no physical relationship between the regressor and predictors, the regression model is hard to justify, and the result (even when it’s showing good R2) is likely to be overfitting of the input parameters.
Line 175: How can the strong subsiding motion be associated with significant CO2 emissions? I could understand how these ancillary variables are physically related to XCO2 data, not emissions. If that’s what the author originally intended, then the regression model should be constructed around XCO2, not emissions.
Line 187: AE is the ODIAC CO2 emissions for the regression model training, correct? If so, AE is not the “anthropogenic” emissions but CO2 emissions from fossil fuel combustion and cement production. Therefore, having fire CO2 emissions as a regressor in the model doesn’t have a physical basis.
Lines 235-237: Because the regression model presented in this study relies heavily on ODIAC emission data, spatiotemporally disaggregated bottom-up emission inventory, the emission estimates from this study are not entirely independent from the conventional “bottom-up” method.
Lines 244-246: How does the MB value of 0.05 gC/m²/month translate into a policy-relevant scale? For example, what are the MB values when the grid pixels are aggregated for annual emissions from large emitting countries (i.e., United States, China, India)? Also, what are the correlation coefficients at each sub-region?
Technical corrections
Lines 11-12: Within the atmospheric science/satellite observation research community, the “top-down” approach usually indicates the emission estimation method using the direct atmospheric observations of trace gas (i.e., Inversion analysis, mass balance analysis).
Line 16: The term enhanced XCO2 seems vague. Please use more specific terms (i.e., enhanced XCO2 relative to the regional background).
Line 27: The words “and removal” seem unnecessary in this sentence.
Figure 9: Map is too small to see some regional features in the mean bias. Increasing the map size or zooming in to -60 to 60 degrees will make the map legible.

Citation: https://doi.org/10.5194/egusphere-2023-1347-RC1
- AC1: 'Reply on RC1', Jia He, 16 Dec 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-1347/egusphere-2023-1347-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-1347-AC1
RC2:
'Comment on egusphere-2023-1347', Tomohiro Oda, 14 Oct 2023

I second what Anonymous Referee #1 discussed. As discussed by Referee #1, this study developed a method to reasonably replicate ODIAC estimates, but the study did not demonstrate its scientific significance. As presented in many previous studies, ODIAC should/does have its own errors like other emission products. For example, Oda et al. (2019) characterized modeling errors in ODIAC. There are many other papers that studied potential errors in ODIAC. ODIAC has a data policy. While users can freely use the product for research purposes, users should read papers suggested before using it mainly to reasonably understand the limitation of the data product. In the manuscript, the potential errors in ODIAC were not discussed at all. Why? Does it not matter?

I also do not see support for the policy relevance of this study. As clearly stated in our papers, ODIAC is primarily designed and developed for accurately prescribing atmospheric CO2 simulations, rather than informing policy. What climate policy do you mean? Sector? How can you possibly inform sectoral emission differences independently just using CO2? I failed to find the relevance of this study to climate mitigation policy.

I also don't think the correlation coefficient is a great metric for emission estimation. You need to know the accuracy of emission estimates if you truly wanted to do emission monitoring.

This study has a fundamental design flaw, and the conclusion is not supported. While Referee #1 kindly provided their detailed feedback, I do not believe the review process is for correcting fundamental errors or addressing knowledge gaps in the authors to make the manuscript something publishable. Our role as a referee is only to evaluate what is presented. I suggest Editor to reject the manuscript. I failed to see any chance for this study to be published in ACP and contribute to the high level science of ACP.
Lastly, the authors should respectfully use the data provided. Where possible, the authors should show original data sources and acknowledge the data properly (e.g. GOSAT). For TCCON data, did you follow the data license policy? https://tccon-wiki.caltech.edu/Main/DataLicense
Sincerely,

Tomohiro Oda (toda@usra.edu)
Reference

Oda et al. (2019) https://link.springer.com/article/10.1007/s11027-019-09877-2

Citation: https://doi.org/10.5194/egusphere-2023-1347-RC2
- AC1: 'Reply on RC1', Jia He, 16 Dec 2023
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2023/egusphere-2023-1347/egusphere-2023-1347-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2023-1347-AC1

Jia He and Bo Huang

Viewed

Total article views: 738 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
481	234	23	738	11	14

HTML: 481
PDF: 234
XML: 23
Total: 738
BibTeX: 11
EndNote: 14

Views and downloads (calculated since 30 Aug 2023)

Month	HTML	PDF	XML	Total
Aug 2023	119	44	3	166
Sep 2023	134	69	8	211
Oct 2023	87	44	4	135
Nov 2023	15	12	0	27
Dec 2023	42	23	3	68
Jan 2024	29	10	1	40
Feb 2024	10	10	0	20
Mar 2024	26	13	0	39
Apr 2024	19	9	4	32

Cumulative views and downloads (calculated since 30 Aug 2023)

Month	HTML	PDF	XML	Total
Aug 2023	119	44	3	166
Sep 2023	134	69	8	211
Oct 2023	87	44	4	135
Nov 2023	15	12	0	27
Dec 2023	42	23	3	68
Jan 2024	29	10	1	40
Feb 2024	10	10	0	20
Mar 2024	26	13	0	39
Apr 2024	19	9	4	32

Viewed (geographical distribution)

Total article views: 719 (including HTML, PDF, and XML) Thereof 719 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 27 Apr 2024

Short summary

A feasible method to accurately estimate CO₂ emissions from satellite-derived column amounts of CO₂ was proposed in this study. The GTWR model is adopted for CO₂ estimation for the first time to account for its local spatial and temporal variability. The monthly CO₂ emissions from 2010 to 2019 are estimated globally using the enhanced XCO₂ from the GOSAT. The newly derived anthropogenic CO₂ emission dataset reveals the variation trend of CO₂ emission on the global scale.


Total:	0
HTML:	0
PDF:	0
XML:	0

Direct Estimation of Global Anthropogenic CO2 Emissions Using Satellite Data

Viewed

Viewed (geographical distribution)

Direct Estimation of Global Anthropogenic CO₂ Emissions Using Satellite Data