the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 26 Jan 2024
Improved definition of prior uncertainties in CO2 and CO fossil fuel fluxes and the impact on a multi-species inversion with GEOS-Chem (v12.5)
Abstract. Monitoring, Reporting and Verification (MRV) frameworks for greenhouse gas (GHG) emissions are being developed by countries across the world to keep track of progress towards national emission reduction targets. Data assimilation plays an important role in MRVs, combining different sources of information to get the best possible estimate of fossil fuel emissions and as a consequence better estimates for fluxes from the natural biosphere. Robust estimates for fossil fuel emissions rely on accurate estimates of uncertainties corresponding to the different pieces of information. We describe prior uncertainties in CO2 and CO fossil fuel fluxes, with special attention paid to spatial error correlations and the covariance structure between CO2 and CO. This represents the first time that the prior uncertainties in CO2 and the important co-emitted trace gas CO are defined consistently, including error correlations, which allows us to make use of the synergy between the two trace gases to better constrain CO2 fossil fuel fluxes. The CO:CO2 error correlations differ per sector, depending on the diversity of sub-processes occurring within a sector, and also show a large range in values between pixels for the same sector. For example, for other stationary combustion the pixel correlation values range from 0.1 to 1.0, whereas for road transport the correlation is mostly larger than 0.6. We illustrate the added value of our prior uncertainty definition using closed-loop numerical experiments over mainland Europe and the UK, which isolate the influence of using error correlations between CO2 and CO and the influence of prescribing more detailed information about prior emission uncertainties. We find that using our realistic prior uncertainty definition helps our data assimilation system to differentiate more easily between CO2 fluxes from biogenic and fossil fuel sources. Using the improved prior emission uncertainties we find fewer geographic regions with significant changes from the prior than using the default prior uncertainties, but they almost consistently move closer to the prescribed true values, in contrast to the default prior uncertainties. We also find that using CO provides additional information on CO2 fossil fuel fluxes, but only if the CO:CO2 error covariance structure is defined realistically. Using the default prior uncertainties, the CO2 fossil fuel fluxes move farther away from the truth for many geographical regions. With the default uncertainties the maximum deviation of fossil fuel CO2 from the prescribed truth is about 7 % in both the prior and posterior result. With the advanced uncertainties this is reduced to 3 % in the posterior.
- Preprint
(2500 KB) - Metadata XML
-
Supplement
(1448 KB) - BibTeX
- EndNote
Status: open (extended)
-
RC1: 'Comment on egusphere-2023-2025', Fabian Maier, 21 Apr 2024
reply
This paper gives a detailed estimation of prior uncertainties in CO2 and CO fossil fuel fluxes. This includes the derivation of spatial error correlation lengths and CO2:CO error correlations from an in-depth analysis of spatial proxy maps. Synthetic CO2-CO multi-species inversion experiments reveal the importance of advanced prior uncertainty definitions to more reliably differentiate between biosphere and fossil fuel CO2 fluxes and to gain advantage from CO in constraining fossil emissions.
Overall, this is a very relevant and good study, which provides the basis for exploiting similarities in the CO2 and CO emission patterns with multi-species inversions. The manuscript is well and illustratively written. Please find below some suggestions for clarification. I recommend acceptance with these minor revisions.
Minor comments:
- Estimating and compiling all the uncertainties shown in this study is a tough task. The authors have done this very carefully. However, as also mentioned in this study, some uncertainty estimates are based on expert judgements; e.g., it was assumed that the representativeness error is of a similar order of magnitude as the proxy value uncertainty. I wonder how this choice affects the results of the closed-loop experiments, considering that the representativeness error may cause spatial correlations. Maybe this could be addressed by performing a sensitivity study with representativeness errors that are slightly varied. However, I do not know how easily such an analysis can be conducted.
- In Sect. 3.3 the authors nicely describe and explain the results of the closed-loop experiments, which are also shown in Fig. 8 – 10. However, I think this section would benefit from some numbers or statistics to back up the statements made. I would also be interested to know whether the differences between the CO2 only and CO2 & CO experiments are significant.
- Unfortunately, I could not fully follow the calculations of the predictor P in Sect. 2.2.2. I think this section needs some revision. Most importantly, please explain for what the vector W stands and indicate over which elements the sums in the denominators in Eq. 2 are calculated. Please also see my specific questions/remarks below. More generally, the predictor was introduced to estimate the error correlation between CO and CO2. Thus, I would have expected a positive correlation between predictor value and the Monte-Carlo based correlation coefficients. However, it turns out that the Monte-Carlo based correlation coefficients decrease with increasing predictor values (e.g. Fig. 6a). What is the reason for this? Explaining this in more detail might also help the reader to better understand the predictor definition.
Specific comments:
Abstract:
L. 26-30: Can you strengthen these statements by providing some numbers? In order to interpret and classify your results, it may be useful to briefly describe here how the true and prior values were determined for your synthetic experiment.
Introduction:
L. 51ff: If you like, you could also mention 14C here, which is the gold standard for separating fossil fuel CO2. You could then further motivate the usage of co-emitted species like CO and NOx by referring to the typically low temporal resolution and poor spatial coverage of the 14C measurements.
L. 73: Since the term “emission factor” is used frequently in your study, it may be useful to provide a brief definition and/or its units here.
L. 77-78: The last part of this sentence is a bit confusing. Do you mean “the emission factor of those (emission) sectors for which the error is not correlated to the error in the CO2 emission factor”?
L. 82: Please add “relative” before “error”, i.e. “So that the relative error in CO2 emissions…”.
Methods:
Throughout: There are several abbreviations, which are not spelled out in full, e.g. “TNO”, “CAMS-REG”, “CEIP”, “LRTAP”, “UNFCCC”, “STEAM”, “IPCC”, “EMEP”, “VISUM”, “GMAO”, “CASA-GFED”. I think the EGU guidelines require to spell out the full names for such abbreviations.
L. 155-157: “Where needed” seems to be a bit vague. Can you give some illustrative examples, e.g. for which situations you selected the largest uncertainties from a given range? This can also be done in the supplements.
L. 169: Maybe you could even say “…, which are the most important contributors to CO and CO2 emissions from area sources…”.
L. 172-173: I didn’t get this sentence. Are there also countries for which emissions are not downscaled by using density population or traffic volume proxy maps? Please clarify.
L. 195: Please explain what “categorical proxies” are.
L. 220ff: It might be useful to briefly introduce the concept of semi-variograms here, as you derive from those the spatial error correlation lengths.
L. 246-247: Do you mean “… the relative contribution of each proxy map … “? Also, what does the vector WD mean? Does the vector WD require subscripts? Do you calculate in Eq. 3 the standard deviation and the maximum of the different WD values of the contributing proxy maps in the respective grid cells? Please clarify.
L. 254: The subscript “C” of “P_C” in Eq. 3 can easily be mixed up with “c” for grid cell. Please clearly indicate that “P_C” and “P_F” are the predictors for the GNFR C and GNFR F sectors. More importantly, why do you need different predictor definitions for the two different sectors?
L. 255: It may be useful to add here a brief description on how the Monte-Carlo approach was applied to calculate the correlation coefficients.
L. 292ff: It would be helpful if you could split Eq. 6 into two (or even three) equations: one equation for the aggregation of the sub-sector emission uncertainty, one equation for the calculation of the grid cell relative uncertainty shown in Fig. 3, and one equation for the calculation of the weighted average correlation length. Then you could also refer to the respective equations in Fig. 2 more properly.
L. 323-324: Do you use the same transport model (GEOS-Chem) for the global simulation? And what is the temporal resolution of the anthropogenic CO2 and CO emissions, i.e. do you apply the diurnal time profiles from TNO?
L. 373: Are you using hourly CO2 and CO observations? Also, considering the goals of the World Meteorological Organization (WMO) for the in-situ CO2 measurement uncertainty of 0.1ppm, I would expect the 2ppm CO2 uncertainty as an upper limit. Maybe you can state this.
Results:
Fig. 5: It would be easier to compare the different panels, if the two panels on the left (and right) side have the same x-axis.
L. 413: This sentence is unclear to me. Could you please explain this with a bit more detail. Maybe you could also refer to the respective equations in the methods and to their parameters (e.g., by “standard deviation”, do you mean “stdev(WD)” in Eq. 5?). This would again allow a better understanding of the predictor definition.
L. 439-440: Do you mean: |prior - true| - |posterior - true|? Maybe you could provide an equation for this metric shown in Fig. 8 (could also be done in the figure caption).
Fig. 9: Maybe you could write the mean values of the error correlations into the respective panels. Are the differences in the mean error correlations of the four experiments significant? Is there a reason, why you have chosen a larger bin-size for the histograms in the upper panels than in the lower panels? Please also correct the caption: “…for four experiments”.
Fig. 10: Please explain the dashed line in the caption. How would these plots look like for the Adv_CO2 and Adv_CO2_CO experiments?
Discussion and conclusions:
L. 557-561: Can you see seasonal differences in the performance of CO to constrain fossil fuel CO2 emissions? I wonder whether CO can provide more or less additional constraint during winter (with heating and traffic emissions in Central Europe) compared to during summer (without heating emissions). Are there seasonal (winter vs. summer) differences in the posterior error correlation between biogenic and fossil fuel CO2?
L. 569-570: Please emphasize that the used in-situ observations are synthetic data.
L. 570ff: Do you have an explanation for why the fluxes in France are hardly altered, although there are some French observation sites? Are those sites perhaps less influenced by fossil emissions? More generally, could the low additional benefit of CO in the Adv_CO2_CO inversion partly be explained by the fact that some of the CO2&CO observation sites are less influenced by fossil emissions? In other words, could CO2 & CO observations from urban sites lead to a greater benefit from the Adv_CO2_CO experiment?
Technical corrections:
L. 159: Add a blank space in the reference “(… August 25, 2022)”
L. 194: “difficult uncertainty”
L. 199-200: “an uncertainty”
L. 210: Add a blank space in the reference “(… September 15, 2022)”
Table 3: Maybe you could slightly change the format of the title column in the table, because there is a lot of space between “CO:CO2” and “error”, which is why it looks as “error” would be a fifth column.
L. 446: “2000 kg s-1”
Supplement material:
L. 16 & L. 25: What is “AP”? Do you mean “AD”?
L. 89-90: Do you mean “simple uncertainty propagation method”?
Citation: https://doi.org/10.5194/egusphere-2023-2025-RC1
Data sets
Data and code related to manuscript 'Improved definition of prior uncertainties in CO2 and CO fossil fuel fluxes and the impact on a multi-species inversion with GEOS-Chem (v12.5)' I. Super, T. Scarpelli, A. Droste, and P. Palmer https://doi.org/10.5281/zenodo.10554686
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 190 | 64 | 14 | 268 | 28 | 10 | 7 |
- HTML: 190
- PDF: 64
- XML: 14
- Total: 268
- Supplement: 28
- BibTeX: 10
- EndNote: 7
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1