The paper addresses two relevant scientific modelling questions that are relevant to GMD, namely the performance of CAMS derived CCN number concentrations with respect to in-situ ground-station observations and possible corrections to improve that performance. Both the evaluation against a large set of ground-station observations from varied environments, and the proposed correction are novel and present important advances for the field.

While the evaluation against in-situ observations is very valuable and well documented, the level of agreement between CAMS and observations is overstated as follows:

• Line 227 states that “simulated temporal variability agrees well for the majority of sites” (line 227), but  R is less than 0.5 for almost half the evaluated stations for monthly averages.
• Line 230 states that “CAMS Nccn reproduces observed variability very well on seasonal and annual time scale and is well suited for climatologies”, but the seasonal and annual agreement with in-situ observations are not directly assessed in the manuscript, only the 11 day and monthly agreement.
• Reproducing variability between Dry and Wet season evaluation is mentioned but has explicitly not been shown (line 236)
• Block et al 2024 is sited as evidence for good agreement (lines 233-234), but this uses only sites SGP, PVC, PGH, GRG, MAG, and only 2 out of these 5 sites have R>0.5

Regarding the method of comparison between CAMS and observations, CAMS derived NCCN are at 0UTC only, but the full diurnal cycle of in-situ observations is used (Section 2.3 Data Treatment). It is worth questioning whether diurnal cycles in observed NCCN lead to a bias here. The approach of longer temporal averaging taken here is not sufficient to remove such a bias. If hourly observations are available from the stations, it should be straightforward to test this.

The method for deriving the correction factor is well justified in terms of justifying an s-dependent k parameter, and in using empirical fits of the Nccn-s spectrum to derive this. Limitations in this method being unable to use the full 3-hourly CAMS reanalysis data or change the underlying size distribution are well justified. However, aspects of the method seem unjustified and sensitivities to the large uncertainties involved not fully investigated. Details of these are as follows:

• The authors clearly note that CAMS derived CCN at latitudes above 70N and 70S are problematic because of the lack of data assimilation at these latitudes. The inclusion of these stations in the derivation of the global correction factor therefore seems poorly justified.
• The k parameter is derived by grouping together median continental, coastal/polar and remote marine stations to find average factors for each environment. However, the variability of S_best within coastal/polar and continental environment groups is greater than between the different environments (Fig. 4). This suggests that the grouping is not valid for this purpose.
• The k factor is derived by averaging the median continental, coastal/polar and remote marine correction factors with equal weighting (line 333-335). While the authors are correct to try to account for the fact that the environmental distribution of stations with available observations does not accurately reflect the global distribution, there is no justification given for assuming that the global atmosphere is made of equal parts coast, ocean and continent.
• There are large uncertainties/variability in the empirical fits for the k-factor (Fig. 4). Some understanding of how this uncertainty affects the resulting Nccn and the bias reduction is needed.
• The PGH site is excluded from the derivation of k(s) because they lead to negative values. This is only noted in the caption for Fig. 6, and no physical justification is given for excluding these data. This is especially problematic given that PGH is the only site from southern Asia. That these data lead to an unphysical k(s) seems motivation to reevaluate aspects of the method used for this correction.

More minor comments are as follows:

• The abstract fails to state the result of the evaluation of CAMS Nccn.
• Line 218 “temporal presentability” - unclear terminology
• Section 2.2 Observed Data - While the number of stations and different environments included in the evaluation is high, and reflects well what is available, it is of note that available observations are still biased toward Europe and North America, with Africa and Australasia remaining particularly underrepresented. This is not a fault of this study, but it would be helpful to note this bias and its’ potential implications in the manuscript.
• The conclusions sections states both that there is no dependence of model performance on environment nor pollution but also that more polluted sites show a positive bias (lines 392-395). These are contradictory.
• Why do the R values for the daily comparisons with SGP, PVC and PGH differ from those presented in Block et al 2024 (see table below)?

              Daily                   11day                 month                Block2024 (daily)

SGP      0.069                  0.133                  0.709                  0.24

PVC       0.296                  (0.098)               (0.360)                0.66

PGH     0.532                  0.623                  (0.480)                0.41

• Line 219 typo “andcorrelations”
• Fig A3 legend hard to distinguish between station and environment specific
• Line 245 “CCN_O are significantly overestimated” – does this means observations>CAMS or other way round? Similar at line 265
• Line 247 “Most stations fall in the respective pollution regime category as defined by observations” unclear – I think the authors mean the pollution categorization from CAMS matches that from observations in most cases - needs to be stated more clearly.
• 6a Impossible to see which datapoints belong to the same station except at the edges, so impossible to see if k does decrease with s for each station as stated at line 320-323. Solid lines for environment specific medians not very visible in 6a either.
• ATT station is used as an example of how the method works (Fig. A3) with some acknowledgement that the global average approach does not improve the predicted Nccn for all stations (lines 344-352). However, ATT has s-best in the middle of the range of all stations, and therefore is likely a case where the global average works best. It would be more justifiable to illustrated also cases that work less well.
• Fig A1 – legend and caption unclear.

The study “Evaluation and improvement of CAMS-derived CCN number concentrations using in-situ measurements”, uses in-situ measurements of CCN concentration in different places over the world to evaluate a model-derived global datasets of CCN number concentration. They performed comprehensive analysis of the bias, and proposed an approach to improve this global CCN dataset. Given CCN is a critical link between aerosol and clouds but with very sparse direct observations in terms of both temporal and spatial coverage, and human activities are an important source of aerosol, a long-term global CCN dataset is of imperative and of great interest to atmospheric and climate communities. The manuscript is well written and structured as well. I would be happy to recommend publishing of this study, and provide my comments below to help further improve the article.

1. Data availability. Given the importance of and of wide interest of this improved global CCN dataset, please be clear about how to access this CCN dataset.
2. I would suggest a bit more discussion of the limitations of this study. For example, as shown in Fig.1 that observations in highly polluted megacities, such as Beijing and Delhi, are not included in the evaluation. This would lead to unquantified uncertainty in high pollution regimes.
3. When interpolate CCN and supersaturation relationship, linear interpolate applied in L132 but power law in later (L140). Better to keep they consistent.
4. The study has clearly stated that volcanic aerosols are not considered and organic matters are an important source of CCN. Given wildfire is one of the major natural source of aerosol, particularly for OM, could authors please elaborate that how wildfire emissions/aerosols are considered in this CCN estimate?
5. I feel L247-251 is unclear, please rephrase it. How does the regime category defined by the bias? When there is a conflict between bias and observation defined category, how should a site be classified?
6. Fig.6 Please provide units.
7. Since the Nccn is estimated based on aerosol species, I wonder if k(s) is also species dependent, and how sensitive (or uncertainty) your results is to this dependence?
8. L380 point-1. Although authors refer to some of previous studies about the evaluation of performance in time-variability, the discussion of time-variability (eg. seasonal variation) is superficial in this study itself. I would suggest not to call point-1 (L380) as specific focus of this study.