Data Assimilation of Ambient Concentrations of Multiple Air Pollutants Using an Emission-Concentration Response Modeling Framework

Round 1

Reviewer 1 Report

This study proposed the response surface model (RSM) with the nudging/assimilation of ground observations for multi-pollutants to improve the WRF-CMAQ model performance on ozone and PM simulations.  This method was implemented over the North China Plain with the NO2, SO2, PM2.5, and O3 observations in 28 key cities for the simulations of four representative months in 2017. The RSM-assimilation framework improved the control PM and ozone simulations performance in terms of decreasing the overall model root mean square errors (RMSE) significantly. It also provided quantitively information about the uncertainties of emission inventory used for chemical transport models. This manuscript is locally written with results mostly clearly presented. I recommend the acceptance for publication with minor changes. I have some specific comments that want the authors to address.

1. Line 28. Define RMSE first in the Abstract.
2. Line 106-107. May use 1-2 sentences to summarize the modeling configurations for this study or provide them with supplemental material.
3. Line 110. I cannot find this reference [18]. The record in the citation section is not complete.
4. Line 130. Is the IP address reference for China National Environmental Monitoring Center proper format for journal publication?
5. Line 136-137. In total 85 sites across 28 cites, which is on average 3 sites per city. This seems inconsistent with the red dots in Figure 1. I understand for Beijing and Tianjin the available monitoring sites are much larger than other 26 cities, so the actual sites used in the RSM-assimilation framework are a subset of the sites shown in Figure 1?
6. Line 156-157. Just wonder what’s the technique difficulty for current RSM design for emission ratios adjustment for gaseous species greater than 2. Is it due to the non-linearity response from gaseous emission to concentration in the RSM? I would assume NH3 and VOC emissions have larger uncertainties in the emission inventory.
7. Line 163. Figure 3. For subplots (d), the scatter plots, are the black dots control case and red dots RSM-assimilated case? It is hard to match from the scatter plot for January with the shown statistics, such as correlation improvement from R2=0.3 to R2=0.6. Also, there are some cut-offs at the right edge of this figure, please fix it.
8. Line 186. Figure 4. Please help to clarify the plots for April 2017 on the PM25 simulation. After nudging, the underestimation of monthly mean PM25 concentrations at stations over Beijing becomes heavily overestimated (from green color with the concentration 30-40 ug/m3 jump to yellow and red color with concentration 70-90 ug/m3). Is it because the large emission ratio adjustment for pPM25 at Tianjin and the prevailing wind at this month is from southeasterly? Or you mentioned dust season, but the dust source should come from northwesterly.
9. Line 195-196. What’s the boundary conditions the WRF-CMAQ used outside the 9km modeling domain?
10. Line 215-217. Still how many sites in each of the 28 key cities were used for the nudging? For example, for the city TS from Figure 6, three random selections yield the almost same RMSE improvement.
11. Line 223-227. Is it possible to include the space observations (e.g. NOX, SOX, NH3, HCHO) into the RSM-assimilation framework to optimize the model emission accuracy in terms of the spatial gradient?
12. Line 283-285. Not mentioned elsewhere for the supplementary materials for Figure S1 and Figure S2, please build the connection in the main content to refer those two figures.
13. Line 347-349. Update the ACPD reference if possible.
14. Line 350. Fix reference #18.

Author Response

We thank the reviewer for the detailed and thoughtful review of our manuscript entitled “­Data assimilation of ambient concentrations of multiple air pollutants using an emission-concentration response modeling framework”. Incorporation of the reviewer's suggestion has led to a much improved manuscript. Please see the attachment that is our response to the issues raised by the reviewer. We also detail the specific changes incorporated in the revised manuscript in response to the reviewer’s comments.

Author Response File: Author Response.docx

Reviewer 2 Report

The authors presented RSM-based data assimilation over China. The topic of data assimilation is powerful tool to fill the gap between model and observation, and such technique is attractive for researchers. Under the current manuscript, I fully cannot catch up two following points. Please address these two concerns.

 

RSM-assimilation framework:

As the first step of data assimilation, it will be in general to constrain the precursors. However, this technique constrained NH3 in the final step as well as PM2.5, and in addition, NH3 is not directly constrained by observation. Why the authors use such flows for multi-air pollutants?

 

Emission adjustment:

Compared to other four species, pPM2.5 is adjusted even over two during polluted days. I am suspicious these results for RSM-assimilation. Generally, CMAQ tended to underestimate PM2.5 concentrations, and therefore, it seems like pPM2.5 is forced to be adjusted to meet observed PM2.5 concentration. In terms of RMSE, this score is improved but just superficial. To validate these results, the evaluation of PM2.5 components is necessary.

Author Response

We thank the reviewer for the detailed and thoughtful review of our manuscript entitled “­Data assimilation of ambient concentrations of multiple air pollutants using an emission-concentration response modeling framework”. Incorporation of the reviewer's suggestion has led to a much improved manuscript. Please see the attachment that is our response to the issues raised by the reviewer. We also detail the specific changes incorporated in the revised manuscript in response to the reviewer’s comments.

Author Response File: Author Response.docx

Reviewer 3 Report

The authors describe an interesting and possibly valuable manner to improve air quality concentration maps through data assimilation of multiple air quality pollutants at once. Several additions seem however necessary to clarify remaining questions readers will have after reading the publication:

 

1. The authors state and discuss the added value of simultaneous data assimilation of both O₃ and PM₅ by optimizations through scaling of the emissions used by a CTM model. I would advise a comparison with the results of data assimilation of a single pollutant at once to quantify the added value.
2. The study looks at 85 stations in urbanized areas of China. The CTM used has a grid resolution of 9 km². As the grid resolution is not all sufficient to explain the air quality gradients within cities, this methodology is expected to work for urban background stations. A discussion of the type of stations used and to what degree their pollution levels are explained by very local emissions is a valuable addition.
3. It is not 100% clear from the publication how the emission adjustment ratio is applied and optimized. Reference is given to a previous paper, I would find it however valuable to add a more thorough discussion to this paper to make it more readable as ‘stand-alone-document’.
4. I lack information on the validation aspects of this study. Is the validation done by leaving the station for which the validation statistics are calculated out of the set-up (leaving-one-out-validation) or represents the validation the data-assimilation including the station itself? I believe the station should be left out, as the statistics are otherways not representative for locations where air pollution is not monitored, which is the final goal of better air quality maps.
5. I would like to ask the author to include that type of emissions used in the CMAQ model. Which type of emissions and emission sectors are used? What spatial variation do the emissions have? Volume sources, line source, point sources, surface sources etc.?
6. Given the goal to have better air pollution information, I would request a discussion on the model uncertainty and how the data assimilation technique can reduce the model uncertainty.

Author Response

We thank the reviewer for the detailed and thoughtful review of our manuscript entitled “­Data assimilation of ambient concentrations of multiple air pollutants using an emission-concentration response modeling framework”. Incorporation of the reviewer's suggestion has led to a much improved manuscript. Please see the attachment that is our response to the issues raised by the reviewer. We also detail the specific changes incorporated in the revised manuscript in response to the reviewer’s comments.

Author Response File: Author Response.docx

Reviewer 4 Report

Dear authors,

this article presents a significant interest. The discussion needs modification.In addition, in my opinion you should add some references . Perhaps it will be very helpful to have an annex with all the abbreviations. In this article you use a lot of abbreviations from models and etc., this is the reason why i believe than a table in annex will be helpful.

Best regards

 

Line 22: The word “which” should be change with the word “that”.

Line 27: The word species should be change with components or pollutants. The word species, usually is used only for living organisms.

Line 28: In line 28 is used an unknown term, RMSE. You should explain the abbreviation.

Line 29: In my opinion in this sentence the word exhibits isn’t the appropriate. Perhaps the word results is more appropriate.

Line 35: The expression “wind – blown dust” may be changed.

Line 36, 206: The word enlarging should be change with the word expanding.

The word “linkage” is used for instance in lines 58 and 59. I can not understand the meaning, what you want to say.

Line 64: Perhaps you can add the word “the”.

Line 78-79: Therefore, a lack ….PM2.5 is observed.

In line 88 I think that you may change “are” with “is”.

Line 95: It is better to avoid “we” in scientific articles.

Line 128: The abbreviation NMB may be explained. Line 152: The word “input” is usually a noun. Perhaps the word introduced is better.Line 156: Normally isn’t used the expression gaseous species. The word gases is usually used.Line 184-185: “ after assimilation” may change as “after the four moth assimilation.Line 199: The expression “gets slightly worse” is very informal.Line 201: Perhaps you can change the word “with” as “compared to”“the” with “a”Line 217: Instead of “degrade” word decrease is more appropriate.Line 220: The expression “at average of 28 cities” isn’t very clearly. I can not understand the meaning.Line 235: “for monthly averages”, for average month valuesLine 247: The word during may be more appropriate instead of “across”Line 252: “also likely corrects for” – is also likely to correct other model..Line 279: The plural isn’t appropriate. In my opinion the appropriate word is impact not impacts.

 

 

 

 

 

 

 

Comments for author File: Comments.docx

Author Response

We thank the reviewer for the detailed and thoughtful review of our manuscript entitled “­Data assimilation of ambient concentrations of multiple air pollutants using an emission-concentration response modeling framework”. Incorporation of the reviewer's suggestion has led to a much improved manuscript. Please see the attachment that is our response to the issues raised by the reviewer. We also detail the specific changes incorporated in the revised manuscript in response to the reviewer’s comments.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

I  appreciate authors' reply  to address my concerns; however, one point still solve my concern.

[RSM-assimilation framework] For NH3, I can understand that the lack of NH3 direct observation.

[Emission adjustment] In my best knowledge, CMAQ v5.2.1 used in this study has thee inline dust scheme. Dust in spring season over China is so familiar, but why authors did not include the inline dust scheme in this study? If the authors know that the dust have the possibility to compensate the adjustment on pPM2.5, it is worthwhile to do the simulation including Asian dust. In addition, the authors explained the missing chemical pathway from S(IV) to S(VI). I have question to this point. The original emission is taken from MEIC 2016, but the simulation year is 2017. I recognize that SO2 emissions have been dramatically decreasing. The difference of emission used in this study and the simulation year could cause the discrepancy between simulated results and observed facts. Could you clearly explain this point?

Author Response

[Comment]: I appreciate authors' reply  to address my concerns; however, one point still solve my concern.

[Response]: We thank the reviewer for reviewing the revised manuscript. We have followed the additional comment and revised manuscript accordingly.

 

[Comment]: [RSM-assimilation framework] For NH3, I can understand that the lack of NH3 direct observation.

[Response]: It is true that the lack of NH3 direct observation is troubling. The NH3 satellite products can be a very useful option to further improve the assimilation method.

 

[Comment]: [Emission adjustment] In my best knowledge, CMAQ v5.2.1 used in this study has the inline dust scheme. Dust in spring season over China is so familiar, but why authors did not include the inline dust scheme in this study? If the authors know that the dust have the possibility to compensate the adjustment on pPM2.5, it is worthwhile to do the simulation including Asian dust.

[Response]:

For the dust scheme, it is true that the inline dust scheme is available in the CMAQ v5.2.1 and we used that for the outer domain simulation (27km) to provide the boundary condition for the nesting domain of NCP. However, we didn’t turn it on for the nesting simulation simply because the wind-blown dust within the NCP is limited, and most of the dust comes from outside of NCP. Also the RSM model requires multiple CMAQ simulations, and the in-line dust model was turn off to increase the computational efficiency.

Meanwhile, our previous study found that the uncertainties of the in-line dust model in CMAQ tends to result in significant underestimation of the dust emission when comparing against with the observed AOD over dust region (Xing et al., 2015). There are some updates in the CMAQv5.2 but still suffering underestimations. As the reviewer mentioned, further improvement of the dust model is necessary to compensate the adjustment on pPM2.5.

We further clarified this point in the revised manuscript, as follows.

(Line 129) “The inline dust model was used to estimate the wind-blown dust emissions in the 27km × 27km China domain. Considering the wind-blown dust mostly comes from outside of NCP, we turned off the inline dust model for the simulation of 9km × 9km over NCP to reduce the computational burden associated with the RSM model development.”

(Line 326) “The uncertainties of pPM_2.5 emissions may be due to the lack of inline dust simulation over NCP domain and underestimation of wind-blown dust emissions outside of NCP [31], as such underestimation of PM_2.5 is more pronounced in April during the dust season.”

Reference:

Xing, J., Mathur, R., Pleim, J., Hogrefe, C., Gan, C.-M., Wong, D. C., and Wei, C.: Can a coupled meteorology–chemistry model reproduce the historical trend in aerosol direct radiative effects over the Northern Hemisphere?, Atmos. Chem. Phys., 15, 9997–10018, 2015.

 

[Comment]: In addition, the authors explained the missing chemical pathway from S(IV) to S(VI). I have question to this point. The original emission is taken from MEIC 2016, but the simulation year is 2017. I recognize that SO2 emissions have been dramatically decreasing. The difference of emission used in this study and the simulation year could cause the discrepancy between simulated results and observed facts. Could you clearly explain this point?

[Response]:

It is true that the emission used in this study is MEIC 2016 while our simulation year is 2017. We agree with the reviewer that the difference of emission used in this study and the simulation year could cause the discrepancy between simulated results and observed facts since the SO2 emission has been continually controlled recently. Our conclusion that the missing chemical pathway from S(IV) to S(VI) might contribute to the low biases of PM2.5 is simply because that the SO₂ concentration was overestimated (see the Figure 7, the adjusted emission ratio of SO₂ is smaller than 1 indicating the overestimation of SO₂ emissions) while the SO₄ concentration is underestimated (as seen in Figure S5, the simulated percentage of SO₄ in total PM_2.5 is smaller than observed percentage in January and July). Such biases are also suggested in previous modeling studies (Fu et al., 2016; Zhang et al., 2019).

We have clarified this point in the revised manuscript as follows.

(Line 336) “This is also suggested from the evaluation of PM_2.5 component concentrations which imply that potential missing chemical reaction pathways for the transition of S(IV) to S(VI) [32, 33] might contribute to the low-biases of sulfate aerosols in January and July (i.e., SO₂ concentration was overestimated but the percentage of sulfate aerosols in total PM_2.5 was underestimated, see Figure S5).”

 

References:

Fu, X., Wang, S., Chang, X. et al. Modeling analysis of secondary inorganic aerosols over China: pollution characteristics, and meteorological and dust impacts. Sci Rep 6, 35992, 2016.

Zhang, S., Xing, J., Sarwar, G. et al. Parameterization of heterogeneous reaction of SO2 to sulfate on dust with coexistence of NH3 and NO2 under different humidity conditions. Atmospheric Environment, 208, 133-140, 2019.

Reviewer 3 Report

The authors have added several aspects to the paper which improve the quality and make it a more comprehensive publication.

A single comment is however not yet addressed. In a CTM like CMAQ, several types of emissions have been included. From the description of the methodology, I understand that all emissions in a single grid cell of the model (9x9 km²) are adjusted with a uniform correction factor.

Have the authors considered a separate correction for the different emission categories (residential, industrial, agricultural etc.)? Would this be a possible further improvement to the methodology or would this introduce too many degrees of freedom?

Please add a short motivation of the choice made.

Author Response

[Comment]: The authors have added several aspects to the paper which improve the quality and make it a more comprehensive publication.

[Response]: We thank the reviewer for reviewing the revised manuscript. We have followed the additional comment and revised manuscript accordingly.

 

[Comment]: A single comment is however not yet addressed. In a CTM like CMAQ, several types of emissions have been included. From the description of the methodology, I understand that all emissions in a single grid cell of the model (9x9 km²) are adjusted with a uniform correction factor. Have the authors considered a separate correction for the different emission categories (residential, industrial, agricultural etc.)? Would this be a possible further improvement to the methodology or would this introduce too many degrees of freedom? Please add a short motivation of the choice made.

[Response]: We thank the reviewer for this good suggestion. It is true that in this study all emissions in a single grid cell of the model (9x9 km²) are adjusted with a uniform correction factor since all emissions has been combined into one value in CTM during the simulation. We agree with the reviewer that the adjustment for each emission category will be more helpful to further improve the emission inventory. The challenge is that to separate the emissions by sector will increase the degree of freedom, which need additional correction functions such as the spatial pattern of each emission category based on observations like satellites.

As the reviewer suggested, we have clarified this point in the revised manuscript as follows.

(Line 368) “Further improvement can be also done to separate the emissions by sectors with additional correction functions such as the spatial pattern of each emission category based on observations like satellites.”