The study introduces a new and easily applicable metric that can be used to determine the occurrence and estimate the strength of atmospheric NPF. This metric can be very useful as the extraction of NPF events is a laborious and time-consuming task with great uncertainties which come from either from the variable intensity of the events as well as personal perspective.

Major comments

While the metric can be useful in some cases there are some significant limitations with its design and application.

• The most important one is the complete exclusion of the growth of the particles in considering an event. According to the widely accepted definition of NPF events (Dal Maso et al., 2005) the new mode of particles formed should present signs of growth. This though cannot be determined with the metric presented as the size range chosen is too small for the growth of the newly formed particles to be confirmed. Additionally, without knowing whether there is growth of the particles, there is no information whether the NPF events extracted are atmospherically relevant (affect the local air quality, new particles to become CCN etc.) or if the particles are quickly lost by mechanisms such as coagulation or condensation. This very important aspect is not discussed at all within the manuscript.
• The metric (according to the Abstract) is considered to help in determining the occurrence of a NPF event. Looking at figure 7 though, there is about 50% uncertainty up until the percentile ranking of 70%. With only a fraction of the days surpassing that ranking (and still with significant uncertainty), how can the occurrence of an event be determined with confidence, especially for lower rankings? Once again the possible events will be considered as undefined.
• Hyytiala is a rural background site far from most of the sources of particles in the size range 2.5 - 5 nm. It is expected that the metric proposed will perform a lot better in such an environment. But for such an environment even a simple observation of either the total number concentration of particles or the particle number concentration in the range 2.5 - 5 nm (without looking at the background and active regions) would provide a similar outcome (in figure 4 the N_2.5-5nm seems rather stable for the non-event days). What is the added value of the specific metric? Please discuss.
• While discussed in the text, the possibility of such a metric being useful in more complex environments is unlikely due to the multiple sources of particles, which apart from that would also make it somehow more difficult to determine the background regions as well. This should be further discussed as examples of such cases are missing.
• Furthermore, Rönkkö et al., 2017 found that traffic can be a source of 3nm particles. Considering that the expected time of a NPF event can coincide in many cases with increased volumes of traffic, how will traffic derived particles be attributed as non-atmospherically formed in busy urban environments and carry on the analysis.

Minor comments

• Line 31. NPF events take place with very variable frequency around the sites studied so far. Please rephrase that as the way this is written now it implies that the frequency of the events is similar everywhere.
• Line 244. This needs to be further discussed and examples of ways to filter the factors mentioned should be given.

Rönkkö, T., Kuuluvainen, H., Karjalainen, P., Keskinen, J., Hillamo, R., Niemi, J. V., Pirjola, L., Timonen, H. J., Saarikoski, S., Saukko, E., Järvinen, A., Silvennoinen, H., Rostedt, A., Olin, M., Yli-Ojanperä, J., Nousiainen, P., Kousa, A., & Dal Maso, M. (2017). Traffic is a major source of atmospheric nanocluster aerosol. Porceedings of the National Academy of Sciences, (29), 7549–7554. https://doi.org/10.1073/pnas.1700830114

The authors present a new method for classifying new particle formation (NPF) based on a simple metric: 

the number concentration between diamters 2.5 and 5 nm. The new method may be useful when analyzing the

occurrence and intensity of NPF events and is definitely more easily automatisized than some of previously

presented methods. Prior to being publishable, several issues need to be addressed.

General comments/questions:

1. Why is the ability to predict NPF event frequency important? It is clear that atmospheric models need

submodels/parameterizations for NPF rates but it is not explained why event frequency is important and

how it could be used?

2. The presented method relies only on number concentrations in the interval 2.5 nm - 5 nm, while ealier

methods (e.g. Dal Maso et al., 2005) view early growth also as important. This raises the question: what

is NPF? How should it be defined? Is it the same as nucleation, or is it apparent nucleation (Kerminen

and Kulmala, 2002) at some larger size, which would also imply that the competition between growth and

scavenging is also important?

3. How do you know that this method works also at other sites, especially polluted ones, when it is

tested only in Hyytiälä? In the abstract you state that "The new method... is expected to serve as a

valuable tool.... at many types of environments" Having only this one site as a test case is, from my

point of view, the main weakness of the manuscipt. At least, such statements without proper justification,

should not be stated. In addition, on lines 54-55 the authors underrate the traditional methods based on

them not likely being generally applicable. At the same time the authors state that this new method likely

is.. based on analysis of one quite clean finnish site. ("My method is likely better than yours. I have

no actual proof, however...")

More detailed comments:

4. Line 97: How was the time interval 21:00-06:00 chosen? Would this be the window also for other sites?

5. Section 2.1.1 NPF "mode" fitting: What is the motivation behind these modes, i.e. why not just use

the deltaN values as boundaries? What then, if at some other site the distribution would be such that it

is not as nicely approximated by a couple of lognormal functions?

6. Line 152: What does "kernel inversion method" mean? Please explain with a few sentences or give

a refence.

7. Equation 2: The notation is confusing. It is stated that J_Dp is the formation rate at 3 nm, i.e. J_3.

But if Dp = 3, then N_Dp on the r.h.s. would be dN3/dt and GR_Dp would be Gr_3. Please be careful with

the notation.

8. Equation 3: The notation is again confusing. Usually when there is a sum, there is some increasing index.

Also, what is N_Dp2 ? Is it a pointwise size distribution density or the number concentration in some

size interval?

9. I don't understand the sentence on lines 178-179. Why is a median value for GR used and not a GR

for each individual day? And how similar is similar in the previous sentence?

10. Lines 222-223: Instead of "a continuous variable", do you actually mean the rolling median? Or do

I understand fig 8 somehow incorrectly?

11. Lines 263-165: Wasn't this recent finding already hypothesized by Kulmala, Pirjola and Mäkelä

in their Nature-paper in 2000 (thermodynamically stable clusters)?

12. Lines 269-272: Again the authors claim that this method is likely to be applicable to many other

sites also, without proper justification. Analysis with data from two sites of very different types 

would be much more convincing than Hyytiälä only.