General Comments

This study documents an extremely valuable resource for the global observational and modeling-based ice nucleating particle communities. The data set is indeed a special one in containing data from multiple sites in the Northern Hemisphere that were collected in some cases for very long periods of time and over diurnal cycles in many cases. I did stumble with accepting the position that a special advantage of the sole use of PINE data removes the need for multiple instruments. While I understand that it could be possible to operate a network of such instruments to emphasize a procedure (not unified in this study) to focus on collecting higher temperature data at INP concentrations higher than about 0.01 per standard liter, the study also makes a compelling statement to me that confirms the need for integrating measurements such as those made offline using filter samples to adequately cover the full span of cloud conditions expected in the atmosphere. The PINE data represent a great advancement on the acquisition of substantial data that has special application to modeling needs but cannot easily access the range of temperatures and lower INP concentrations that may be most critical to representing and simulating mixed phase clouds. Am I wrong to conclude that? I also emphasize below the need to be a little more explicit about the methodological analysis of INP concentrations using raw PINE data. It may be obvious to the PINE community, but it bears attention for detailing to those not versed in using this instrument. Further, discussion of inferred sample volumes will help the casual reader understand better the extent to which the global atmosphere can be interrogated using real-time sampling instruments. My recommendation is that this paper could be modestly revised and be fully acceptable for publication.

Specific Comments 

Abstract

Lines 4,10: The statements regarding hours of sampling or numbers of INP measurements are important and relevant, but also important and relevant are the volumes collected. Certainly, that information is more revealing about the sampling possible with real-time measurements. Perhaps give it in standard liters or standard cubic meters. Is that possible? I think, based on what I learned later in the discussion around Figure 2 it means that the data set herein represents an assessment based around 800 m³ of air at defined conditions. It is a lot by the standards of historical INP measurements but not necessarily by the standards of trying to represent what is present in the global atmosphere. This is not a negative point I am asking to be documented, just a practical one.

Introduction

Line 51: I find the statement that INP concentrations are “known to vary by several orders of magnitude during a single diurnal cycle” to be a bit misleading regarding the more typical scenarios at one set of conditions. I suggest "can sometimes vary" describes the situation more accurately. Certainly, with air mass changes there can be such a variation in a diurnal cycle and I believe this has been captured in a few earlier papers regarding changes following atmospheric aerosol perturbations (e.g., ones documenting changes following rain events, or passing through smoke plumes when operating continuous flow chambers) besides the one referenced. But such data, including the reference cited, also shows sometimes days at a time without a major diurnal cycle. This can vary by site and the season. Your point though is understood and valid, nonetheless, so only recommending this small change.

Line 62: Is the “less than 0.01 L^-1” lower bound before or after consolidating measurements into the 6 h time intervals?

Line 91: Before the next paragraph on uncertainties, I expected to learn how the concentrations are defined. Are they continuously measured during the expansion and somehow integrated over narrow temperature ranges or is some time used around the lowest pressure and temperature of the expansion for which concentrations are defined by the volume sampled? This information is in older papers I believe, but it is not in section 2.3. The size distribution is mentioned, but over what time interval and acquired volume is that defined in each expansion?

Campaigns

Considering the previous comment, could information on single INP measurement sample volume also be included in the information for each project. Otherwise, there are details relevant to PINE users about times spent flushing and expanding and refilling, but not on effective sample volumes per sample. Is it always the same in each study?

PINE analysis and software

Lines 227-228: Is the OPC particle size distribution mentioned as critical information that is measured at the minimum temperature? Or over what time to that point? Again, trying to confirm understanding of how concentrations are determined, exactly, and what volume is represented.

Data subsetting

Line 244: This is the first mention of the 2 L volume. Please integrate this into the above discussions to make it clear how this is determined.

Figure 2: This is a very nice figure!

The GLoPINE INP data set

I understand that this is a database paper emphasizing the utility of a unique and comprehensive data set for constraining model representations of INPs. Still, is there a need to make any statement regarding the suitability of the method for providing more measurements at above 255 K, or if this method can be stand alone to characterize that regime where there are still many time periods below detection limits? I have also not raised the issue that concentration measurements are but one way to characterize INPs, though surely even the modeling community is going to ultimately want to know more information about validating their representation of sources and if ways are imagined for integrating PINE technology with other techniques to address that question.

Conclusions

The last comment above should perhaps be addressed here instead. At least one statement on the deficiencies of single instrument use could be helpful. The limitations at higher temperatures seem clear and a continued need for the community. One imagines that PINE data is no less valuable if combined with another, especially for “robust” constraint of model representation of INPs. Otherwise, it seems that model parameterizations are wholly trusted for comparison to total INP concentrations.

Herbert et al, present a collation of PINE-based INP measurements for the community to use. This is an excellent resource for atmospheric science community and then manuscript is generally well written. That said, some clarifications on the INP concentration calculation and counting statistics should be included in the manuscript to make it more assessable for the community, who has for the most part, are used to using INP measurements from offline techniques. Below are some additional comments the authors should consider before publication.

General comments:

There is no description of how the humidity in PINE is controlled or measured. If ambient air is introduced into the chamber and the walls are set to subzero temperatures, what keeps ice from forming on the walls of the chamber or for a cloud to already form due to isobaric mixing before the expansion even begins? Are cloud droplets in the OPC a necessary precursor (immersion freezing) for ice crystals to be counted as INP?

Is there a reason the flush varies so much between campaigns?

For the postprocessing software, has a comparison been done on the same dataset to see the impact of how the size threshold is obtained. This would be an important thing to check for harmonizing the dataset.

Is the INP concentration only reported at the minimum temperature of the experiment and is it calculated from the number of ice particles detected at said minimum temperature during an expansion? Or is the number of ice particles detected during the expansion used to measure the cumulative INP concentration until the minimum temperature? If it is the latter, how much of the volume sampled through the OPC is truly coming from the minimum temperature during the expansion? Is there a correction accounting for this when the minimum limit of detection is calculated. More specifically, if during an expansion only 0.5 L of air are measured through the OPC while the chamber is at its minimum temperature, then the limit of detection at T_min is not based on the entire 2 L sampled during the expansion but on 0.5 L. This makes calculating the true limit of detection at T_min extremely challenging, especially when the ability of an aerosol to be an INP is so strongly dependent on temperature. This should be clarified and discussed in the manuscript (see specific location below).

Following up on the previous comment, is the probability of an INP being removed before the minimum expansion temperature accounted for i.e. 2 L of the original volume has been removed before the end of the expansion in the ExINP-GVB example (lines 248-250)?

The calculation, description and purpose of CV could be made clearer. It sounds like N_INP,ss is based on the volume of air sampled during the 6 hourly period, so the minimum detectable INP concentration should be a function of the total air sampled during all the expansions at the given temperature in that period (it should be made clearer that CV should be calculated at each temperature interval). Therefore, should there be a metric like CV to calculate the minimum sampled volume of air needed to make a measurement significant? Using CV (essentially a metric for the variability over the 6 hourly period) seems a bit counter intuitive when the motivation of the dataset is to capture the variability in INP concentrations at the sub-diurnal/plume scale. That said, it seems like 0 ice crystals would lead to an undefined CV and therefore, would be removed from the dataset regardless of the thresholding.  Is that intended? Either way, CV should be explained better, especially as it is left up to the user to choose a threshold when using the dataset. In addition to showing  the amount of data removed for each campaign (Figure 3), it would be helpful to see how the CV thresholding impacts the data presented in Figure 2b, so it becomes clear which points would be lost and when.

As the temperature space for the different campaigns and even with an individual campaign varied, can you provide a recommendation on how to handle working with the 6 hourly 2 K binned data when there are different numbers of observations for each of the 2 K bins.

With the previous comment in mind, including the sampled volume of air in each interval (both time and T) would fix this and therefore, should be provided in the dataset to allow for volume-weighted averages rather than just "observation-number" means.

Minor comments:

Line 24-27: Not sure if including details on the ranges in aerosol ERF are really needed here.

When describing the different campaigns, consider omitting “but no accompanying paper is available” as it is quite repetitive and is not really needed

Line 78-79: What is the importance of the flow “by-passing” the instrument during the expansion? Consider omitting

Line 85: Does this mean that no aerosol are seen by the OPC? This could be rephrased to hydrometeor size distribution if it is excluding interstitial aerosol.

Line 92-94: It should be made clearer if each expansion yields one INP concentration at a prescribed temperature or if the INP concentration is reported at multiple temperatures during an expansion.

Line 100-105: I understand that these larger particles may be lost but if they do make it, are they misclassified as ice crystals due to their large size? After all, one large aerosol in an expansion would already result in an INP concentration of 0.5 L^-1.

Line 230: Here it is still not clear how these measurements are made. By using the minimum temperature and pressure does this mean that all ice crystal counts down to this temperature are considered as the cumulative number of INPs at the minimum temperature? If yes, and in line with the general comment, what volume is used for calculating a concentration? The total volume of air sampled by the OPC is not representative of the INP at the minimum temperature as only a very small fraction of the air sample is observed at that temperature. Is this somehow accounted for? This seems different than other immersion freezing techniques where a fixed volume of air is sampled and then the cumulative INP concentration is measured at a range of temperatures.

Line 254-257: This section could be streamlined as it is a bit bulky for motivating the “regular intervals” used in this dataset.

Line 260: Consider making it clearer that these expansions and resulting subset data are different than previous immersion freezing data sets in that each temperature is considered independently here. So if the subset data set covers e.g. 16 different temperatures, then the actual number of 6 hourly samples is closer to 2250 “6-hourly spectra” that cover the entire temperature range e.g. compared to drop freezing techniques. Either way, it would be worthwhile mentioning how much volume of air was sampled here as well to put the measurements into context.

Line 265-266: It is not clear why limN_INP is used here? Is this just meant to say above this limit? Please clarify what is meant here.

Line 266-268: It is also not clear why a moving average is needed for limN_INP? Is this not just dependent on the volume of air sampled in the single expansion and then summed over the 6 hourly period?

Figure 2b: It is great that the error bars are included here, but an error in INP concentration of an order of magnitude due to the OPC seems quite high?

Line 283-285: The way this is formulated is a bit misleading as it implies that the dataset includes orders of magnitude more INP information than previous studies. I’m not saying that this isn’t an amazing resource and dataset but an implied comparison like this should be based on the volume of air sampled between the measurements techniques conducted and not just based on the number of observations. This is especially the case when each N_INPss (or N_INP) is for a given temperature while immersion freezing data from a drop freezing technique covers the INP concentration over ~ 20 °C. If this dataset were to be divided up to produce an INP spectra, then it would also be on the order of 1000s of spectra.

Figure 4: The nonzero N_ss values in the corners of the subpanels are not immediately clear. Are these numbers the number of 6 hourly ss

Figure 5: It would also be nice to show the number of subset observations in each T interval.

Line 308-310: So if a single expansion during an interval led to no ice, the entire interval is set to 0? This seems inconsistent with the information shown in Figure 2, consider reformulating.

Editorial comments:

Using the ss subscript could be a bit confusing for the community who are used to having ss represent supersaturation. Consider using a different subscript.

Line 227: consider using hydrometeor size distribution (see comment above)

Line 265: ‘the’ is repeated twice before Level 1