the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
A global CO2 flux dataset (2015–2019) inferred from OCO-2 retrievals using the Tan-Tracker inversion system
Abstract. Accurate assessment of the various sources and sinks of carbon dioxide (CO2), especially terrestrial ecosystem and ocean fluxes with high uncertainties, is important for understanding of the global carbon cycle, supporting the formulation of climate policies, and projecting future climate change. Satellite retrievals of the column-averaged dry air mole fractions of CO2 (XCO2) are being widely used to improve carbon flux estimation due to their broad spatial coverage. However, there is no consensus on the robust estimates of regional fluxes. In this study, we present a global and regional resolved terrestrial ecosystem carbon flux (NEE) and ocean carbon flux dataset for 2015–2019. The dataset was generated using the Tan-Tracker inversion system by assimilating Observing Carbon Observatory 2 (OCO-2) column CO2 retrievals. The posterior NEE and ocean carbon fluxes were comprehensively validated by comparing posterior simulated CO2 concentrations with OCO-2 independent retrievals and Total Carbon Column Observing Network (TCCON) measurements. The validation showed that posterior carbon fluxes significantly improved the modelling of atmospheric CO2 concentrations, with global mean biases of 0.33 ppm against OCO-2 retrievals and 0.12 ppm against TCCON measurements. We described the characteristics of the dataset at global, regional, and Tibetan Plateau scales in terms of the carbon budget, annual and seasonal variations, and spatial distribution. The posterior 5-year annual mean global atmospheric CO2 growth rate was 5.35 PgC yr−1, which was within the uncertainty of the Global Carbon Budget 2020 estimate (5.49 PgC yr−1). The posterior annual mean NEE and ocean carbon fluxes were −4.07 and −3.33 PgC yr−1, respectively. Regional fluxes were analysed based on TransCom partitioning. All 11 land regions acted as carbon sinks, except for Tropical South America, which was almost neutral. The strongest carbon sinks were located in Boreal Asia, followed by Temperate Asia and North Africa. The entire Tibetan Plateau ecosystem was estimated as a carbon sink, taking up −49.52 TgC yr−1 on average, with the strongest sink occurring in eastern alpine meadows. These results indicate that our dataset captures surface carbon fluxes well and provides insight into the global carbon cycle. The dataset can be accessed at https://doi.org/10.11888/Meteoro.tpdc.271317 (Jin et al., 2021).
- Preprint
(4085 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on essd-2021-210', Anonymous Referee #1, 24 Aug 2021
This paper describes a global terrestrial biosphere and oceanic CO2 flux dataset (2015-2019) inferred from OCO-2 b9 retrievals using Tan-Tracker inversion system. The fluxes are provided daily at 2° (latitude) x 2.5°(longitude) spatial resolution. The dataset was evaluated against OCO-2 retrievals and column CO2 observations from TCCON observation network.
While high quality carbon flux estimates are urgently needed to provide policy support and to advance global carbon cycle science, I find four major caveats in generating the dataset:
- The data thinning and filtering method. The study used several steps to remove outliers by comparing to model simulations, and to make sure the number of observations is less than 20000. There is nothing wrong to filter the observations before assimilations. However, the observations spatial gradient after filtering (Figure 2) is so different from the original observations that the assimilated obs could hardly be treated as OCO-2 retrievals anymore. For example, the mean value over northern Amazon is about 405 ppm before filtering, but it is equal or less than 400 ppm after filtering. The study used prior fluxes that were constrained by surface observation network, which is very sparse in the tropics. Thus, the strict filtering can filter out the real signals in the data. It is not clear in the paper what the bases are for the filtering threshold (e.g., 2ppm absolute differences between observations and model simulated values).
- Relative biases between ocean glint and land observations. This study used both ocean glint observations and land observations in the flux inversion. However, it was shown that ocean glint and land observations of the ACOS-OCO2 b9 retrievals have relative biases, and normally only inversions using land observations are discussed (1–3).
- The inferred flux spatial pattern. This study showed that the net fluxes across the 11 land TransCom regions are net sink except in tropical south America (Table 5), which is very different from previous studies that showed large efflux over northern tropical Africa (2, 4). This difference is most likely due to the aggressive filtering method. Since the inferred fluxes from this study are very similar to the prescribed prior fluxes (Figure 8), it is hard to say how much the posterior fluxes are really constrained by OCO-2 retrievals.
- Uncertainties for the posterior fluxes are not provided. Even though the regional aggregated fluxes have uncertainty estimates, the uncertainties are not provided in the gridded dataset. I would suggest including the uncertainty estimates.
- Evaluation dataset. The study used independent OCO-2 retrievals and TCCON observation network to evaluate the posterior fluxes. However, it is not clear how the independent OCO-2 retrievals were selected. Since OCO-2 retrievals were all retrieved with the same methodology, they are not totally independent even some of those retrievals were not assimilated. Normally, aircraft CO2 observations and surface CO2 flask observations are used as independent observations to evaluate posterior fluxes.
Crowell,S., Baker,D., Schuh,A., Basu,S., Jacobson,A.R., Chevallier,F., Liu,J., Deng,F., Feng,L., McKain,K., et al. (2019) The 2015–2016 carbon cycle as seen from OCO-2 and the global in situ network. Atmos. Chem. Phys., 19, 9797–9831.
https://doi.org/10.5194/acp-19-9797-2019
Liu,J., Baskaran,L., Bowman,K., Schimel,D., Bloom,A.A., Parazoo,N.C., Oda,T., Carroll,D., Menemenlis,D., Joiner,J., et al. (2021) Carbon Monitoring System Flux Net Biosphere Exchange 2020 (CMS-Flux NBE 2020). Earth Syst. Sci. Data, 13, 299–330.
https://doi.org/10.5194/essd-13-299-2021
Peiro,H., Crowell,S., Schuh,A., Baker,D.F., O’Dell,C., Jacobson,A.R., Chevallier,F., Liu,J., Eldering,A., Crisp,D., et al. (2021) Four years of global carbon cycle observed from OCO-2 version 9 and \textit{in situ} data, and comparison to OCO-2 v7. Atmos. Chem. Phys. Discuss., 2021, 1–50.
https://doi.org/10.5194/acp-2021-373
Palmer,P.I., Feng,L., Baker,D., Chevallier,F., Bösch,H. and Somkuti,P. (2019) dominate pan-tropical atmospheric CO 2 signal. 10.1038/s41467-019-11097-w.
https://doi.org/10.1038/s41467-019-11097-w
Citation: https://doi.org/10.5194/essd-2021-210-RC1 -
RC2: 'Comment on essd-2021-210', Anonymous Referee #2, 26 Aug 2021
An accurate grided global terrestrial ecosystem carbon flux data is of great importance in global carbon cycle research. This study presents a global CO2 flux dataset (2015–2019) inferred from OCO-2 retrievals using the Tan-Tracker inversion system, which uses a dual-pass inversion strategy and the nonlinear least squares four-dimensional variational data assimilation (NLS-4DVar) method. The characteristics of the dataset, including annual and seasonal variations, and spatial distribution at global and regional scales, and over Tibetan Plateau, are analyzed, and the evaluations against unassimilated XCO2 retrievals and TCCON observations are performed. The validation showed that posterior carbon fluxes significantly improved the modelling of atmospheric CO2 concentrations, with global mean biases of 0.33 ppm against OCO-2 retrievals and 0.12 ppm against TCCON measurements. However, actually, the observation data used for evaluations is not completely independent from the assimilated XCO2 data, the unassimilated and assimilated XCO2 data are obtained by the same satellite and the same inversion algorithm, and there is a great correlation between adjacent data. The TCCON data has also been used during the inversion and deviation correction of the XCO2 retrievals. Moreover, the CO2 mixing ratios and CO2 fluxes are optimized synchronously with the dual-pass scheme, thus the improvement in atmospheric CO2 concentrations does not entirely represent the improvements in CO2 fluxes. In addition, the difference between the posterior flux and the prior flux in the regions outside Europe is very small, and the interannual variations (IAV) are almost the same in all 11 Transcom regions, indicating that the dataset presented here basically does not improve our understanding of carbon fluxes in different regions. In summary, I think the reliability of this dataset needs to be further verified, and its usefulness in global and regional carbon cycle research needs to be further elucidated.
Other issues:
- Line 64, the response of carbon fluxes in the Tibetan Plateau ecosystems play an important role in global carbon cycle. This description is inaccurate, because the carbon source or sink of the Tibetan Plateau is very small compared with those of global ecosystems.
- Section 2.2, “fossil fuel emission and biospheric inventories used to count CO2 precursor species (CO, CH4 and other carbon gases) as direct CO2 emissions at the surface”, How is this item considered? How was that adjustment performed? The author needs to describe it clearly. In addition, the inventories of fossil fuel emissions, biomass burning emissions, ship emissions, aviation emissions, biofuel burning emissions also need to be introduced, including the name, source, time range, and spatial resolution, etc. Moreover, as shown in Table A1, there are no IAVs in the biofuel burning emissions, ship emissions, and chemical source, which are inconsistent with the real situation, therefore, the impact of these assumptions on the inverted CO2 fluxes also needs to be discussed in depth.
- Line 141, “The prior NEE was obatined from CarbonTracker CT2019B”, CT2019B released its prior flux (i.e., CASA simulations) and posterior flux, which one is used in this study?
- Section 2.3, XCO2 data thinning and filtering were conducted in this study. It is necessary. However, “differences between OCO-2 XCO2 and corresponding model-simulated XCO2 were considered. The observation was discarded when the absolute difference exceeded 2 ppm”. This step is unreasonable and unfounded. The essence of inversion is to correct the flux error based on the deviation between simulation and observation. Now that the observations with large model-data mismatch errors are discarded, the effect of observation on flux is significantly reduced. In addition, with the accumulation of errors, the deviation between simulation and observation may become larger year by year (Figure 11), which may result in very few assimilated observations in the last few years.
- Section 2.4, the uncertainties used in this system may be too small, as show in Figure 5 and Table 5, the monthly uncertainty and uncertainty in each region are very small. This setting may be one of the reasons why the fluxes of each area in this study have little changes.
- Section 2.5, the TCCON data was used in this study, but almost no references are cited. It is recommended that the author carefully read the TCCON data usage.
- Section 4.4.1, It is not appropriate to directly compare the inverted NEE with GCP and JCS results, because there are large differences about the rest fluxes in the terrestrial ecosystems among these studies. This study considers the biomass burning and biofuel emissions, GCP considered the land use change emissions and had unbalanced item, while in JCS, only biomass burning emission was excluded. Additionally, the inverted ocean sink is too large, which exceeds the uncertainty range of the GCP estimate.
- Figure 4, Why is the NEE of CT2019 not given?
- Section 4.1.2, This study only reduces carbon sinks in summer and carbon emissions in winter, which means that it reduces the seasonal amplitude of NEE. The author needs to provide evidence to show that this reduction in amplitude is reasonable. In addition, 2015/2016 have a strong El Niño event. Many studies have shown that the carbon sinks in 15/16 have been significantly reduced, but from this result, the carbon sinks in the summer of these two years are almost the same as in other years, and even larger. the author needs to demonstrate it.
- Section 4.1.3, The URs in this study is significantly lower than the results of previous studies (40~70%), and the URs on the ocean is greater than that on land, which is contrary to the results of others. Such a low UR (<10%) shows that the system that the system has a limited effect on the reduction of carbon flux uncertainty, which reduces the feasibility of the results of the paper.
- Section 4.3, The author analyzed and discussed the inverted carbon sinks in the Qinghai-Tibet Plateau (QTP) in detail. Usually, in plateau areas, due to the influence of complex terrain and high albedo, the retrieve error of XCO2 may be relatively large, and the amount of observation data may be relatively small. The author needs to check the data amount and errors in this area. As shown in Figure 9, the difference between the prior and posterior are very small, suggesting that the constraint of XCO2 on the NEE of QTP is very weak. Therefore, this result may not make much sense.
- Section 6, till now, there are other NEE datasets inferred from OCO-2 retrievals, like the CMS-Flux NBE 2020 (Liu et al., 2021@ESSD) and the OCO-2 v9 MIP (https://gml.noaa.gov/ccgg/OCO2_v9mip/). I suggest that the author make a comprehensive comparison with other existing results.
Citation: https://doi.org/10.5194/essd-2021-210-RC2
Status: closed
-
RC1: 'Comment on essd-2021-210', Anonymous Referee #1, 24 Aug 2021
This paper describes a global terrestrial biosphere and oceanic CO2 flux dataset (2015-2019) inferred from OCO-2 b9 retrievals using Tan-Tracker inversion system. The fluxes are provided daily at 2° (latitude) x 2.5°(longitude) spatial resolution. The dataset was evaluated against OCO-2 retrievals and column CO2 observations from TCCON observation network.
While high quality carbon flux estimates are urgently needed to provide policy support and to advance global carbon cycle science, I find four major caveats in generating the dataset:
- The data thinning and filtering method. The study used several steps to remove outliers by comparing to model simulations, and to make sure the number of observations is less than 20000. There is nothing wrong to filter the observations before assimilations. However, the observations spatial gradient after filtering (Figure 2) is so different from the original observations that the assimilated obs could hardly be treated as OCO-2 retrievals anymore. For example, the mean value over northern Amazon is about 405 ppm before filtering, but it is equal or less than 400 ppm after filtering. The study used prior fluxes that were constrained by surface observation network, which is very sparse in the tropics. Thus, the strict filtering can filter out the real signals in the data. It is not clear in the paper what the bases are for the filtering threshold (e.g., 2ppm absolute differences between observations and model simulated values).
- Relative biases between ocean glint and land observations. This study used both ocean glint observations and land observations in the flux inversion. However, it was shown that ocean glint and land observations of the ACOS-OCO2 b9 retrievals have relative biases, and normally only inversions using land observations are discussed (1–3).
- The inferred flux spatial pattern. This study showed that the net fluxes across the 11 land TransCom regions are net sink except in tropical south America (Table 5), which is very different from previous studies that showed large efflux over northern tropical Africa (2, 4). This difference is most likely due to the aggressive filtering method. Since the inferred fluxes from this study are very similar to the prescribed prior fluxes (Figure 8), it is hard to say how much the posterior fluxes are really constrained by OCO-2 retrievals.
- Uncertainties for the posterior fluxes are not provided. Even though the regional aggregated fluxes have uncertainty estimates, the uncertainties are not provided in the gridded dataset. I would suggest including the uncertainty estimates.
- Evaluation dataset. The study used independent OCO-2 retrievals and TCCON observation network to evaluate the posterior fluxes. However, it is not clear how the independent OCO-2 retrievals were selected. Since OCO-2 retrievals were all retrieved with the same methodology, they are not totally independent even some of those retrievals were not assimilated. Normally, aircraft CO2 observations and surface CO2 flask observations are used as independent observations to evaluate posterior fluxes.
Crowell,S., Baker,D., Schuh,A., Basu,S., Jacobson,A.R., Chevallier,F., Liu,J., Deng,F., Feng,L., McKain,K., et al. (2019) The 2015–2016 carbon cycle as seen from OCO-2 and the global in situ network. Atmos. Chem. Phys., 19, 9797–9831.
https://doi.org/10.5194/acp-19-9797-2019
Liu,J., Baskaran,L., Bowman,K., Schimel,D., Bloom,A.A., Parazoo,N.C., Oda,T., Carroll,D., Menemenlis,D., Joiner,J., et al. (2021) Carbon Monitoring System Flux Net Biosphere Exchange 2020 (CMS-Flux NBE 2020). Earth Syst. Sci. Data, 13, 299–330.
https://doi.org/10.5194/essd-13-299-2021
Peiro,H., Crowell,S., Schuh,A., Baker,D.F., O’Dell,C., Jacobson,A.R., Chevallier,F., Liu,J., Eldering,A., Crisp,D., et al. (2021) Four years of global carbon cycle observed from OCO-2 version 9 and \textit{in situ} data, and comparison to OCO-2 v7. Atmos. Chem. Phys. Discuss., 2021, 1–50.
https://doi.org/10.5194/acp-2021-373
Palmer,P.I., Feng,L., Baker,D., Chevallier,F., Bösch,H. and Somkuti,P. (2019) dominate pan-tropical atmospheric CO 2 signal. 10.1038/s41467-019-11097-w.
https://doi.org/10.1038/s41467-019-11097-w
Citation: https://doi.org/10.5194/essd-2021-210-RC1 -
RC2: 'Comment on essd-2021-210', Anonymous Referee #2, 26 Aug 2021
An accurate grided global terrestrial ecosystem carbon flux data is of great importance in global carbon cycle research. This study presents a global CO2 flux dataset (2015–2019) inferred from OCO-2 retrievals using the Tan-Tracker inversion system, which uses a dual-pass inversion strategy and the nonlinear least squares four-dimensional variational data assimilation (NLS-4DVar) method. The characteristics of the dataset, including annual and seasonal variations, and spatial distribution at global and regional scales, and over Tibetan Plateau, are analyzed, and the evaluations against unassimilated XCO2 retrievals and TCCON observations are performed. The validation showed that posterior carbon fluxes significantly improved the modelling of atmospheric CO2 concentrations, with global mean biases of 0.33 ppm against OCO-2 retrievals and 0.12 ppm against TCCON measurements. However, actually, the observation data used for evaluations is not completely independent from the assimilated XCO2 data, the unassimilated and assimilated XCO2 data are obtained by the same satellite and the same inversion algorithm, and there is a great correlation between adjacent data. The TCCON data has also been used during the inversion and deviation correction of the XCO2 retrievals. Moreover, the CO2 mixing ratios and CO2 fluxes are optimized synchronously with the dual-pass scheme, thus the improvement in atmospheric CO2 concentrations does not entirely represent the improvements in CO2 fluxes. In addition, the difference between the posterior flux and the prior flux in the regions outside Europe is very small, and the interannual variations (IAV) are almost the same in all 11 Transcom regions, indicating that the dataset presented here basically does not improve our understanding of carbon fluxes in different regions. In summary, I think the reliability of this dataset needs to be further verified, and its usefulness in global and regional carbon cycle research needs to be further elucidated.
Other issues:
- Line 64, the response of carbon fluxes in the Tibetan Plateau ecosystems play an important role in global carbon cycle. This description is inaccurate, because the carbon source or sink of the Tibetan Plateau is very small compared with those of global ecosystems.
- Section 2.2, “fossil fuel emission and biospheric inventories used to count CO2 precursor species (CO, CH4 and other carbon gases) as direct CO2 emissions at the surface”, How is this item considered? How was that adjustment performed? The author needs to describe it clearly. In addition, the inventories of fossil fuel emissions, biomass burning emissions, ship emissions, aviation emissions, biofuel burning emissions also need to be introduced, including the name, source, time range, and spatial resolution, etc. Moreover, as shown in Table A1, there are no IAVs in the biofuel burning emissions, ship emissions, and chemical source, which are inconsistent with the real situation, therefore, the impact of these assumptions on the inverted CO2 fluxes also needs to be discussed in depth.
- Line 141, “The prior NEE was obatined from CarbonTracker CT2019B”, CT2019B released its prior flux (i.e., CASA simulations) and posterior flux, which one is used in this study?
- Section 2.3, XCO2 data thinning and filtering were conducted in this study. It is necessary. However, “differences between OCO-2 XCO2 and corresponding model-simulated XCO2 were considered. The observation was discarded when the absolute difference exceeded 2 ppm”. This step is unreasonable and unfounded. The essence of inversion is to correct the flux error based on the deviation between simulation and observation. Now that the observations with large model-data mismatch errors are discarded, the effect of observation on flux is significantly reduced. In addition, with the accumulation of errors, the deviation between simulation and observation may become larger year by year (Figure 11), which may result in very few assimilated observations in the last few years.
- Section 2.4, the uncertainties used in this system may be too small, as show in Figure 5 and Table 5, the monthly uncertainty and uncertainty in each region are very small. This setting may be one of the reasons why the fluxes of each area in this study have little changes.
- Section 2.5, the TCCON data was used in this study, but almost no references are cited. It is recommended that the author carefully read the TCCON data usage.
- Section 4.4.1, It is not appropriate to directly compare the inverted NEE with GCP and JCS results, because there are large differences about the rest fluxes in the terrestrial ecosystems among these studies. This study considers the biomass burning and biofuel emissions, GCP considered the land use change emissions and had unbalanced item, while in JCS, only biomass burning emission was excluded. Additionally, the inverted ocean sink is too large, which exceeds the uncertainty range of the GCP estimate.
- Figure 4, Why is the NEE of CT2019 not given?
- Section 4.1.2, This study only reduces carbon sinks in summer and carbon emissions in winter, which means that it reduces the seasonal amplitude of NEE. The author needs to provide evidence to show that this reduction in amplitude is reasonable. In addition, 2015/2016 have a strong El Niño event. Many studies have shown that the carbon sinks in 15/16 have been significantly reduced, but from this result, the carbon sinks in the summer of these two years are almost the same as in other years, and even larger. the author needs to demonstrate it.
- Section 4.1.3, The URs in this study is significantly lower than the results of previous studies (40~70%), and the URs on the ocean is greater than that on land, which is contrary to the results of others. Such a low UR (<10%) shows that the system that the system has a limited effect on the reduction of carbon flux uncertainty, which reduces the feasibility of the results of the paper.
- Section 4.3, The author analyzed and discussed the inverted carbon sinks in the Qinghai-Tibet Plateau (QTP) in detail. Usually, in plateau areas, due to the influence of complex terrain and high albedo, the retrieve error of XCO2 may be relatively large, and the amount of observation data may be relatively small. The author needs to check the data amount and errors in this area. As shown in Figure 9, the difference between the prior and posterior are very small, suggesting that the constraint of XCO2 on the NEE of QTP is very weak. Therefore, this result may not make much sense.
- Section 6, till now, there are other NEE datasets inferred from OCO-2 retrievals, like the CMS-Flux NBE 2020 (Liu et al., 2021@ESSD) and the OCO-2 v9 MIP (https://gml.noaa.gov/ccgg/OCO2_v9mip/). I suggest that the author make a comprehensive comparison with other existing results.
Citation: https://doi.org/10.5194/essd-2021-210-RC2
Data sets
Tan-Tracker global daily NEE and ocean carbon fluxes for 2015-2019 (TT2021 dataset) Jin, Z., Tian, X., Han, R., Fu, Y., Li, X., Gao, J., Mao, H., Chen, C. https://doi.org/10.11888/Meteoro.tpdc.271317
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
1,132 | 491 | 81 | 1,704 | 73 | 90 |
- HTML: 1,132
- PDF: 491
- XML: 81
- Total: 1,704
- BibTeX: 73
- EndNote: 90
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1