|The manuscript by Kuyper et al. is greatly improved in its organization and presentation of the results. However, I have found some issues that I think need additional clarification. These are related to the analytical methods and calibration. I thank the authors for providing an actual spreadsheet file of their standards and air analysis. This definitely helped me understand the procedure much better, but also brought up some new issues. In my view, though the authors show a good fit to the bromoform peak area vs mixing ratio calibration, the random errors are very large. Furthermore, there are a few statements in the description that are not accurate. First, a linear regression is not a measure of accuracy, so the claim of 99.6% accuracy is not appropriate. In terms of regression analysis, I put the authors’ data, not including the measurements from August, into a statistical program (SYSTAT) and found an R^2 of 0.86 and a regression line (similar to the authors’) of Area = 129 (-+168) + 227 (-+7.5) * Std Mixing Ratio, where the -+ is the standard error of the estimate. The regression analysis was used by the authors to calculate all of the air measurements. The authors apparently did not use the standards interspersed with the samples to calculate any drift as claimed in the manuscript. The same area/mixing ratio relationship was used for all calculations. If the standard responses were used for drift calculation, then the calculated data would be much different. One point regarding this method is that a zero area produces a mixing ratio of 0.83 pptv. It may be worth pointing that out.|
There are a number of troubling aspects to the calibration that suggest problems with their technique. The variability of each standard level is remarkably high, with ranges over a factor of 2 or more in the response factor at each level. Another issue is the variability of the blank. As indicated on the spreadsheet about 25% of the blanks produced measurable peak areas, and these range from 1.36 to 8.85 pptv bromoform. Clearly there was some issue with carryover or some other problem that needs to be examined. The next major issue is that the response variability at the 21 ppt level (near the mean ambient concentration) is about 55% for one standard deviation. Since the authors have no data on replicate ambient analysis, one must assume that the variability of the ambient data is also in this range. An overall variation in the response factor considering all standards is 43% (1 SD), still remarkably high, and this variability must be considered for the ambient samples. Or, if the authors think the analytical system for ambient samples somehow has a better precision than that demonstrated by the loop injections of the standards, there should be some explanation of why. Or they should assume that the response variation in the standard injections reflects actual response variations of the detector and correct the ambient data for that response change. For the future, my suggestion would be to work with one or more secondary ambient air samples that are calibrated against a primary standard. (And to determine the source of the variability in the loop injections!).
The authors note that the drift was measured from day to day using the 1 – 3 loop injections. However, this day to day “drift”, sometimes calculated from one standard only and sometimes from multiple standards, varies over a factor of 3. Apparently, though this daily “drift” was not taken into account in the calculations. This should be discussed in the methods section.
Another discrepancy, based on the spreadsheet information, is the frequency of the standard injections. The manuscript describes standard injections every five samples. However, this does not seem to be the case for each day. Some days have no standard injections, others multiple, and others with 7 – 8 samples run with no standard injection. Typically, no standard injection ended the daily runs. Please correct this description of standardization frequency in the text.
Finally, I am curious about data that appeared in the spreadsheet, but was not included in the manuscript. The spreadsheet shows several values of bromoform > 60 ppt that are not shown in the manuscript, and apparently were also not included in calculating sample average. It is not uncommon to have “flyers” sometimes in the data, but it would be helpful to mention if some data are excluded.
I thank the authors again for sharing their raw data with the reviewers. Based on this data, I request a more thorough and accurate description of the analytical procedures, especially related to calibration and calculation, and how the variability observed impacts the uncertainties associated with the reported measurements. At this stage, I can’t be sure that the lack of drift correction hasn’t masked features in the data, and I am also unsure about the absolute calibration, which makes it difficult to compare this data with others. If these measurements are to be published, there needs to a much clearer description of the various uncertainties and limitations.