Application of a Common Methodology to Select in Situ CO₂ Observations Representative of the Atmospheric Background to an Italian Collaborative Network

Round 1

Reviewer 1 Report

In this manuscript, the authors proposed a method to screen out CO2 measurement samples so that screened CO2 data can be more suitable for background concentration estimation for Italy. The crux of the manuscript, from my point of view, is the development of BaDSfit based on the BaDS procedure.  Overall, the authors focused on the aim of the study fairly well. The manuscript is easy to read and well-written in terms of English. However, I have two major concerns about this manuscript.

1. It is not clear what BaDSfit itself is and how it is considered as a stand-alone method that can be applied in other places. It will be helpful if the authors make good efforts to show the proposed algorithm in a diagram and/or equations. In addition, if it is only applicable to Italy, I don’t think the method is not more than just a data fitting method for a specific area. 
2. There are a lot of existing methods to discern signal from noise. Also, there are a lot of methods to separate signals for each periodic cycle (e.g., annual, monthly, daily, etc.). The authors should discuss why those methods were not compared with BaDSfit. The author may argue that BaDSfit is really a preconditioning algorithm or procedure. If so, in my opinion, the authors need to clearly demonstrate BaDSfit is really independent of the subsequent signal processing, vice versa. It may go back to my concern 1 above. Speaking of which, why did not the authors compare results with BaDS even?

My other comments are as follows:

• The legend should be explained clearly. For example, I am not sure what “CO₂_camm_detrend” is although I can infer. However, I recommend that the authors clearly describe each legend item in the caption.
• All references with URLs should be properly cited following the reference style of Atmosphere.
• The introduction can be shortened by about 40%. For example, Lines 92-112 is a mixed body of texts among Introduction and Method.
• There are unnecessary blank lines and typos such as a double-quote symbol at the end of Line 314. The authors may want to ensure there is no typos.

Author Response

Response to Reviewer 1 Comments

Point 1: It is not clear what BaDSfit itself is and how it is considered as a stand-alone method that can be applied in other places. It will be helpful if the authors make good efforts to show the proposed algorithm in a diagram and/or equations. In addition, if it is only applicable to Italy, I don’t think the method is not more than just a data fitting method for a specific area.

Response 1: We really appreciate the time and efforts dedicated in reviewing this manuscript. In the revised version of the manuscript, we better explained that BaDSfit is not different from the original BaDS. The novelty of our work is related to the following points:

• BaDS was used to different measurement sites in respect to PRS
• These measurement sites are very different in term of environmental conditions in respect to PRS
• We implemented a more objective methodology to define the setting parameter n which largely determine the selection results

For these reasons, we decided to keep the name BaDS recognizing that the selection algorithm was not changed in respect to Apadula et al. (2020), actually. The methodology is detailed in the Section 2.2 of the revised manuscript. Obviously, BaDS is not only applicable to Italy. As demonstrated by this work, after an accurate tuning, it is effective in selecting background data at measurement sites very different from that for which it was originally developed (PRS). This point was better highlighted in the Discussion section (line 641): “The BaDS capability to identify the CO₂ measurement period representative of the atmospheric background in different environmental conditions (from the high mountain to a coastal site influenced by anthropogenic emissions, and to a remote marine site) may allow further studies concerning the long-term CO₂ variability by using a larger number of measurement sites, even outside the Italian territory.”

Point 2: There are a lot of existing methods to discern signal from noise. Also, there are a lot of methods to separate signals for each periodic cycle (e.g., annual, monthly, daily, etc.). The authors should discuss why those methods were not compared with BaDSfit. The author may argue that BaDSfit is really a preconditioning algorithm or procedure. If so, in my opinion, the authors need to clearly demonstrate BaDSfit is really independent of the subsequent signal processing, vice versa. It may go back to my concern 1 above. Speaking of which, why did not the authors compare results with BaDS even?

Response 2: The comparison of BaDS with other data filtering methodologies is a very interesting topic. Unfortunately, this is well outside the scope of this manuscript which aim is to assess the effectiveness of using BaDS at other measurement sites in respect to PRS (for which it was originally developed). We already shared BaDS with other colleagues who are working on this topic, and we are confident that a paper comparing different data selection methodology for filtering the background signal from in-situ CO₂ time series will came out soon. However, in the revised version of the manuscript (“Discussion” section), we provided the comparison between the results of the original BaDS and our optimized tuning for PRS, to evaluate the impact and the consistency between the two approaches (i.e. n obtained by subjective choice of the station operator and n derived by a more objective process based on the minimization of the diurnal cycle CO₂ amplitude). Moreover, we further stress in the “Discussion” section that BaDS is a pre-conditioning algorithm for selecting time period representative of the atmospheric background from time series of CO₂ in-situ observations. Finally, in the “Introduction” section, we better differentiated between methodologies for time series decomposition (i.e. HP-Spline, STL, CCGRV, least square fitting) and methodologies for background data selection (i.e. ADSV, COV, REBS, SD, BaDS, Giostra et al., Thoning et al.).

Point 3: The legend should be explained clearly. For example, I am not sure what “CO₂_camm_detrend” is although I can infer. However, I recommend that the authors clearly describe each legend item in the caption.

Response 3: We updated the figure with the correct legend.

Point 4: All references with URLs should be properly cited following the reference style of Atmosphere.

Response 4: done.

Point 5: The introduction can be shortened by about 40%. For example, Lines 92-112 is a mixed body of texts among Introduction and Method.

Response 5: the introduction has been reorganized.

Point 6: There are unnecessary blank lines and typos such as a double-quote symbol at the end of Line 314. The authors may want to ensure there is no typos.

Response 6: done.

Reviewer 2 Report

Review comments for atmosphere-1050942-v1

General Comment:

The overall manuscript is well written and generally easy to follow for the general reader. The tables and figure captions appear correct. There are just a few minor issues with some of the figures. Spot checking of the references found no errors. The following specific comments are for the authors consideration.

Specific Comments:

Line 53. To this aim it is …. (consider inserting the word ‘it’)

Lines 89 – 91. Beginning ‘With respect to [19], we posed …….’

After reading the entire manuscript the reviewer finally understood (he thinks) this sentence. The problem is possibly the use of the word “posed”. One believes the authors are talking about identifying a robust method to identify the proper threshold value used to discriminate the CO2 variability.

Is this what the authors intended:

“With respect to [19], we paid particular attention to developing a robust method to identify suitable threshold values with which to discriminate the CO2 variability ….”

Lines 104-106. This sentence listed the main objective of the work. It would be good to see a specific reference again to this objective in the Conclusion statement. Perhaps the reviewer missed it, but one doesn’t really see a definitive statement that the authors met their original objective in the Conclusions.

Line 227. The word “instead” is not necessary and can be removed.

Line 278. The phrase “It is interesting to note ….” should be avoided. Let the reader decide the significance. The sentence can be written in a much more direct manner. Is it necessary to actually have the sentences stand alone? If so, is more context needed for the reader to help them decide that these observations are important to focus on?

Line 314. End of sentence – seems that the ‘ “ ‘ mark not needed.

Figures.

1. The x-axis of several of the figures is in Italian or using Italian abbreviations. For example, figures 5 and 6. It is up to the editor and the authors whether the x-axes need to be changed. It is obvious for the most part what is intended and a quick Google search provided verification.
2. Perhaps a more significant issue with the figures is the shading to illustrate confidence limits. Reading the document on screen and even magnified it was not immediately apparent shading was present and was still hard to see. This seems to be an issue with Fig. 3. The error bars in fig. 5 are better but even they were difficult to view on screen without magnification.

Author Response

Response to Reviewer 2 Comments

General comment: The overall manuscript is well written and generally easy to follow for the general reader. The tables and figure captions appear correct. There are just a few minor issues with some of the figures. Spot checking of the references found no errors. The following specific comments are for the authors consideration.

Response: We appreciate the time and efforts dedicated in reviewing this manuscript. We have addressed all issues indicated in the review report, and believe that the revised version meets the journal publication requirements.

Specific comments:

Point1: Line 53. To this aim it is …. (consider inserting the word ‘it’)

Response 1: done

Point 2: Lines 89 – 91. Beginning ‘With respect to [19], we posed …….’
After reading the entire manuscript the reviewer finally understood (he thinks) this sentence. The problem is possibly the use of the word “posed”. One believes the authors are talking about identifying a robust method to identify the proper threshold value used to discriminate the CO2 variability.
Is this what the authors intended:
“With respect to [19], we paid particular attention to developing a robust method to identify suitable threshold values with which to discriminate the CO2 variability ….”

Response 2: That’s right: the verb “posed” is inappropriate, it is a wrong translation from Italian. This has been corrected in the revised version of the manuscript.

Point 3: Lines 104-106. This sentence listed the main objective of the work. It would be good to see a specific reference again to this objective in the Conclusion statement. Perhaps the reviewer missed it, but one doesn’t really see a definitive statement that the authors met their original objective in the Conclusions.

Response 3: We add a specific phrase about this point in the Conclusions (lines 526-530).

Point 4: Line 227. The word “instead” is not necessary and can be removed.

Response 4: done.

Point 5: Line 278. The phrase “It is interesting to note ….” should be avoided. Let the reader decide the significance. The sentence can be written in a much more direct manner. Is it necessary to actually have the sentences stand alone? If so, is more context needed for the reader to help them decide that these observations are important to focus on?

Response 5: thanks for raising the point. We decided to eliminate the consideration.

Point 6: Line 314. End of sentence – seems that the ‘ “ ‘ mark not needed.

Response 6: done

Point 7: Figures

Response 7: we updated the figures with the correct month abbreviations on the x-axis. The shadings for the confidence interval have been darkened.

Reviewer 3 Report

The authors present an interesting and valuable work where a common algorithm is applied to quantify background CO2 levels at several distinct observing sites in the Mediterranean. Given the difference in location and spatial sensitivity, each site undergoes its own parameter optimization to remove non-background data while also trying to preserve the most data possible. Stationary observation platforms require careful and precise quantification of the background to interpret the datasets, thus this work serves to address an important area of study.

The manuscript is laid out in a logical order and each site is nicely described to provide the reader background. References are cited throughout which, at least qualitatively, show how these results fit into the context of other studies of the region and using data from these specific sites. Some details require clarification throughout the manuscript and my specific comments to these areas are provided below. I have also indicated some language issues below that I thought detracted from the quality of the text but were simple to remedy.

Specific comments:

Line 47: The use of “albeit” is a bit distracting at the beginning of a sentence. “While” or “recently” would fit better.

Line 83: The authors state that the method of Thoning et al [25] is widely used, but there is no description of what the method entails. Why is widely used?

Lines 207-209: How does the algorithm handle data gaps due to calibrations or outages? Is the uncertainty in the background estimate different between sites due to varying amounts of data?

Line 263: Does “data coverages” mean temporal coverage, or is there a spatial element as well?

Line 266: Is this line meant to reference Table 2, instead of Table 3? The discussion about the impact of seasonality (Table 3) comes later in the manuscript.

Lines 271-274: Some of these percentages do not match either table 2 or 3.

Line 307: Again, this appears to be a reference to Table 2, not 3.

Line 313: The line should read: “Only for PRS is there no difference between the two approaches.” Why do think this site is different? Is it more consistently representative of the background because of location/meteorology/etc.? Is this what you expect?

Table 4: A couple of the numbers do not match (even with rounding) when comparing with tables 2 and 3. Is there a reason for this?

Table 5: This should be closer to where it is first mentioned in the text.

Line 339: How are the 95% confidence intervals calculated? This is important information to include.

Line 395-369: The time series are “more consistent with each other than the original datasets.”

Lines 397-405: This section of the text is highlighted for some reason. It is stated that a multi-day peak observed at CMN is caught by the “stricter flagging strategy adopted at this laboratory.” How does this flagging strategy compare to those used at the other sites? In what way(s) is it stricter and does your algorithm require that sites which are particularly susceptible to local or regional sources/sinks operate with these stricter flagging schemes?

Figure 5. What are the semi-transparent vertical boxes in this figure? This should be clarified in the figure caption. It also looks like the letter denoting each month is in Italian (ex. G for January and June, L for July).

Line 438-440: This sentence is structured in a confusing way. I would recommend changing it to “The outstanding agreement of the CGR background to the other sites demonstrates the ability of the BaDSfit algorithm to extract the background signal even if the measurement site is strongly impacted by local/regional emissions.”

Line 519: The use of “testified” in this sentence is awkward and distracts from one of the main takeaways of the paper. I would recommend changing it to “testified to” or “shown.”

Author Response

Response to Reviewer 3 Comments

General comment: The authors present an interesting and valuable work where a common algorithm is applied to quantify background CO2 levels at several distinct observing sites in the Mediterranean. Given the difference in location and spatial sensitivity, each site undergoes its own parameter optimization to remove non-background data while also trying to preserve the most data possible. Stationary observation platforms require careful and precise quantification of the background to interpret the datasets, thus this work serves to address an important area of study.

The manuscript is laid out in a logical order and each site is nicely described to provide the reader background. References are cited throughout which, at least qualitatively, show how these results fit into the context of other studies of the region and using data from these specific sites. Some details require clarification throughout the manuscript and my specific comments to these areas are provided below. I have also indicated some language issues below that I thought detracted from the quality of the text but were simple to remedy.

Response: We appreciate the time and efforts dedicated in reviewing this manuscript. We have addressed all issues indicated in the review report as detailed below.

Specific comments:

Point 1: The use of “albeit” is a bit distracting at the beginning of a sentence. “While” or “recently” would fit better.

Response 1: done

Point 2: Line 83: The authors state that the method of Thoning et al [25] is widely used, but there is no description of what the method entails. Why is widely used?

Response 2: we phrased the sentence again, with more details about the method of Thoning et al.:” Among the different approaches, there is the one of Thoning et al. [25], initially developed for the measurements of Mauna Loa Station. It consists of two steps which analyze the variability of the hourly CO2 data (i.e. their standard deviation) and the difference between consecutive hourly means. Then an iterative algorithm removes any values which differ from the weighted spline curve by more than a given threshold. Due to its simple usability and due to the lack of any strong pre-requisite regarding the measurement site environmental conditions, it was applied to other remote stations like, for instance, Schauninsland, a site in the southwest of Germany [34], Junfgraujoch (Switzerland), Puy de Dôme (France) [35], and four WMO/GAW stations in China [36]”.

Point 3: Lines 207-209: How does the algorithm handle data gaps due to calibrations or outages? Is the uncertainty in the background estimate different between sites due to varying amounts of data?

Response 3: We thank the reviewer for raising this point. As concerning the question of the impact of data availability to the background estimates, we added a specific comment in the discussion section of the manuscript (lines 477-485): “Among the different approaches, there is the one of Thoning et al. [25], initially developed for the measurements of Mauna Loa Station. It consists of two steps which analyze the variability of the hourly CO₂ data (i.e. their standard deviation) and the difference between consecutive hourly means. Then an iterative algorithm removes any values which differ from the weighted spline curve by more than a given threshold. Due to its simple usability and due to the lack of any strong pre-requisite regarding the measurement site environmental conditions, it was applied to other remote stations like, for instance, Schauninsland, a site in the southwest of Germany [34], Junfgraujoch (Switzerland), Puy de Dôme (France) [35], and four WMO/GAW stations in China [36].”. Short data gaps due to calibrations or outages are filled using a simply linear interpolation. This information have been added in the revised version of the manuscript (Section 2.2).

Point 4: Line 263: Does “data coverages” mean temporal coverage, or is there a spatial element as well?

Response 4: It means “temporal data coverage”, this is now specified in the revised manuscript.

Point 5: Line 266: Is this line meant to reference Table 2, instead of Table 3? The discussion about the impact of seasonality (Table 3) comes later in the manuscript.

Response 5: That’s right, we corrected the typo

Point 6: Lines 271-274: Some of these percentages do not match either table 2 or 3.

Response 6: That’s right, we corrected the typo

Point 7: Line 307: Again, this appears to be a reference to Table 2, not 3.

Response 7: done.

Point 8: Line 313: The line should read: “Only for PRS is there no difference between the two approaches.” Why do think this site is different? Is it more consistently representative of the background because of location/meteorology/etc.? Is this what you expect?

Response 8: We added a specific comment at line 301 of the revised manuscript:” Only for PRS is there no difference between the two approaches: this is probably due to its remote location, more consistently representative of the atmospheric background and less affected by local source/sink through the seasons of the year”. We somewhat expected this, considering that PRS is a very high mountain sites located near the homonymous Alpine glacier with terrain covered by snow all year round (this would minimize any possible local natural emission/sink able to affect CO₂ variability ).

Point 9: Table 4: A couple of the numbers do not match (even with rounding) when comparing with tables 2 and 3. Is there a reason for this?

Response 9: It is a typo.

Point 10: Table 5: This should be closer to where it is first mentioned in the text.

Response 10: We moved it into the text near its first mention.

Point 11: Line 339: How are the 95% confidence intervals calculated? This is important information to include.

Response 11: confidence level is calculated through bootstrap simulations (more details can be found in Carslaw et al., 2019).

Carslaw, D.C. (2019). The openair manual — open-source tools for analysing air pollution data. Manual for version 2.6-5, University of York.

Point 12: Line 395-369: The time series are “more consistent with each other than the original datasets.”

Response 12: done

Point 13: Lines 397-405: This section of the text is highlighted for some reason. It is stated that a multi-day peak observed at CMN is caught by the “stricter flagging strategy adopted at this laboratory.” How does this flagging strategy compare to those used at the other sites? In what way(s) is it stricter and does your algorithm require that sites which are particularly susceptible to local or regional sources/sinks operate with these stricter flagging schemes?

Response 13: The flagging strategies to which we refer was adopted by the single laboratories which provided the dataset analysed by BaDS and it is not part of the BaDS algorithm. BaDS analyzes the dataset that these laboratories (CAMM, ISAC; ENEA and RSE) provided and it does not require any strict flagging scheme. Basically at CMN-CAMM the station operators rejected the data points that deviates from the overall data population (more specifications are provided in the revised manuscript, see lines 522 – 525: “: i.e., data points that deviates from the rest of the dataset are rejected by a data screening performed on the raw data by the station operators “). At the other sites, according with the ICOS data flagging strategy, data are invalidated only when a robust reason is provided (i.e. recorded sampling problems, observed interferences near the sampling inlet, recorded instrumental failures, …). We intended that a stricter strategy is adopted at CMN – CAMM (in respect to CMN-ISAC), because “anomalous” CO₂ data (i.e. data affected by high deviations or variability in respect to the overall data population) are typically rejected from the CMN – CAMM data set.

Point 14: Figure 5. What are the semi-transparent vertical boxes in this figure? This should be clarified in the figure caption. It also looks like the letter denoting each month is in Italian (ex. G for January and June, L for July).

Response 14: we updated the figure with the correct month abbreviations on the x-axis. The shading is for the confidence interval.

Point 15: Line 438-440: This sentence is structured in a confusing way. I would recommend changing it to “The outstanding agreement of the CGR background to the other sites demonstrates the ability of the BaDSfit algorithm to extract the background signal even if the measurement site is strongly impacted by local/regional emissions.”

Response 15: done, thanks for the suggestion

Point 16: Line 519: The use of “testified” in this sentence is awkward and distracts from one of the main takeaways of the paper. I would recommend changing it to “testified to” or “shown.”

Response 16: done

Round 2

Reviewer 1 Report

I don't concur that the authors addressed my major concerns adequately. 

Author Response

Response to Reviewer 1 Comments

Point 1: I don't concur that the authors addressed my major concerns adequately.
Response 1: In the “Conclusion” section of the manuscript, we specified the need to test BADS with other monitoring sites and other data selection methodologies present in literature. This will be done in a future work (lines 642-646).

“In this paper, the performances of BaDS methodology have been evaluated on a set of data produced by an Italian collaborative network. To better assess the BADS effectiveness, the application to a large number of monitoring sites, such as the European infrastructure network ICOS, as well as the comparison with other methodologies presented in literature (e.g. [20, 30-33]) is needed: this will be the focus of a future work.”

We evaluated as not necessary to add equations or diagram about BADS, since a detailed description of each single step of the algorithm is already provided in the text.