I found that the authors of this paper designed a thoughtful online field instrument to measure Oxidative Potential  (OP) of aerosols. They discussed the chemistry principles and physical characteristics underlying their instrument design.  They demonstrated methodologies’ functionality with calibration and case studies.  At the end, it’s revealing to see their method obtained a good correlation of PM2.5 data to support its technological significance.  The work is reported in good alignment with the journal’s theme.  For these reasons I would recommend its publication.

To inspire potential improvement, I have a few questions on a major aspect. The online and off-line OP measurements both had a reaction time of 20 min. Although it’s discussed the selectivity of 20min being its optimality toward DHA’s stability and relevance to mimicking conditions in lung cells.  Can the author discuss how different reaction time will impact the measurement outcome more clearly?  Also, Fig 3 suggests the online measurement can capture data within one min; I wonder how does this 1-min reaction be captured by a 20-min residence time? 

In addition to this, I am slightly concerned with the linear fitting on some data sets. In figure 4, it appears to me that the linearity of the intensity-DHA calibration starts deviating after 20 uM, maybe even more at a higher concentration range.  Can the author justify the linearity suitable range of the calibration maybe by adding more data points?  What will be the statistical confidence of the slope and what magnitude of error can be caused?  How to handle measurements larger than 100 uM DHA when working with ambient samples?  Another is Figure 5B, where it appears not very appropriate to fit the data by a line anymore- can the authors discuss the implication of the seemingly S-shaped trend?

With these, I recommend some revision of the work to be formally published.

This paper describes an instrument developed to measure PM2.5 oxidative potential with a single-component chemical assay, ascorbic acid (AA).  The instrument is impressive and the topic well suited for publication in this journal.  Some points for the authors to consider:

In the past, the AA assay performed on filter extracts have been conducted in two ways, 1) pure aqueous AA assays in which AA is the only antioxidant and 2) assays in which there are other antioxidants along with AA present, these typically involve monitoring AA in synthetic lung fluid (SLF). The authors chose method 1) since it increases the method detection limit.  This is reasonable, but it would be very beneficial if there was some discussion contrasting the reported results of methods using pure AA or AA in SLF, such as what chemical species or sources are associated with AA depletion in each case, what are the similarities and what are the differences, and if possible, what contrasting health effects are observed.

The authors find that a-pinene SOA has components that are highly unstable and can only be measured in an AA assay with an online instrument, such as the one described here.  The question is, are these species important, ie, toxic?  In real lung fluid there may be components that minimize these species and so lessen their toxicity (eg, antioxidants beyond AA) that suppress the ROS transported into by the particles.  It has been argued that it is the aerosol particle species that catalytically produce oxidants in vivo that may be the ones that are most important at driving oxidative stress and an inflammation response, such as PAHs and related compounds, and metals, since they can generate ROS without being consumed and can be enhanced in aerosols of mixed chemical components (see more on this below). These compounds also tend to be stable. There is an implicit assumption that all species that react with AA in a pure assay are equally toxic,( eg, see lines 222- 230), maybe actually a filter measurement that only measures the more stable species is a more health relevant measurement?  Something to consider.

Last line of Abstract, how is the OP LOD units ug/m3?

Line 195-196, why is the PILS collection efficiency so low, 20-25%.  It is not clear what liquid flow rates are very low which is stated to be the cause?  Please clarify. Could one add flow rates (air and liquids) to the schematic of Fig 2?  Where do the 75 to 80% of the missing particles go?  Is the resulting ROS measurement corrected for this low sampling efficiency?

Line 208, what is the pore size of the cellulose grade 1 filter, which will define the size of insoluble particles that pass through this filter.

Do the authors know if the denuders are necessary?  Eg, what possible gases would interfere?  Do the denuders actually produce more reactive gases?  One might wish to compare with other forms of denuders.  Why were such a large number of denuders in series used?

Is it true that particles deposited in the lung fluid will be at the lung fluid pH?  What about if incorporated into specific components of the lung fluid, such as macrophages?

Line 284. The point is not clear.  Metals are not SOA.  Will metals have the same issue as fresh SOA, ie, if not measured immediately their contribution to OP AA will be under-measured?

Line 325-330.  The conclusion that the AA assay response is x100 higher than that for a-pinene SOA, but SOA concentrations are much higher than metals so this evens things out is an overly simplistic analysis since the experiments were based on single pure components.  Real aerosols are a mixture of many compounds, such as metals and PAH-(and related compounds) that can synergistically affect OP.  As just one example, Fenton reactions that produce ROS are enhanced in the presence of semi-quinones that cycle Fe(III) back to Fe(II). 

Line 341 to 350, the tests with Beijing filters, is not clear.  Was the idea that the online analytical system was being compared to a manual analysis, both following the same protocol?  It seems like the filter was extracted in a water solution containing AA and then that extract analyzed by the online and manual methods for a comparison.

Fig. 7 caption is cut off.

Equation 1 has no correction for PILS collection efficiency; does this mean the measurement is about 75% too low?

Line 429 to 431, there is no proof for this statement.  As noted above, the authors are assuming that the more unstable components measured with the online AA system, but not the filters are health relevant, but what is the justification?  Maybe qualify this by stating something like, assuming all species, stable and unstable are of equal toxicity…. Or add to the line; severely underestimate health-relevant OP for cases of high concentrations of relatively fresh SOA.

The development of an automated online field instrument to quantify the oxidative potential of aerosol via ascorbic acid oxidation is presented in this manuscript. The authors conducted a series of calibrations of aerosols with known sources or compositions as well as reported data from ambient measurements. The use of fluorescence detection is important to avoid background interference from absorbance-based measurements. I think the manuscript is well written and the data is carefully analyzed. I have a few clarifying questions for the authors listed below.

•    How did AA maintain freshness for multiple days of field measurements?

•    Would the change in AA concentrations in the instrument over time (via natural decay) affect the kinetics of AA-aerosol oxidation reactions?

•    A limit of detection for urban ambient aerosol measurements was reported to be around 5 µg/m3. Given that different types of aerosol components may have very different reactivities towards AA oxidation, do the authors have any information about what major aerosol composition contributes to the current observation?

•    How was the slope for ambient measurements determined? (Line 410)