The manuscript by Lefrancois and coauthors evaluates a method to extract quantitative data from mass chromatograms of trace gases in air samples.  The analytical method they study is based on cryo-enrichment gas-chromatography mass spectrometry, using a Time of Flight (TOF) mass spectrometer.   In contrast to target compound analysis that is common using quadrupole MS in selected ion mode, the TOF analysis obtains full mass spectra over a given range continuously over the chromatographic analysis time.  As demonstrated in this manuscript, this ability allows one to examine past chromatograms for mass spectra of compounds not identified at the time of the analysis.  The problem addressed in this manuscript is how to obtain quantitative measurement of those compounds without contemporaneous standards, and to determine the uncertainty with this retrospective approach.  The authors do a very good job of laying out the procedure and how it was evaluated through several years of analysis of samples collected in flasks and in-situ at the Taunus Observatory.   The work is logically presented, thorough, and appropriate conclusions are drawn.

The method that the authors describe is equivalent to the use of internal standard quantitation that is common in analytical chemistry.   In conventional use, a known amount of internal standard (or multiple IS’s) is added to each sample and relative response of the standard(s) to the sample components are used to obtain quantitative data.  The assumption is that absolute response variations will occur, but relative response between internal standard and sample component will be constant.  Since no standards are added to each sample, the technique described by the authors uses previously quantified trace gases as internal references for comparison to a series of hydrochlorofluoro-olefins (H(C)FOs). They carry out a rigorous study of the stability of relative responses of gases contained in their calibration standards to find periods of high reproducibility and low variation, and which gas(es) in the calibration mix demonstrates the best properties.   Given the choice of reference standard and identification of periods of stability, the author’s then do the retrospective quantitation of several H(C)FOs and compare them to more recent analysis that used a direct calibration.  They estimate accuracy of the retrospective analysis at better than 25%.   Given the very low mixing ratios that are found, this level of accuracy seems more than adequate.

The analysis raised a question for me (but that doesn’t need to be addressed in the manuscript).  The question is why don’t all compounds work equally well as internal standards?  And if they don’t all work well, what is the uncertainty of the non-target compounds for any specified period?  They might behave well or might not, it seems to me.  Also, given the stated precisions of the calibration standards, it is then surprising that the correlations between standard responses show percent errors in the 10% range.  I’m not sure how to interpret that.

Other comments are related to some clarification and minor editorial suggestions.

L 81.  Suggest eliminate “preceding” or change to “…(ppt)), sample trace gases are enriched by cryofocussing in a sample loop.”

L113.  Though I think I know what you mean, could you better describe what a “target” standard is?

L120.  It would be helpful to describe some more detail of the method.  For example, I couldn’t find anywhere how the quantitation was accomplished with the QTOF data.  Was a single selected ion used for each compound, or the sum of major fragments, or some integration of a peak that has been deconvolved from the total ion chromatogram?  Were different methods of sample integration tried?  Perhaps there is also a relationship between mass and total ion current that could be used to quantitate certain classes of compounds (at least within 25%)? Other questions: mass resolution of TOF?

L177.  Although HCFC_141b elutes near water, it shows excellent precision.  So not sure why this might be excluded.   Or it might be interesting to learn how water vapor might influence the results.

L179. Not sure if the plots are artificial data from some “arbitrary substances”  or if the actual compounds are just not named here.   Could you clarify?

L182.  It is not clear how the 10% criterion for rejection is applied.  Is this from point to point, or relative to some average?

L197.  Not sure if you mean “exemplary”, as in “best example of the group”, or are these just examples of some of the compounds.  (also in Figure caption).

Figure 2.  As I understand it, this figure compares the peak areas of compound pairs in the same calibration standard over the time of the study.  Could you comment on the very large range of peak areas that were observed?   Is this a characteristic of the TOF?

L204.  Note that the independence of rRF will also depend on linearity and any zero offset.

L209.  The observed shift of 152a relative to 133a deserves some comment.  Presumably there was no similar shift in the time series ambient measurements of either gas.  So, this behavior, though maybe rare, would seem to be a major red flag in applying the proposed method with confidence.

L213.  There are a number of compounds that have drifts or anomalies that prevent them from being used as “reference” compounds.  Does this have any implication on how these are used for direct calibration?   Do these standards cause the sample mixing ratios to be flagged?   The authors also suggest that there are a number of potential factors that will influence the relative responses.   In cases of outliers or large shifts (such as 152a), have the authors determined specific causes for the deviations?

L214.  Not sure of meaning…change “suited” to “suitable”?

L251.  I was interested to see that 152a was selected as a reference standard for the in-situ measurement evaluation, though there was a problem with this compound in the canister analysis.  As noted, this is disturbing and deserves comment.

The manuscript "Non-target analysis using gas chromatography with time-of-flight mass spectrometry: application to time series of fourth generation synthetic halocarbons at Taunus Observatory (Germany)” by Fides Lefrancois, Markus Jesswein, Markus Thoma, Andreas Engel, Kieran Stanley, and Tanja Schuck, is a well written account of their efforts to produce an alternative indirect calibration method for use in detecting short-lived compounds with large atmospheric trends in historical GC-TOF-MS data archives. This is a novel and interesting idea and presents a strong impetus for other research groups in this area of research to try out this kind of analytical approach. In short, the method circumvents the problem that arise when the calibration gas in use at the time of the air-sample analysis (canister or in-situ), did not have detectable amount of the target analyte or this amount in the calibration gas wasn’t quantified. The method thus enables retrospective analysis of these kinds of datasets which can then be regarded as a type of digitized air archived. This method proves especially useful when considering several 4th generation CFC replacement compounds (HFOs/HCFO) which, up to recently, have only been present at very low mixing ratios in the atmosphere and with infrequent detections in the air archival data. The authors describe in detail the individual steps in their data evaluation for selecting data with good underlying quality predictors (based on relative response factors… i.e., extends standard analytical textbook internal standard GC-methods) and demonstrate the application to the detection of two common HFOs and one HCFO, in in-situ measurements and flask samples collected at the Taunus Observatory near Frankfurt am Main, Germany. These compounds have only become detectable in atmospheric samples over the last few years and thus the paper is of both timely and scientific importance. The manuscript is well apportioned and a delightful read with very few errors and a clear narrative. I have only trivial and minor comments that should be addressed before the paper can be accepted for publication.

General comment: The manuscript contains a large number of figures and perhaps one or two of those could be relegated to SI ?

Abstract – line 7: “thus” can be omitted in the sentence.

p. 2 line 36: Typo in spectrometry (spectrometry).

p.2, line 53: It is incorrect to say “TFA is known to cause negative environmental impacts”. Large concentrations of TFA will do that, but the authors should consult the references they cite themselves, especially Solomon et al., for a precise characteristic on this matter, what impact the current and predicted future levels of TFA in the environment will actually have.

P4. line 115. Mole should be capitalized ion the beginning of the sentence.

p.6 line 159-161: This sentence should be rewritten for clarity. It would benefit form some commas and perhaps start out with “Using equating 3, the rRF for the species of interest, which is not…..

p.6, line 165: replace “should” with “is assumed to”

p.7, line 173: I guess the selected compounds listed in table 1 could be termed “a training” set for the method. As such, the authors, and other researchers, will adopt similar or dissimilar training sets, for the technique to be “calibrated’ on for application to their datasets.

p.7, line 182: 10% - this this value arbitrarily chosen? Why not based on a statistical parameter such as sigma(s)?  

Figure 4 caption: replace “used cylinders” with “used calibration gas”.

All figures could in general benefit from being made color “agnostic”. Several of them, e.g. fig 6 and 8, are only legible in color print out, whereas there are no limitations on symbols shape that dictates the necessity of using colors.

P.15, line 254-255: Please comment on the fact that a larger fraction of data from the in-situ measurements than from the flask measurements were selected. Was this expected? Any predictable reason behind this outcome?

P.16, line 262: What does “partly different” mean here? Could a different description be used?

p. 16 line 275, Suggest replacing detectability with detection frequency.

p.16, line 278 and 281: inset “ weekly” before “mole fractions”.

p.18, line 285-287: How would this indirect retrospective method likely work out trained on a data set like that collected at Jungfraujoch – i.e., a setting with largely clean background air? Any ideas if it would work there as well?

p. 19, line 295-297: Does this mean that indirect values obtained from datasets involving flask measurements and in-situ measurements, respectively, shouldn’t really be directly compared? I.e., if the variability is quite unpredictable, which sample dataset is "generally” more likely to produce good indirect values.

p.21, line 311: “sufficiently” is probably a better word here than “rather”.

p.21, line 311-312: regarding “reference species with similar retention times” – has the sensitivity to the retention-times been tested? Is it just assumed “likely”- I’m not saying that this is not a good assumption , just wondering if the authors have tested this – if not, this should be an easy test within the GC-TOFMS data sets.

p.21, line 317: “… retainage a sufficient number of measurements”. Sufficient here means a very low number? Ref. Table 4 where e.g. 2018 has 3-6 observations?

p.22, line 325: These quoted different are much lower than what shows up for the annualized values in the tables. Is the large discrepancies for the annualized values not an issues since those are what are often cited?

Review for manuscript by Lefrancois et al., 2021.

The manuscript by Lefrancois et al. presents a method to derive quantitative information from the measurement of trace gases made by preconcentration, GC-MS, even in cases where the substance in question is absent from the calibration gas. This method answers a growing need for making both retrospective screening and non-target screening quantitative. The method is then applied to retrospective screening of emerging HFOs, and therefore reconstructs quantitative molar fractions values for the period 2014-2018, for the site of Taunus Observatory in Central Germany. We can expect that the presented method will be widely applied in the future, to accompany the deployment of high mass coverage, high mass resolution instruments at monitoring stations.

The paper is generally well organised and Figures well chosen. It is evident from the data presented that the method has been well tested, and the results are of good scientific value. However, a few minor improvements would help future readers. Reformulation of a few sentences is needed (specific suggestions hereafter). Also, where possible, it would be very interesting to mention potential causes for the rejection of about half of the data points.

Comments are detailed hereafter.

I would suggest to use a title that contains the main purpose of the article, which is to make non-target screening quantitative. Usually, articles with "non-target screening" in their title contain discovery of new substances, which is not the scope here.

Suggestions for title: "Quantitative non-target ...."

l. 8: I would simplify the grammar: "This archive can be used if", or even "This archive can be used for retrospective screening"

l. 11: "or the amounts in the calibration gas may not have been quantified." 

Introduction, l. 18-20: "The application of the indirect calibration method on several test cases can result into accuracies around 13% to 20 %. For H(C)FOs accuracies up to 25% are 20 achieved." I would be good to reformulate these sentences to really convey the meaning that low values represent a better accuracy. Maybe you can replace the word "accuracy" by "uncertainty" here: "The application of the indirect calibration method on several test cases can result into uncertainties around 13% to 20 %. For H(C)FOs, of particularly low mole fraction values, uncertainties up to 25% are observed.".

l. 26 "which are part of or affiliated to"

l.31, maybe: "is not well covered"

l. 33: "by the installation"

l. 35: maybe you want to specify the mass range coverage of the TOF instrument (minimum and maximum measured masses).

l. 45-47: may you can consider if you would like to leave out "HFC-1234yf" and "HFC-1234ze(E)". The HFC nomenclature is actually not made for compounds with a double bond.

l.51 : "have no ODP": "have an ODP value of zero." "have no ODP" suggests that the computation of the ODP is impossible.

l. 57: the magnitude of what? Amplitude of annual cycles, magnitude of mole fraction of pollution events?

l. 80: "and each pair of measurements is bracketed"

l. 81: "range of parts per trillion (ppt)": range of picomole per mole, pmol/mol or hereafter part per trillion (ppt)"

l. 102: add comma: "For each measurement, approximately"

l. 117: you can leave out the sentence about calibration scales, it is already mentioned l. 92-94.

l. 139 : "before calibration standards containing measurable amounts of these substances were used".

l. 140: tense concordance, not sure, check with native speaker. "When these compounds were detectable in ambient air, the peak areas could not be converted to mole fractions using Eq. 2 because neither numeric values for Acal nor rR were available."

l. 141: you surely mean: "between another compound which is measurable in the standard"

l. 144: "that means that the ratio of signal per amount of analyte for the two compounds is constant with time." I'm not sure about the meaning of this sentence. We know that the response of a MS instrument may vary strongly over time, for example the instrument response increases after source cleaning. However what is important here is that the instrument response behaviour should vary similarly over time for all substances, as you clearly write afterwards. I would rephrase as: "Ideally, the sensitivity of the analytical system for two different species should behave similarly over time. In such a case, the ratio of responses R of two given species should be close to constant."

l. 146: "this ratio should be the same for any sample.": maybe too general. Suggestion to write more specifically: "this ratio should be constant over time for any chosen pair of compounds".

l. 155: "It must be stable over time". Check entire manuscript.

l. 164-166: meaning not clear. A non-stable sensitivity does not necessarily imply a non-stable relative sensitivity, this is something you are going to investigate next. Suggestion to rephrase: "The methodology outlined in 3.1 is based on the assumption of a constant rRF in Eq. 4. In reality, the absolute sensitivity of a mass spectrometer is known to vary over time, in particular after tuning the mass spectrometer or after modifications of the analytical system such as replacement of filaments, columns or sample loops. It is therefore an open question whether changes in the relative sensitivity rRF should also be expected or not. Thus, to evaluate [...]"

l. 169: "need to separated": "need to be evaluated separately".

l. 181-185: difficult to understand, suggestion to rephrase: "To identify periods of stable rRFevalu for a given pair of compounds, timeseries of rRFevalu are reviewed. To do so, for each measurement or data point of rRFevalu in the timeseries, we compute the sum of other rRFevalu data points that do not deviate from the chosen data point by more than 10%. The data point with the highest number of matching data points is used as a reference (shown with red cicle in Figure 1, panel (b)) and all data points that fall outside the 10% interval are excluded (shown as grey data points in panel (b)).

Note: I would not use "independant measurement", since the measuring instrument is the same of course the results are not fully statistically independent, and we actually need the results not to be independent for this method to work.

To make it more clear, on Fig. 1 please mark with e.g. a red circle the data point that was selected as most likely rRF value.

Table 1: add bibliographic reference to all scales where needed. 

METAS-2017: Guillevic et al., 2018 (ok, already done). 

EMPA-2013: for HCFC-133a: Vollmer, M. K., Rigby, M., Laube, J. C., Henne, S., Rhee, T. S., Gooch, L. J., Wenger, A., Young, D., Steele, L. P., Langenfelds, R. L., et al. (2015), Abrupt reversal in emissions and atmospheric abundance of HCFC-133a (CF3CH2Cl), Geophys. Res. Lett., 42, 8702– 8710, doi:10.1002/2015GL065846. 

EMPA-2013 for HFOs: Vollmer et al., Environ. Sci. Technol. 2015, 49, 5, 2703–2708.

SIO-05, SIO-14: Prinn et al., J. Geophys. Res., 105, 17,751-17,792, 2000, and Prinn et al, Earth Syst. Sci. Data, 10, 985–1018, 2018.

l. 195: you probably need "the" in front of all "MAPE", check through the manuscript. I would add the equation for the computation or a reference (e.g. the Wiki page).

l. 199: "Except for HFC-227ea"

Section 3.2.1, general question: could you find explanations for the outlier rRFevalu data points?

l. 200-201, I would try to reformulate in an easier way. E.g.: "To test which pairs of substances produce the highest correlations, all possible pairs of substances have been tested. The obtained values for r2 and MAPE are shown in Fig 3".

l. 208, typo: "all cases where HFC-152a is involved." Also, it seems to me more to be a drift in the rRF value, that started before the change in standard tank, and stabilised after some runs of the new standard. Such a drift (albeit much smaller) can also be seen in the HFC-125 data points. So I'm really not sure that you can link this for the standard tank change. I would remove this sentence.

l. 211: maybe you can comment on why HFC-227ea and HFC-245fa? HFC-227ea seems logical to be a bad one, as its measurement standard deviation given in Table 1 is one the highest. However why HFC-245fa? Or, alternatively, you can explain later why some are good ones?

l. 220-222: "To quantify the differences between the selection of data of main reference and test substance via main reference substance and an evaluation substance we compared the relative standard deviations of the resulting filtered data sets." I don't understand this sentence. Please clarify. You may also want to cut into smaller sentences. Maybe, adding the equation you use will help to understand what you compute here. 

Usually there are two quantitative values to characterise a result: its standard deviation, which reflect the random noise, and the average difference between two values (usually a test value and a reference value), which is a systematic bias. A bias not equal to zero means that the method causes a systematic error.

Now based on Fig 5, maybe what you want to express here is a precision loss, that you express via the difference in standard deviation? If this is really the case, here is my suggestion:

"To quantify the precision loss between direct calibration and calibration via a transfer substance, we compare the relative standard deviations of the resulting filtered data sets.", or something similar.

Another important quantity to evaluate is if your method creates a bias or not? i.e. what is the average value of the distance (or difference) between the true and reconstructed value? It should be (close to) zero to show no bias. (cf see below comment on Table 3)

l. 237: if you mean precision loss, use: "the difference between the standard deviations".

l. 241: "As test cases to apply the indirect calibration method, we chose..." or "As test cases to be applied the indirect calibration method, ...".

l. 243: "mole fractions of HFC-227ea show..."

Table 3: average relative difference: this is your metric for the bias, right? Please write the equation somewhere in the text (e.g. around l. 245). Also: usually, if the bias or systematic offset value is within the 2 sigma standard deviation, it means within uncertainty, there is no bias. This is an important point to show here. But in Table 3, the "av. rel. difference" value is systematically more than the value of "standard deviation". Can you comment on this?

l. 275: typo: "HFO-1234yf"

l. 276: concordance of tenses, "increased continuously up to 100%"

l. 311-312: "Further, it is likely that using reference species

with similar retention times as the target species provides more stable results." Can you give an example here? No retention time data are provided.

l. 313: "good results" is subjective. Maybe use a quantitative value instead, e.g. "which yield the minimum number of rejected data points".

l. 330, typo: "is the measurement", "which are expected".

Your data show a rRF that is mostly not stable over time. Can you discuss the possibility to use a running-mean rRF value over time, instead of assuming a constant value over a short time period? Also, at least for some time periods, could you assign a (hardware?) cause to the non-stable rRF?

Figure 5, legend: "Illustration of data selection for the weekly flask sampling measurements..."

The paper by Lefrancois et al. addresses a relevant question in the field of atmospheric background observations and scanning using GC/MS. Basis for every high-quality measurement is a good calibration of the instrument. While for GC FID systems, applying C-response factors, a calibration is achieved also for substances not available in the reference gas mixture, for GC/MS systems, the existence of a compound in a reference gas mixture is usually mandatory.

The here presented method to derive calibration factors from known and well calibrated substances provides a solution for new gases which are not or have not been available in calibration gases at the time of sampling. The relative response factors usually applied for the Internal Standard concept is here extended to an external standard. Similar methods are certainly used in labs as a work around for compounds not available in working standards gases, but Lefrancois et al. provide a description of the concept and present the results of elaborated testing of this method. Application of this concept leads to a new data set of recent HFO data observed at the Taunus Observatory. HFOs were quantified until back 2014 and the quality of the data was assessed proving the applicability of the here presented method.

The overall presentation is well done, presented results support sufficiently the interpretations and conclusions. The authors have clearly indicated their work and contributions from Co-Authors and as far as my understanding goes the presented results are traceable. Some (minor) issues that require some discussion though.

The title seems quite long. The relative response concept used for the indirect calibration and its validation is not mentioned even though it the major aspect of this paper as it is written. Therefore, the authors might think of a revised title.

The language is basically good. I am not a native speaker myself – as you can easily read here – but in in some parts the use or non-use of the definite and indefinite article seem a bit arbitrary and some sentences should be re-formulated for clarification.

Close work with European metric institutions has brought up a discussion on “correct vocabulary”.  Among this the terminus “mole fraction”. As “mole” is a unit, NMIs requested the use of “amount fraction” instead und the correct unit would be “nmol/mol” (instead of ppt) - just a remark, as I came across this discussion often recently.

As already stated, the overall presentation is good, however, some parts suffer slightly from not always clear structure leading to some open question (especially Section 3). Examples are listed below.

Introduction:

l. 47, “These hydro(chloro-)fluoroolefines are the so-called fourth generation of synthetic halocarbons…” - there are no other fourth generation synthetic halocarbons?

l.51:  “However, some HFO, as some HFC and HCFC, can form the very persistent and toxic trifluoroacetic acid (TFA) as…” check commas and the use of "some “

Are the three substances of this paper among the TFA forming HFOs?

l. 63 & 64: I think it has to be “Section” with “S” à please check!

l.70: “locations of industry” à maybe better “industrial areas” ?

Section 2:

l. 77: What does “approximately weekly” mean?

l. 86: “ …the sample loop is heated to approx. 200 °C…”  - approximately?

l. 99 “mud dauber” …an insect screen, I guess?

l. 107ff: Are the sample inlets mounted in a separate heated box or are all parts in one single box? The description for the continuous instrument jumps a bit. Starting with the sampling procedure (flows, sample volume, desorption temperature) you continue with the unit set up (heated box, materials) and then return to the flushing procedure. The adsorption temperature is the same as in 2.2. (-80°C)? Streamlining this paragraph might improve the reading.

l. 133 : You neglect data for calibration intervals which deviate more than the weekly 1s-precision, ok – I am just interested: Did you ever take in account/discuss to add the additional error to the uncertainty?

Section 3:

Figure 1: as I understand this figure is explanatory only, in order to describe the method how stable periods were defined. Therefore the plotted compounds are not mentioned. Nevertheless, the question bothered me while looking at the plot, what substances are plotted. Later on, in Figures 5 and 7 you show the similar plots again. I wonder, if you could combine those?

l. 181: Why did you choose the stability criterion to be 10%?

l. 195 ff.  I had to read this sentence several time. Just to be sure: MAPE is the difference “A_measured – A_odr”, with “A_odr” being the ODR fit of peak areas forced through the origin

l. 210: “Using HFC-227ea and….” You already present a single result of the analysis that follows in the next paragraph. I found this is confusing. From my point of view you can delete this sentence.

Figure 5 (&7): I understood that you derived stable periods from calibration measurements as well. However, in the caption of Figure 5 you mention “data of the weekly flask sampling measurements“. Please clarify.   

l. 221ff: “To quantify the differences…” I do not understand this sentence, please explain. As a consequence please also explain panels d and h in Figures 5 and 7. Not clear to me.

l. 229ff: “In summary…” These are the characteristics you expect from your relative Response factor to achieve a good calibration and which set the frame for your checks you presented in the preceding lines. Might be good to have it earlier in the text?

l. 237: “Using HFC-125 as evaluation substance with HFC-143a, the difference standard deviations of the mean rRF selected via the test substances and selected via itself ranges between 1 and 10 %.” Do you mean “different”?

l. 243: “In this test case, mole fractions of HFC-227ea shows the best correlation …” “show” instead of “shows”.

l. 249:  You end this paragraph with a statement about the importance of having a constant sensitivity. You take up this point but then do not really discuss it. It would be nice to have all the presented results of 3.2.2 “wrapped up” at this point in final short summery

Table 2: With standard deviation you mean the standard deviation of the average relative difference? Is it this really necessary to have this table or could you add this information into Fig.6. ?

Figures 8 and 6: When you calculated the amount fractions via the indirect way you applied the method for data in the stable periods (as derived in Figure 5 and 7 respectively). The constant rRFs you used to “attach” the test substances to the reference substance are determined using the average rRF in the calibration measurements also during the stable times? This is not clear to me?    

Why did you use the measurement precision as error bars? How does this reflect the uncertainty factors of the different calibration methods?

Figure 6: What happens if you omit the two single data points in January 2018 for HFC-32? Any idea what happened here?

Figure 8: What happens to the regression lines when you omit the high amount fractions?

Tables 2+3: Do you use this deviation to derive the uncertainty of the method?

l. 262: “This is caused by long-term drifts…” What does this tell you about the possible errors arising from differences between the evaluation and test substances, even though you have filtered out periods with a larger variability?

Section 4:

Figures 9 and 10: maybe you keep the zoomed in plot, only? What do the error bars represent?

l. 291: “These larger amounts could be….” So, this is the effect of non-linearity or a larger integration error for the small calibration peak?

Conclusion:

You have presented methods to derive the best possible reference substances for your indirect calibration and you evaluate this indirect calibration procedure regarding its performance.

It would have been nice to see an assessment of how uncertainties arising from the determination of the rRF and evaluation of stable periods (by the evaluation) are reflected in the results of the test compound analysis. E.g. do expected errors match with observed differences between indirect and direct calibration?