the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 25 Mar 2025
XCO2 observations compared to km-scale ICON-ART simulations indicate an underestimation of Thessaloniki’s emissions in the ODIAC inventory
Abstract. An accurate inventory of CO2 emissions is important for the implementation of effective reduction measures and thus for climate change mitigation. Most current inventories are based on reported activities and rely little or not at all on atmospheric data. However, these inventories have large uncertainties, especially for smaller scales such as urban areas. For example, for the city of Thessaloniki, Greece, the EDGAR inventory reports 3.1 Mt, which differs by 72 % from the emission estimate of the ODIAC inventory (1.8 Mt) for the same area for the year 2019. With a measurement campaign in the framework of the Collaborative Carbon Column Observing Network (COCCON), we collected observations for three months in October 2021 and summer 2022 in Thessaloniki. A total of 30 days of column averaged molar fractions of CO2 (XCO2) were recorded. We combine these data with km-scale simulations from the numerical weather prediction model ICON-ART. The ODIAC inventory was used for simulating the emission of CO2. We optimized the simulated atmospheric time series of XCO2 to best match the observed data by scaling the prior emissions using a least-squares approach. With different configurations, we found a consistent up-scaling of the prior emissions, with total emissions ranging from 2.9 to 4.4 Mt in the urban area of Thessaloniki. This estimate is significantly higher than the emissions reported in ODIAC. The result demonstrates the potential of including ground-based column measurements of CO2 in the construction of emission estimates to reduce uncertainties at the urban scale.
- Preprint
(4361 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-639', Anonymous Referee #1, 29 May 2025
The study of Feld et al. uses measurements of total column CO2 (XCO2) from a combination of two ground-based EM27/SUN spectrometers placed in and around the city of Thessaloniki, Greece, to estimate the emissions of the city. Emissions are estimated by simulating the total columns with a mesoscale atmospheric transport model (ICON-ART) driven by inputs from a high-resolution emissions inventory and optimally scaling the simulated columns (and hence the emissions) to the observations.
The task is challenged by multiple factors: The column enhancements due to emissions from the city are relatively small, differences in the enhancements between the two sites are small compared to measurements uncertainty, the data set is limited to 30 days of observations, and the meteorology in Thessaloniki is complex due to land-sea breezes and topography.
The study builds on previous studies using EM27/SUN spectrometers to quantify urban emissions. The main scientific contribution of this study is to demonstrate the utility of such spectrometers to estimate emissions from a mid-sized city in a complex environment for which, as the authors argue, high-quality atmospheric transport simulations are necessary rather than simple box-model approaches.
Although I agree with this last point and the model seems quite capable in reproducing the meteorology during the study period, the authors did not convince me that the chosen approach is appropriate and the collected observations sufficient to quantify the emissions of the city. The study requires substantial revisions before it can be considered for publication. More details are provided in the following.
Main issue
The study makes the strong assumption that the background does not change over a given day. All variability during the day is thus assumed to be due to emissions and biospheric fluxes within the model domain. For each day, a background is subtracted from the observations and only the enhancements above this background are compared with the model. Correspondingly, the model only simulates the signals from CO2 fluxes within the model domain, but no background. Treating the model in this way makes sense under the assumption of a daily constant background, but whether this assumption is justified in the first place is highly questionable. The publication provides no convincing evidence that this assumption holds. If this was not the case, any variability in the background would be translated into emission signals, which could substantially affect the results. Unfortunately, there were no measurements upstream of the city to test the hypothesis. I fully acknowledge that measuring upstream is complicated in a coastal city, but nevertheless it seems to me that it would have been possible to place the mobile instruments outside in one of the agricultural areas to the south or across the bay to the southwest of the city. It is unfortunate that the 2nd instrument was always placed inside the city.
There are several factors that may lead to a changing background: Advection of air masses differently influenced by vegetation (note that most of the measurements were taken during the growing season) or anthropogenic emissions, changes in air mass origin during the day, or changing influences of the land-sea breeze evolving over the day.
The authors acknowledge the influence of land-sea breezes and the complex topography, but there is no deeper analysis of their potential impact on XCO2. The wind rose at Thermi (Fig. 8) suggests that there was indeed a prominent land-sea breeze during most of the measurement days, potentially amplified by the hills to the northeast of the city. At low altitude, the sea breeze transports air from sea to land, but at higher altitude potentially from land to sea. There could thus be a recirculation of CO2 from Thessaloniki at higher levels. Was the model domain large enough to capture such effects? The full model domain had a width of 2° east-west, but all figures only show a subdomain. It would have been helpful to see a figure presenting the full model domain and its topography to better judge potential influences of land-sea contrasts and topography.
On many days, there was a marked decrease of XCO2 over the course of the day (see Fig. 9). There is no discussion of this important phenomenon at all. Was this due to an increase in wind speeds or due to the take-up of CO2 by the vegetation over the day?
In my view, a more careful analysis of the mesoscale flow and its impact on CO2 is needed. The analysis presented in Section 3.2 is extremely limited and doesn't provide any insights into the overall flow situation. Furthermore, a more thorough assessment of the hypothesis of a daily constant background is needed, for example by looking at CAMS CO2 global reanalysis data upstream of the city. I am also missing an analysis of the contribution of biospheric CO2 to the signals shown in Fig. 9. Wasn't this simulated as a separate tracer? Interpreting the differences between model and observations only in terms of emissions from the city but ignoring the influence of potentially wrong assumptions regarding background and biospheric fluxes is very risky, in my view too risky.
Another issue is that there is no discussion of the strategy of how the station pairs were selected on the different days. Typically, such studies use an upwind-downwind configuration (or at least a positioning along the main wind direction) to quantify the emissions based on the difference in XCO2 levels between the sites. However, it seems that in this study the selection of the mobile site was not guided by actual wind conditions but rather by the idea of covering all parts of the city over the course of the campaign. This approach needs to be better motivated. In fact, the study makes very little use of the fact that there was a pair of sites. Rather each site contributes to the results independently. The chosen approach would thus be equally valid (or invalid) when applied to a single site, e.g. the stationary site that has been measuring over a much longer period.
Another inversion study for a coastal city using total column observations (but not cited) was performed in Los Angeles by Hedelius et al. (https://doi.org/10.5194/acp-18-16271-2018). In their study, they used a TCCON site outside Los Angeles as background and analyzed only the difference to the TCCON site within the city. This seems to me to be a much more robust approach.
Further important points
- Several data sets used in the study are not described in sufficient detail. These include the CEDS data set mentioned on page 2 and the MODIS, SMAP and FLUXCOM-X-BASE data set mentioned on pages 7/8. Without any further details, their role and utility for the study is difficult to judge.
- CAMS has produced the CAMS-REG emission inventory (Kuenen et al., 2024; https://doi.org/10.5194/essd-14-491-2022) that has higher resolution than the CAMS-inventory used in this study. It would make much more sense to use that inventory.
- Figure 3 summarizes the measurements at the two sites. Considering the importance of the gradients between the site pairs, it would be useful to add another panel, in which the median of the instrument at Campus is subtracted from each pair. All box plots for the Campus instrument would thus have a median of 0. Furthermore, I was strongly confused by the labels on top of this figure. As described in the text, one of the instruments remained at the same location (at Campus), but the figure shows two different labels for this instrument, "P" and "M". I also found it confusing that the same site was named differently in different contexts. The campus site, for example, is sometimes called Campus, sometimes P (Physics), and sometimes "A". Why not use the same name throughout? I don't see the point of using a label "A", for example. Writing Campus instead of "A" would not make the publication much longer.
- The tables need to be improved:
Table 1 has four columns but only two titles. "Vaisala" or "Davis" seem to be instruments rather than sites. The site name should come first, then the instrument. Why is there no entry for the accuracy of the pressure measurement at C and D? It would also be good to add a column "wind" and show which sites have wind measurement (e.g. with yes/no or an "X").
Table 2: There should be separate columns for site and instruments. It doesn't make sense to have a label "B" referring to a site location in a column called "instrument". It would also be good to add a column for the units. When comparing model results with observations, it is more common practice to express the bias in terms of "SIM – OBS" rather than "OBS – SIM".
Table 4: The word "Instrument" should be centrally placed above the instrument columns (SN52 etc.), not to the left. - The simulation setup looks far from ideal to me: Running a simulation for each measurement day separately is a valid option, but starting these simulations only at 3 UTC, i.e. shortly before sunrise, doesn't seem appropriate to me. The XCO2 columns shown in Fig. 9 often show a marked decrease over the course of the day starting from a high value in the morning. These high values could be due to CO2 buildup over the night (e.g. due to vegetation respiration fluxes in the domain), which the model is not able to capture when initialized only at 3 UTC. Probably each simulation would have to be run over 2 days with only the 2nd day being used for the analysis.
- The reverse analysis of the potential influence of distant sources on page 11 is awkward. The influence of sources outside the city should be directly available from the simulation, unless no anthropogenic sources outside the Thessaloniki were included. Furthermore, the short simulation time with initialization at 3 UTC may not be sufficient to see impacts at 70 km distance. Finally, the flow is quite strongly directed due to the land-sea breeze and the mountain circulations. A reverse view in such a situation makes little sense.
- The motivation for the measurements at a small distance of 500 m is not very clear. Of course, with increasing distance it can be expected that difference in XCO2 increase. What is the conclusion of the fact that these measurements show a slightly larger scatter than the side-by-side measurements?
Small points and corrections
Line 43: change "with the total emissions" to "and the total emissions"
L45: "nearly double" would be more precise
L64: Use past tense when citing publications: "Viatte et al. (2017) used ..". L65: "They also derived"
L79: Change to "Here, we apply"
L88: "... spatial gradients were observed …". This gives the wrong message, because the paper doesn't make any use of spatial gradients between sites.
L91: Change to "school of the physics building on the campus .."
L94: Change to past tense: "This allowed us ..".
L105: Do stations C and D also provide pressure? The description of the meteorological data is confusing.
L118: Use past tense: "was 2.03 ppm" and "amounted to only 0.17 ppm".
L120: Change to "… observed gradients of up to .."
L124: Change to "the daily variations measured in the dataset were"
L125: Again, use past tense to describe the conditions or actions during the campaign: ".. were changing too much .."
L125: Better "using a simple model approach such as a box model .."
L126: Better ".. is required to interpret the observations"
L131: Past tense: "setup was similar". Use "finer" instead of "denser" simulation grid.
L137: What do these number of degrees correspond to in terms of kilometers?
L142: This is an odd justification for treating CO2 as a passive tracer. I would simply say that this was justified by the fact that CO2 is not chemically reactive within the troposphere.
L146-147: Past tense: "a re-mapping was needed", "grid structure was re-mapped".
L147: The emiproc tool seems to have been published by Constantin et al. (2025) https://joss.theoj.org/papers/10.21105/joss.07509
L156: Singular: "mole fraction" .. "is not directly calculated"
L169: XCO2 is simply the ratio between VC_CO2 (eq. 4) and VC_dry_air (eq. 5)
L170: Change to "toward the sun"
L188: Change "in 10 min bins" to "to 10 min bins"
L193: It is unclear here and later whether Q(5) was computed for each spectrometer separately or whether it was computed from the combined time series of the two.
Section 2.4: Only here we learn that many different tracers were simulated each representing the emissions from different grid cells in the city. This would better fit into Section 2.2. describing the simulation setup.
L202: Here and elsewhere: The factors are scaling factors not weighting or "re-weighting" factors. Weighting is not the same as a scaling.
L203: It is sufficient to say ".. can be found by minimizing the cost function".
L205: "the the". Better ".. the corresponding vector of contributions from all pixels i"
L210: Better " .. not rely on assumptions of the prior uncertainty"
General comment on pixel scaling method: It needs to be made clear that a single scaling was computed from all observations on all days rather than a scaling was computed for each day separately (at least that's what I assume).
L214: past tense: "remained unchanged"
Caption Fig. 6: Change "They latter" to "The latter". The caption includes some discussion/conclusions from the figure that do not belong into a figure caption but in the main text.
Table 2: Change "The unit of the three rightmost columns is" to "The units of the three rightmost columns are"
L244-247: Why is the comparison with MUSICA IASI and TROPOMI data relevant here? This needs to be better explained. Are these particularly reliable instruments? What do you mean by "agreement between the observations"? Between which observations?
L249: "impressively"?
L251: Better: "although only a few grid cells apart, the wind roses .."
L257: What do you mean by "the observed bias"? A bias in the observations?
L258: The analysis of how well the differences in XH2O between the two sites are captured is interesting. Why is there no similar analysis for XCO2? Why not simply optimizing the differences in XCO2 between the two sites instead of optimizing the absolute enhancements? This approach would be much less dependent of variations in the background (and actually has already been applied in CO2 inversions in Paris).
L283: More elegant: "… varies between individual days".
L286: It is unclear how the correlation was computed. As the average of the correlations for the two spectrometers?
L300: Change to "This is remarkable, because XH2O is not .."
L306, 314, 315: These are scaling factors not "weights".
L312: Change to "It is still a remarkable fact"
L343: Could "lead to discrepancies"
Citation: https://doi.org/10.5194/egusphere-2025-639-RC1 -
RC2: 'Comment on egusphere-2025-639', Anonymous Referee #2, 15 Jun 2025
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-639/egusphere-2025-639-RC2-supplement.pdf
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 462 | 57 | 20 | 539 | 12 | 27 |
- HTML: 462
- PDF: 57
- XML: 20
- Total: 539
- BibTeX: 12
- EndNote: 27
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1