Review of: “A WRF-Chem study of the greenhouse gas column and in situ surface mole fractions observed at Xianghe, China. Part 2: Sensitivity of carbon dioxide (CO2) simulations to critical model parameters”, Callewaert et al., submitted to EGUsphere

General comment:

The manuscript presents an overview of the performance of the WRF-GHG against CO₂ observations collected for one year near Xianghe, China (within 100 km from Beijing). The authors evaluate the basic model performance and present a series of sensitity experiments in order to assess the impact of critical configuration settings. The paper is well-writen and mostly well-stuctured, and of high editorial quality.

The authors find that signals emitted from anthropogenic sources (primarily from industry and energy sectors, but with contriubutions from residential heating & transportation) dominate the local signal variability in both column and in situ model realizations of the atmosphere. The authors use the framework developed in the tandem paper and presented for CH4 there. Major additions here are the vertical emission structure (important for anthropogenic sources of CO2), and VPRM biosphere model. The evaluation of their performance is interesting. Seasonal biospheric signals (modelled using online VPRM module) were found to be playing a seasonal role, but the sensitivity experiments have shown a critical necessity for update of the VPRM model parameters, with the default settings not providing accurate predictions over China. This is also not surprising considering the incomplexity of the model.

I think the study is overall good report on the simulation results, but my primary concern is that it might be of relatively low interest. The findings do not go much beyond previously published papers (those given in the introduction, but also the companion paper) neihter in terms of methods, data used, nor the model techniques. The main innovation as compared to the companion paper is a new gas investigated over East China. The authors conclude that the study “demonstrates the value of the WRF-GHG model framework to interpret both temporal and sectoral variations in surface and column CO2 observations”, and further states that model accuracy is strongly dependent on appropriate configurations choices” – but this is known, and I personally feel that the contents might not warrant a full research paper in ACP.

I recommend for the study to be published only after major revision. This should include expansion of the paper material, especially in terms of more robust sources of error evaluation. The impact of the urban areas on the CO2 fluxes also deserves more attention, as we are analysing an area in the immediate vicinity of one of the largest cities in the world – Beijing. Application of an inversion framework, even a simple form, should be possible with the results available here, and could strenthen the scientific content substantially (see, for example “2.1. Bayesian Synthesis Inversion” in Peylin et al., 2002), allowing for the quantitative source error estimation.

Specific major comments:

Structure: section 3 is titled “Results”, but it contains a subsection “3.5.3. Discussion”. I assume, based on the text, that this is meant to be Section 4 “Discussion”, and I’ve put remarks assuming that this was the targeted structure. If this is not meant this way, Sec. 3 should be titled “Results and Discussion” – in which case I ask to adjust responses to my “move to Discussion comments” accordingly.

L120-121, tables 1, 6 & 8: (major) I suggest to restructure the naming of the sensitivity experiments. While reading the Results and Discussion sections, I kept being confused by the BASE simulation occurring in SFC and PROF flavour. As it is right now, the clarity would benefit if you had BASE, SFC, LC, PARAM as four simulations for sensitivity experiments. Tables 6 & 8 could then be merged, as the same simulation is referred to as PROF and BASE in these, with same numbers given.

L161: Division of data into afternoon and nighttime is questionable. The typical daytime that is considered in the CO2 analysis papers 12-15 or 11-15 LT, to assure developed boundary layer throughout the year; Xianghe is only at 40N south, with a distinct seasonal difference in daylight time. In winter, at 18 LT, the stable boundary layer might already be developing, with the surface measurements more fitting into the nighttime category already. It's much worse, however, with what the authors have selected as "nighttime"; in late December the sunrise is at 7.30 am, in summer 5 am is already daytime. As applied, the division mixes very different atmospheric states in terms of PBL dynamics, and the results of analyses are likely to be affected by diurnal cycle of PBLH rather than by, which doesn’t help in evaluation of “nighttime” model performance.

Other comments:

L39-L45: Please also provide a reference to Beck et al. here. Note: WRF-GHG is an evolution of WRF-VPRM used by Ahmadov et al. 2007 in the paper you cite. In L44 I suggest to add “(there referred to as WRF-VPRM)” next to Ahmadov et al. citations.

L75, and also L79-80: Measurement uncertainty over what timescale? A timescale comparable to instantenuous values of the model output should be provided, unless the model output was averaged over hourly time steps.

L88: How many vertical levels below 3km? What is the thickness of the lowest layer?

L93: Biomass burning emissions are highly variable in both space and time. How was this assured? What kind of errors are expected?

L99: What does 00:00 UTC correspond to in local time? Why was this time selected?

L110: Were results momentary or averaged in time? How does that compare to the observational data?

L112-113: How is the column above the model top handled in this study? Even if discussed in the Part 1, this should be mentioned here. This is important for comparing against (total column) TCCON measurements, with possibility of significant, and artificial, biases.

L160: How was the deseasonalization done?

Figure 2 (also concerns Figure 4):

1. Consider only keeping the bias corrected version, and move the uncorrected to the supplement. The difference between 2 and 4 is minor and cannot be distinguished with a naked eye.
2. I suggest to include all model data in panels a) and b) – it would be interesting to see the model behaviour for the gap periods, especially the July 2019 event, which could shed some light on the temporal extent of the event (since model performs well).

L166: Which correlation coefficient? Pearsons? R? R2? Be precise.

L175-176: “This bias likely originates from a similar error in the background data, inaccuracies…” – conclusion precedes the results. Remove here and move to discussion please.

L201: I disagree that MBE changes only “slightly”. That’s 30% of absolute bias, and 0.7 ppm can be significant.

L202-203: Negligible – please provide numbers (can be “smaller than XXX”) for reproducibility.

Figure 3: is this needed? Consider moving to supplement.

L219: “VPRM-computed fluxes suggest…” – this requires some comment. What kind of crops are grown there? Is such a net flux from croplands realistic? What about the urban-area fluxes?

L235: “… ppm before… after” – please state explicitly which periods were used as “before” and “after”

L239-240: “enhacement below / above the background” -> negative / positive enhancement

L242: “a less strong” -> weaker

L244 onwards: from here the text turns into Discussion and should be moved to appropriate section of the paper.

L247: “a warm air mass with relatively high biogenic CO2 levels,” this sounds as if there is a lot of biogenic signal from respiration, whereas in fact this is a lack of signal. Needs rewording.

L249 (and elsewhere in the text): “biogenic CO2 flux” – consider referring to it as “NEE” as in the original VPRM paper, or “net biogenic CO2 flux”. Especially the latter would improve clarity for the reader.

L261-275: For clarity of text, I suggest to give the absolute values only for observations and discuss the model over/underestimation together with model results presented as differences. Will some restructuring of the paragraph, but would make for easier reading.

L276-280: While entirely true and important for CO2 -- what is the relevance of this part here? Feels disconnected, while it should be part of broader PBL-biosphere dynamics discussion. Also - it's more discussion than results.

L284: “Applying these values to China likely introduces regional mismatches.” - Please discuss why. This is part of rationale for this sens. test.

L288: “In WRF-GHG…” – Lack of CO2 biogenic flux in those cells possibly has a major impact on low-level mole fractions in the vicinity of Beijing, strongly affecting model performance. How large are the areas within reasonable distance (100km) around the station that provide NEE fluxes of zero (classified in VPRM as urban). Land-use data from SYNMAP and Copernicus LCC are available to the authors. It could give some insight into the magnitude of the issue.

L298: “Here, the focus is instead…” - With the rectifier effect important for CO2, wouldn't it be wise to study PBL dynamics in detail? This is important -- because of this connection we cannot be sure about the fluxes unless we are sure of PBL. This requires – at the very least - discussion about the accuracy of the PBL in the current scheme, and potential errors that can stem from using the selected PBL scheme Some conclusions can perhaps be drawn based on available literature without the need to run extra experiments, but it is opinion of this reviewer that this topic has been treated too lightly here.

L301: See specific major comment 2.

L349: “In contrast, …” – Wording. This sentence also shows that the respiration is higher, so there is no contrast (?).

L387: “specific adjustment – such as (…) – can substantially improve model performance” – I’m not convinced that the change is substantial. In MBE for LC and PARAM yes, but other metrics don’t show that much of a change. And Table 6 the results show that MBE in most cases that actually deteriorates (!) when emission profiles are applied. It’s more grounded in data to say that “the model becomes more realistic in its setting”, but substantial improvement is only seen when PARAM simulation is compared to BASE, in the afternoon data (easiest in PBL). I do not really trust nighttime comparisons due to improper selection of hours for analysis – see comment above.

L418: “thereby reduced” - Wording: the importance was not "reduced", since it was always smaller for XCO2. Please rephrase.

L421-422: “Likely causes include…” – what about PBL bias in the model?

L424-426: “Adjustments to the VPRM vegetation parameters substantially affected near-surface mole fractions, underscoring the critical role of appropriate parameter selection—especially in the absence of a standardized VPRM configuration for China.” – please provide numbers here.

L430: “while emissions dominate” - anthropogenic or biospheric+anthropogenic? Please be precise.

References:

2002, “Influence of transport uncertainty on annual mean and seasonal inversions of atmospheric CO2 data”, Philippe Peylin, David Baker, Jorge Sarmiento, Philippe Ciais, Philippe Bousquet, Journal of Geophysical Research: Atmospheres, Volume 107, Issue D19 pp. ACH 5-1-ACH 5-25