The manuscript presents observational gridded datasets over Greece, covering daily total precipitation and daily mean, maximum, and minimum temperatures. The authors have applied quality control and homogenization procedures to the input data. They also examined the use of different statistical methods for spatial interpolation. In addition, they incorporated numerical model output to address gaps in the observational network, which is relevant given the complex topography of the region. The datasets have been evaluated through cross-validation using independent observations and compared with existing gridded products available for the same area. The figures included in the paper are informative and clearly presented. The results support the conclusions drawn by the authors.

There are a few points that may require clarification or expansion. First, the manuscript does not include a sensitivity analysis regarding the use of WRF model output for a year other than 1999. While this analysis may not be essential, the authors could expand the discussion around lines 139–141. For example, they might consider whether a regional reanalysis product, such as CERRA, could have been used, or if WRF simulations were tested for other years. Second, certain methodological choices could be described in more detail. This is outlined in the comments below.

Overall recommendation: The study provides a useful dataset and analysis for the region. I recommend publication after the authors have addressed the comments that follow.

Comments:

1. Regarding the gridding of temperature data: It is likely that the station locations, your grid, and the CHELSA grid differ in elevation for the same geographic points. This is expected, but it is unclear how these differences were handled during the spatial analysis and subsequent comparisons. Did you interpolate all datasets onto a common grid before comparison? This point could be clarified in Sections 3.3 and 3.4. Also, discussing elevation differences may help with the interpretation of results in Section 4.2.1. Please consider revising that section accordingly.

2. The choice of FRK as the final spatial analysis method is only briefly mentioned in lines 289–291. This decision is important and could be stated earlier and more clearly. For example, it could be introduced in the abstract (e.g., after “against withheld observational data,” add a sentence about the method used). Additionally, you could move the relevant lines to the beginning of Section 4.1, rather than introducing FRK in the section discussing temperature results. Consider also whether the conclusion should briefly mention that FRK performed best among the methods evaluated. It may also be useful to explain why a single method (FRK) was chosen for both temperature and precipitation, despite indications that SVM performed well for precipitation. A short explanation of the reasoning behind this choice could be helpful.

3. Lines 42–44: The phrase “model-generated” could be clarified by adding that these were generated using statistical methods, to distinguish them from output from dynamical models.

4. Section 2.1: Please specify the definition of a “day” for each variable (e.g., whether it spans from 00 UTC to 24 UTC). Even if this follows a standard convention, it should be stated explicitly.

5. Line 165: Consider whether this line should be part of the previous paragraph, as the new line may not be necessary.

MAJOR COMMENTS

The manuscript presents a comprehensive dataset derived from regional climate downscaling addressed to the Greek territory using advanced machine learning techniques. The primary objectives are to enhance the spatial resolution over a 39-year period and to improve the representation of daily temperature and precipitation fields across the complex geography of continental Greece and its islands. The authors employ a hybrid approach combining geostatistical interpolation and statistical downscaling methods with atmospheric modelling, validated against observational datasets. Four methods (FRK, GAM, KNN, and SVM) were evaluated, with FRK ultimately chosen. Evaluation against CHELSA-W5E5 and withheld station data supports the improved spatial accuracy and bias reduction of CLIMADAT-GRid, especially in mountainous regions. The validation strategy, multiple error metrics, and comparison with an established product (CHELSA-W5E5) strengthen the study’s robustness. The outcomes suggest improved accuracy in temperature and precipitation at regional scales, supported by the analysis of suitable climate indicators. However, while the objectives are met mainly, certain aspects require further clarification to substantiate the claims thoroughly.

The introduction chapter might discuss similar datasets produced for the same purpose. This would help clarify the expectations surrounding this exercise, including its benefits, drawbacks, and potential challenges. The datasets E-OBS and IBERIA01 referenced by the authors in Chapter 3 could serve as a starting point. In the subsequent chapter, the datasets are delineated without any preceding explanation or introduction regarding their presentation, which may lead to confusion for the readers. For example, WRF parachutes in Section 2.3 without any justification or preangle (abstract is not part of the manuscript), leading readers to assume that the model will be employed for temperature and precipitation analysis. I recommend incorporating an introductory paragraph between items 2 and 2.1 to bridge this gap, as was implemented in Chapter 3.

Continuing with the discussion on WRF, only during Section 3.2 (the second section before the results), the readers are informed that the atmospheric model was utilised exclusively for the temperature field, which may reduce the audience rejection regarding precipitation. The decision to employ solely one year to represent the overall study period is somewhat contentious, as a significant amount of variability is forfeited. Nevertheless, this approach is permissible, given that the model ultimately functions as a spatial interpolator driven by physical laws, subsequently manipulated to incorporate the seasonal and interannual variations delineated by the observational data. It may be beneficial for the authors to include a map illustrating the participation of the observational data at each grid point, as it could mitigate the discussion concerning the employment of the atmospheric model to cover regions lacking a station. Furthermore, this addition would facilitate the analysis of the specific areas to which the results can be attributed through the model. It would be essential to formally present the domain's limits, since they may circumstantially restrict the representation of some large-scale atmospheric phenomena transiting from the boundary condition. Furthermore, the authors reference several studies on WRF applications in Greece but do not elaborate on their findings or the model calibration expressed by the selected set of physical parameterisations. This potentially makes the WRF application quite questionable for these purposes.

The description of geostatistical interpolation and ML methods, lacking detail and overly restricted to their respective R packages citation, is both awkward and limiting (Sect. 3.1 and 3.2). This narrow focus obscures the wide variety of options and parameters defined in each method and, as a result, hinders and prevents the proper reproducibility of the study. Thus, presenting these details in the manuscript is crucial. Concerning the metrics employed in the evaluation, specifically RMSE, MAE, and KGE, it is noteworthy that the most fundamental among them, Bias, was not considered. Relying solely on RMSE and MAE does not allow assessing whether the method underestimates or overestimates the observed values. This significantly impacts the analysis of temperature fields, particularly precipitation fields. Therefore, including this metric is also mandatory.

The results chapter is well-structured, and its figures and tables are clear, readable, and sufficient to justify the main findings. Nevertheless, it is also true that omitting some information and making certain choices in preparing the outputs reduces their respective impacts. For example, why was Figure 4 presented only for 2016 and not for the 10 years of validation (2010-2019)? Besides lacking a direct relation to Table 3, this leads readers to believe that the differences pointed out by the authors are limited and specific to that year and may not be as pronounced in an overall analysis. Furthermore, including a fifth panel (e) with the map that highlights only the station locations using colours representing the observed values on the same scale as precipitation ​​would significantly enhance this analysis and assist in determining which method was definitively superior. Keep in mind that spatial variability does not imply a better result.

Another relevant point regards the sentence “Despite this, CHELSA underestimates the observed number of SU by about 10 days/yr, while CLIMADAT-GRid closely aligns with the observed values.” (L319-320). Which results substantiate this conclusion? It may be pertinent to present and discuss the evaluation metrics (Bias, RMSE, MAE, KGE) for both CLIMADAT-GRid and CHELSA throughout the entire study period. Only a brief and vague text in L350-354 may not be sufficient. Finally, the conclusions chapter effectively fulfils its intended role, although it does not provide commentary on an essential aspect of the work regarding the various geostatistical and machine learning techniques employed in developing the temperature and precipitation datasets. This oversight may be attributed to the insufficient detail in the preceding chapters. Incorporating these elements, whether in the methodology or the conclusions, would substantially enhance the manuscript's value. Nonetheless, it is essential to underscore that the analyses provided are devoid of any fallacies or significant flaws and, in any case, compromise the integrity of the study. It is simply a matter of refinement.

Given the innovative approach and the potential contributions to regional climate, I recommend acceptance upon a comprehensive review of major comments. The paper presents a high-quality, methodologically sound dataset likely to be of great use in regional climate research, impact modelling, and policy work in Greece. The authors must tackle the previously mentioned concerns to enhance their transparency, reproducibility, and broader significance. Addressing these issues will fortify the manuscript and increase its contribution to the scientific community.

MINOR COMMENTS

L23, L26      CHELSA still CHELSA-W5E5 up to this point.

L31             The phrase “… are becoming…” requires modification. This citation originates from 2012. Currently, it represents a prevailing reality.

L32-33        Add a comma in “(Herrera et al. 2012)”.

L55             Remove the E-OBS citation that was previously introduced in L41.

L63             Citation missing for “MeteoSerbia1km”.

L73             The acronym "CLIMADAT-GRid" is used without prior definition. Please define it upon first use.

Fig 1           Figures 1 and 2 are redundant; only Figure 2 is sufficient if it replaces Figure 1. Furthermore, the blue colouration on terrain elevation maps is typically attributed to regions situated below the mean sea level (h<0). Consequently, it is advisable to redefine the scale to initiate with green tones.

L120           Citing Skamarock et al. (2019) may be sufficient.

L136           Remove the term “approximately” once ERA5 has a precise horizontal resolution of 0.25º. In this case, the approximation regards the resolution in km, which varies from 25 to 31 km, roughly estimated at 28 km.

L137           Following the standard presented in the manuscript, replace “USGS (United States Geological Survey) (Slater et al., 2011)” with “United States Geological Survey (USGS, Slater et al., 2011)”.

L137           Following the standard presented in the manuscript, replace “CORINE (Coordination of Information on the Environment) database (2010)” with “Coordination of Information on the Environment (CORINE, CLMS 2018)” if the authors have used the latest version (https://doi.org/10.2909/960998c1-1870-4e82-8051-6485205ebbac). Additionally, this citation may be included in the references.

L153           The acronym "GMTED2010" is used without prior definition. Please define it upon first use (Figure’s caption doesn’t count).

L202           Remove the comma in “Papa and Koutroulis (2025,)”.

L205           Replace “Climate Change Detection and Indices (ETCCDI) (Zhang et al. 2011).” with “Climate Change Detection and Indices (ETCCDI, Zhang et al., 2011).”.

Fig 2           In addition to the aforementioned comments regarding Figure 1, it is advisable to change the colour used for the markers on the evaluation stations, as they tend to blend with the background.

L217           Replace “The values of the root mean square error (RMSE), the mean absolute error (MAE) and the KGEs…” with “The values of RMSE, MAE and the KGEs…”, once they were defined previously.

Fig 8,9,11    Since the objective of these figures is to compare the temperature and precipitation fields of two different datasets, wouldn't it be better to present the difference between them instead of the entire fields? This way, the differences pointed out by the authors would be clearer. Furthermore, it would enable the presentation of TN, TG, and TX in the same figure without losing quality. That is, Figures 8 and 9 would be merged, with the addition of TG, which was omitted without explanation.

General      The recursive use of “hereafter” is inappropriate in most instances. Typically, this expression is employed to redefine a name or acronym. Only in L199 does it appear to be correctly utilised to redefine the acronym CHELSA-W5E5 as CHELSA.

General      In the scientific literature on climate and meteorology, the prevailing terminology for the temporal aggregation of precipitation over a day is “daily accumulated precipitation” or simply “daily precipitation”. Although the procedure is referred to as the precipitation sum, its use can lead to different interpretations.