General comments

This study uses Earth System Model output, in-situ observations, reanalysis data, back-trajectories, and machine learning to investigate the effects of recent precipitation on the aerosol size distribution at three locations. Specifically, the authors evaluate the models with respect to a hypothesis from the literature that the removal of large particles by precipitation can trigger new particle formation events at these sites. The study represents a novel and valuable contribution to a subject in the scope of ACP: aerosol-cloud interactions and new particle formation processes, the climate effects of which are notoriously uncertain due to the need for complex, unconstrained parameterisations in ESMs.

The authors make very interesting conclusions about new particle formation-precipitation interactions, which lead to important recommendations for future research. However, the results in this article are complex due to the fact that the authors have undertaken a very thorough analysis of the data available, (including several different variables, models, locations, seasons, data products and sensitivity tests). Hence, the article would benefit from some simplification, including more synthesis of how these multiple strands of evidence lead to specific conclusions about individual physical processes in ESMs. I also recommend more discussion of the sensitivity of the results to the use of different meteorological datasets. As such, I recommend that this article should be published following some revisions to address these points. Specific recommendations are given below.

Ruth Price

Specific comments

1. Different meteorological datasets: at the high-latitude stations, ERA5 precipitation data are used along HYSPLIT back trajectories to compute correlations in the observations. This differs from the approach in Khadir et al., where GDAS1 precipitation was used. Could the authors clarify why this change was made? There is some indication that results are sensitive to this choice (e.g. Figs. S9–S15, lines 394–395, 688–690). Given this sensitivity, it would be helpful to include a more explicit discussion about the robustness of results to the choice of meteorological dataset, particularly in the conclusions. For example, NorESM and EC-Earth appear to overestimate precipitation at ATTO; could this contribute to the overestimated particle formation and growth rates observed in these models (Figs. 5/6)? Similarly, while Section 2.5 argues that GDAS1 and ERA-Interim dynamics are sufficiently similar, it would be helpful to show a comparison over a subset of the data to validate that assumption.
2. Presentation of key result: the article highlights the inability of the models to capture CCN replenishment as a central result (e.g. abstract, lines 655–656), along with possible causes and implications of this finding for model processes. However, Section 3.2, where this is largely discussed, also includes extended treatment of NPF inhibition and particle removal processes. This mix reduces clarity. The structure of Section 3.2 (and therefore the article as a whole) would be improved if it focussed solely on CCN replenishment (i.e. positive correlations between precipitation and particles <50 nm) to make the key result clearer and more concise. For example, the paragraph beginning at line 376 contains only one sentence on the replenishment signal before shifting to NPF inhibition and offline oxidant fields, despite these topics being already discussed in Section 3.3 (e.g. beginning at line 455).
3. Section 3.3: this section would benefit a clearer structure to support the reader’s understanding of its contributions.
   • Section 3.3.1: detailed, speculative discussion in this section is sometimes difficult to follow. For example, lines 473-474 and subsequent discussion reference a nighttime/daytime difference in N10–30 vs precipitation at HYY, whereas it is difficult to see any such strong difference between the observed relationships at HYY in Figure 7b. Moreover, the language used to describe Figure 7 is at times somewhat imprecise (e.g. lines 469–470 “NorESM remains unchanged” could be changed to “the relationship between N10-30 and accumulated rainfall remains unchanged in NorESM”). Overall, since the section does not identify a key result (e.g. line 490: “It is hard to draw conclusions about model skill”), much of this content could be moved to the supplement, with only essential findings retained in the main text.
   • Figure 7: this figure itself is difficult to interpret because the overlapping lines are hard to distinguish; using separate subplots for “all” and “nighttime” could improve this.
   • Section 3.3.2: this section introduces seasonal data that had not been discussed earlier in the paper, adding considerable complexity. However, Figure 8 does not allow for individual seasonal relationships to be easily interpreted (it is also not clear why some sections of lines are faint). These issues make it difficult to follow how the conclusions at the end of Sections 3.3.2 are derived. A clearer, more structured presentation that links explicitly to earlier sections would help.
4. Temperature dependence of particle scavenging: section 3.4 contains the very interesting result that an observed temperature dependence of particle scavenging is not captured by models. However, similarly to comments above relating to section 3.3.2 and figure 8, it is very difficult to interpret Figure 9 due do the way data from different seasons are included but cannot be distinguished. Please consider revising Figures 8 and 9, perhaps limiting the data to just the seasons that are already presented in section 3.2.
5. XGBoost analysis: like section 3.3, this section presents a lot of new information without a clear link to the key results of the article. For example, while the authors state that the XGBoost analysis supports the correlation results, Figure 11a does not show an increase in small particles at longer time lags, which seems to contradict previous findings. It would be helpful to clarify this apparent discrepancy and articulate more clearly how the machine learning results complement the core conclusions.
6. Section 2/table 1: there are significant repetitions of the information in Table 1 and section 2.1. To make the article as a whole more concise, consider restricting the information to Table 1 and making only brief references in the main text, or further giving details given where necessary.
7. Supplement: the supplement currently contains a large number of figures, some of which are not referenced in the main body of the article. To make key information easier to find and comprehend, consider removing figures and/or information from the supplement that is not directly referred to in the article.

Technical corrections

• Line 3: et deposition should be wet deposition
• Line 49: cloud-born should be cloud-borne
• Line 120: arctic should be Arctic
• Table 1, first row and column: cloud-born
• Lines 347-348: “For ATTO, note that the satellite product used for the observations has fewer instances of low precipitation rate compared to the models, and has more higher values instead.” My interpretation of Fig 2 is that the TRMM product has many more instances of 0 values of precipitation rate than the models, and fewer instances both of low (0-0.05) and higher values, I struggle to see how Fig 2 shows that TRMM “has more higher values instead”. Please clarify this.
• Lines 348-349: “This is likely because to the input data for the observations at ATTO have a resolution…”
• Line 429: close bracket missing after (10-30 nm
• Lines 453-454: “Additionally, whatever relationship is observed, we see will be a combination” seems like it should be “whatever relationship is observed will be…” or “whatever relationship we see will be…”
• Figure 9 caption: suggest it is more logical if it reads “Figure 9. Same as Fig. 8a for cold versus warm conditions at the station separately” rather than “Figure 9. Same as Fig. 8b but for N100 and for cold versus warm conditions at the station separately”
• Line 674: “the smallest particlethe NPF”

This paper by Blichner et al looks to investigate how much of an impact precipitation can have on the CCN budget. Precipitation is typically seen as a CCN sink and so research into how precipitation can lead to new aerosols that can then grow and potentially become part of the CCN budget once again is important. They explore how precipitation can lead to good conditions for new particle formation and an increase in smaller aerosols due to downdrafts brought about by precipitation events. They explore how different processes affect the particle number size distribution (PNSD) after rainfall. This work is building on ideas suggested by Khadir et al (2023) and introduces three Earth system models to see how well they predict changes to PNSD along back trajectories from precipitation events.

They have selected three different models, nudged to ERA interim analyses, which use different in-cloud and below cloud processing schemes. They also look at how the models treat phase dependency, redistribution of aerosols and aerosol transport differently. This gives them a range of approaches to modelling aerosols and clouds which can then be attributed to different sensitivities that the models exhibit, compared to observations taken at three different sites (two high latitude sites and one tropical). They primarily explore the correlation between precipitation and number concentration and how this differs for different aerosol sizes, with different times away from precipitation event. They discuss their results process by process to gain an understanding of how each affects the aerosol number size distribution, back through time. They also use XGBoost, a machine learning algorithm that is good at predicting outcomes from a small number of input features. This is used to support the arguments their main methods suggest by looking at whether the same patterns seen in the correlation figures can be picked out from the predictions of these machine learnt models. They also do some feature analysis to help pick out potential correlations that shouldn't be attributed to precipitation.

The main results from this paper show that models are underpredicting new particle formation and growth after precipitation, compared to observations. They perform better at the ATTO site, where the increase in small particles is linked to downdrafts instead. It is unclear however whether improved model performance is for the right reasons, with some existing hypotheses impossible within the models available. They also see further verification of previous work that showed particle growth at the ATTO site was much too high in several of the models. XGBoost presented an interesting new way to further analyse these correlations and potentially helped to identify questionable correlations seen in other methods.

This paper is written very well with a clear narrative path throughout the sections, introducing all the important concepts as well as previous work that this is building upon. This paper will fit within the current literature around this topic, exploring some new novel results that provide further understanding of how models perform when looking at the CCN budget after precipitation.  The conclusions are clear and show scope for future research, outside of this paper. The title, abstract and results reflect the work that has been done. I believe this paper should be published, once minor comments (outlined below) have been addressed.

General comments:

Thought 1 (Introduction): The discussion of the paper is primarily on a weather timescale (4 days) or so however the paper starts by talking about the impact of aerosols on climate and human health. I know the general discussion after the first few lines is in aerosol dynamics which are at a weather timescale but I think it would be good to introduce aerosol impacts on weather timescales in these first few lines.

Thought 2 (Introduction into Methods): HYSPLIT and GDAS1 are introduced in both sections without any references to the specific model/method used with regards to HYSPLIT and no reference for GDAS1. I think a more substantial introduction and explanation of these two things would be appropriate as well as a reference to the GDAS1 dataset and HYSPLIT.

Thought 3 (Methods): I think that the models are well outlined however I think the table for comparison is too complicated and a lot of it is repeated in the individual models sections. I struggle, using the table and the individual sections to fully grasp the differences between the models. I believe a simpler table that just outlines each part and an additional section at the end, comparing the models in terms of schemes and complexity (linked back to the introduction when wet deposition scheme complexity was discussed) would be useful to support later discussions of model performance.

Thought 4 (results): I think that the figures within figures could benefit from labelling with a), b), c) etc so that when they are referred to, they can be referred to as 2b) and so on to make it clear which figure is being referenced. Also some of the figures have faint lines for some reason (see figure 8). Figure 7 could do with being split into two plots (one for nighttime and one for all) as it is hard to read the results. Overall the spacing between the figures and size of the figures could be tweaked to remove some of the white space too.

Specific comments:

[L3] “Wet” rather than “et”

[L5] can do general circulation models (GCM) in brackets here rather than later

[L10] new particle formation (NPF) in brackets here rather than later

[F1L1] “Illustration” not “Illustratration”

[F1L6] “points” rather than “point”

[F1L11] Reference Khadir et al (2023) at this point as this plot type is taken from that paper

[L50] “Borne” not “born”

[L62] A reference to a paper explaining why precipitation formed via ice formation is expected to scavenge less than liquid precipitation formation

[L120] A suitable reference for each of the different sites related to NPF or precipitation

[L126] This second hypothesis contradicts the one by Zhu et al. Which one is more plausible or are both plausible in different scenarios? A further comment on this contradiction would be beneficial.

[L128] Linked to F1L11 but this should be in figure caption aswell/alternatively

[L131] “,for example,”

[L171, L204, L244] Why these relaxation times? Further detail would be useful.

[L191] “area” not “are”

[L209] “scavenging” to “scavenging”

[L229] “includes” not “include”

[T2L4] “Jun-Aug” instead of “Jun-Jul”

[T2L11-13] Might read better if you swap Dry and Dry-to-Wet around as then goes through the year.

[L351] swap S1 and S2 around

[L418] “at least in” not “just at least”

[L427] “S7))” not just “S7)”

[L429] “10-30nm)” not “10-30nm”

[L468] remove anyway

[F7] Night needs to clearly be a star in the legend

[L496] “one can see here” not “one can here see”

[L497] need something between “stations, the models” like a “whereas” or change the sentence structure

[F8] Could you differentiate between the seasons? Also address the white out effect of some of the lines

[L526] “Fig S22” not “Figs S22”

[L530] “to be more” not “to more”

[L581] struggles to reproduce two of the models, not NorESM and observations instead of current wording

[L586] “XGBoost” not “XGboost”

[L618] “72-80h” not “72-8h”

[L658] “to be a lower” not “to be lower”

[L673] this needs fixing to make sense (remove the NPF?)

[L682] “much too large” to “too many”

[L692] “Is” not “are”

[L699] Need to make it more clear that XGBoost agrees with overall correlations but attributes different reasons – some more explanation would be useful