Time-Dependent Downscaling of PM2.5 Predictions from CAMS Air Quality Models to Urban Monitoring Sites in Budapest
Round 1
Reviewer 1 Report
General and major comment
I find the work presented by the authors is interesting. However, the authors need to have a discussion section. A discussion section to compare the results of this study with similar methods of downscaling to predict PM2.5 on the local scale is useful for the readers. I find this 'downscaling' method is related to assimilation method of observed data. The fusion method as presented here could be qualified as an assimilation method. Comparison of this method with others such as An artificial neural network (ANN) in the discussion section would be helpful.
Suggestion and minor comments on the manuscript is included in the attached Pdf file.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Overall this is an interesting manuscript looking at multiple different air quality models utilizing data from the CAMS research. While this work has the potential to impact other studies and future investigations it is currently not fully articulated by the authors. The primary concern with this work is the very limited interpretation of results, discussion, and comparison to other research.
Introduction
PM2.5 should be defined the first time it is used (line 37) also standard notation is to subscript the 2.5 which does not appear to be done throughout the manuscript.
Providing additional information on the need and background for downscaling prediction models would be useful.
Methods
Additional context regarding the sampling locations/sampling period is needed. A section indicating the time period that the samples were collected would be helpful.
Figure 1: overlaying topography on this map should be considered
The information on the end of page 4 and top of page 5 may be more reader friendly in a table.
Results
The legends should provide additional details to make the figures understandable without looking into too much detail in the text (i.e. in figure 5 the red line is not defined in the observed concentration – it appears to be the mean concentration across the time period but this should be clearly stated)
Figure 4: were p-values calculated for any of these? (particularly correlation coefficients)
Figure 5: what is the x-axis for the time series fitted model weights? The same as the observed concentrations? If wo additional discussion is needed related to how these models match the observed concentrations both in trends as well as statistical analysis.
There is a lack of discussion for the results. Comparisons to previous literature in other regions should be drawn upon to compare. As well as interpretation of the results.
Conclusions
What applications can this study have to other research?
Author Response
Dear Professor,
Thank you for sending us the very helpful comments on our manuscript entitled „Time-dependent downscaling of PM2.5 predictions from CAMS air quality models to urban monitoring sites in Budapest”. In the following, we provide our point-by-point answers to the specific queries. For clarity, the Reviewer’s comments and questions are in italic.
Overall this is an interesting manuscript looking at multiple different air quality models utilizing data from the CAMS research. While this work has the potential to impact other studies and future investigations it is currently not fully articulated by the authors. The primary concern with this work is the very limited interpretation of results, discussion, and comparison to other research.
Thank you for your valuable comment. A Discussion section has been added to the manuscript to present the relationship with other methods and to compare the results with some similar studies.
PM2.5 should be defined the first time it is used (line 37) also standard notation is to subscript the 2.5 which does not appear to be done throughout the manuscript.
PM2.5 has been defined as “particulate matter smaller than 2.5 µm diameter” and subscript notation has been implemented throughout the manuscript.
Providing additional information on the need and background for downscaling prediction models would be useful.
Verification of European CAMS measurements in the winter 2018-2019 found a large negative bias of PM2.5 predictions with mean relative biases between -0.22 and -0.02. This provided motivation for downscaling with the purpose that reducing systhematic underestimation could improve the relatively large hourly RMSE values of 8–11 µg/m3.
The above paragraph has been added to the manuscript.
Additional context regarding the sampling locations/sampling period is needed. A section indicating the time period that the samples were collected would be helpful.
A paragraph describing the sampling locations and the sampling period has been added to the text.
Figure 1: overlaying topography on this map should be considered
Figure 1 has been improved with topography contours.
The information on the end of page 4 and top of page 5 may be more reader friendly in a table.
A table has been added to present selected temporal subsets.
The legends should provide additional details to make the figures understandable without looking into too much detail in the text (i.e. in figure 5 the red line is not defined in the observed concentration – it appears to be the mean concentration across the time period but this should be clearly stated)
Legends were improved in Figures 2 and 3 and a new legend was added to Figure 5.
Figure 4: were p-values calculated for any of these? (particularly correlation coefficients)
The p values of correlation coefficients were very low (less than 0.001) in all cases due to a large number of observation-prediction pairs in the hourly time series. However, due to the strong autoregression of consecutive measurements, the p values are not interpretable in terms of significance in this case.
Figure 5: what is the x-axis for the time series fitted model weights? The same as the observed concentrations? If so additional discussion is needed related to how these models match the observed concentrations both in trends as well as statistical analysis.
All graphs in Figure 5 share the same horizontal axis. This is now explicitly stated in the caption. Statistical analysis of each model is presented in Figure 4.
There is a lack of discussion for the results. Comparisons to previous literature in other regions should be drawn upon to compare. As well as interpretation of the results.
A Discussion section has been added to the manuscript.
What applications can this study have to other research?
CAMS models are widely used in Europe and the experiences presented here might be beneficial for applications in other European cities. Furthermore, an updated smog alert regulation is under ongoing debate in the country and enhancing model prediction performance might trigger the incorporation of CAMS air quality predictions into regulatory practice.
This completes our reply. We thank the Referee for your comments, suggestions and critical reading of our work. We hope that the manuscript is now in an acceptable form.
Round 2
Reviewer 2 Report
All of my comments were adequately addressed, thank you.
