General comments: The author presents a newly developed gridded, high-resolution climate dataset comprised of daily, monthly and yearly precipitation and temperature for Mexico that they have developed using stations data and Kriging with External Drift on a local neighborhood (KEDl) interpolation. The study presents a new dataset that can be very useful in understanding different aspects of climate change and its impacts at relatively high spatial and temporal resolution across Mexico.

This dataset appears to be a major improvement over existing datasets for Mexico. The station counts for the 1999-2020 period are much higher than in CRU and GPCC datasets; it would be helpful to demonstrate that they are higher than the Daymet dataset. We conducted a visual comparison of the MexiHiRes January and July 1981-2020 normals to the WorldClim, CHELSA, Daymet, and US PRISM products, and found that the MexiHiRes surfaces were relatively free of artefacts and appeared to be more credible than WorldClim, CHELSA, and Daymet, and generally more compatible with the adjacent US PRISM normals.

Our main concerns relate to the filtering of the original datasets—specifically, the criteria used, the quality of the input data, and the details of the interpolation method. The data and methods section needs substantial revision. Clear explanations, potentially a flowchart, and a discussion of how each step in the process might influence the results would be highly beneficial. The validation and comparison presented are limited and only applied to specific extreme cases. The manuscript lacks discussion of potential limitations of the datasets. It is not sufficiently demonstrated that the newly generated datasets are better than existing alternatives, despite multiple claims by the author, though we are confident this can be done with more convincing analyses.

Although this paper requires major revisions, it appears that this is an exciting and necessary improvement to climate data available for Mexico, and we commend the author for the effort.

Major comments:

• Description of method: The author mentions in multiple instances that this is better data. However, the methods is not clear on the process of interpolation of stations data. How the observation data, how the interpolation works, what are benefits or limitations of the method they choose and why they chose the method need to be clearly explained.
• Validation and comparison: The intercomparison with other datasets wasn’t convincing. The author needs to make a more comprehensive comparison of their new dataset to other datasets, not only for selected extreme events.  Further, they have to present how their data represents climate differently than other datasets and why it is so. I also suggest that the author discusses their methodological and data limitations.
• Climate normals: The focus of this paper is on the daily time series. However, the climate normals are also an important contribution and deserve some description, validation, and intercomparison.
• Content Refinement: The Introduction is ambiguous and there are several instances when concise refinement of the writing is required. They have to present the need for this work, and how this dataset will be better than existing datasets. Overall writing can be more concise.

Other comments:

In the comments below. L indicate line number, * indicate major comments.

• L3: remove terms like largest , highest etc and provide specific values like what temporal coverage, especially in abstract.
• *L10-17: not clear if the author is just presenting these values or saying that these values are more realistic than values from other datasets. I am not clear what the author is trying to say by saying ‘a summary that was not available before’. It is available from other datasets like ERA5land. Reliability of the values can be different, but it is available.
• L12-17: for which period are these values true, 1951-2020? specify.
• L18-19: generic
• L39: define for the first use
• L46-48: redundant
• L46 -48: superfluous. This whole paragraph can be summarized to few sentences, much of the details may not be relevant to the paper
• *L63 (whole paragraph): I suggest that the author create a table with name of dataset, resolution ( spatial , temporal), data period, region and citation . such table will give this exact information from this and previous paragraphs and will be more succinct
• L75: “monthly surfaces of precipitation, Tmin and Tmax for the 1910–2009 period (i.e. 12 surfaces in total)”. If these were normals, wouldn’t this be 36 surfaces? But also, the term “monthly surfaces” is ambiguous about whether it is a monthly gridded time series or climatological average/normal. As suggested, and table would be more succinct and explicit.
• *L80: what about Daymet / ERA5 or other regional and  global datasets that covers Mexico and available at daily resolution  but may not be as same spatial resolution as this dataset ? However,  What problem their spatial resolution creates that this MexHiresClimDB will better address ? Need to explain why this work is important and what value it adds in more detail.
• L93: use numeric for XXth
• L95: what is the link between the description of study area and the datasets the author developed ?
• L97-99: move this paragraph after the data description
• L100: what aforementioned refer to ? Provide the name of dataset. Throughout the paper try to use ‘aforementioned’ less often as it creates confusion
• L100-101 : rewrite: how many stations originally were there , how many were used ? what are the criteria to select the stations
• L108: what might have caused less number of stations after 2021 ?
• *Figure 2. The station coverage in this dataset is much better than in the CRUts and GPCC datasets, which report a dramatic (>90%) decline in station coverage after 1998. This is a compelling advantage of the MexiHiRes product and it would be useful to highlight it.
• *L110 onwards to paragraph: need more description of the process and how this method works. For example, what happens to temperature lapse rate and what value is used ? A flowchart with all the steps involved from data collection to final output will be useful.
• *L125-128: can the author show how sensitive their used values are ?
• L129-133: can be more succinct
• L134: recommend re-expressing 1.227×10⁶ as 28 months so readers don’t have to do the math. It’s a lot of computation!
• L129: Is 26GB minimum requirement? not clear
• L146: “if R2=0.80, then the model explains 80% of the variability”. Reword: it explains 80% of variance, not variability. since variance is calculated from squared deviations, this interpretation will overestimate the amount of variability/dispersion that is explained.
• L170: is it value or variable, be specific.
• L172: it’s true that MAE for raw precipitation (in mm) is meaningless, since an error of 50mm has a different significance in a wet vs dry climate. however, MAE is meaningful if precipitation is log-transformed prior to analysis of error. By the same token, I would expect that the other metrics (R-square, COE, and IOA) are confounded by raw precipitation values in mm; wouldn’t they primarily represent wetter regions—and wet anomalies—where absolute errors (in mm) are larger.
• *L200-239: I suggest the author move the description of these datasets to introduction and summarize there. Here they directly present the results of comparison.
• *L290-291: This is not well justified statement but based on selected extreme event and a single coefficient.   How this statement compares with what they presented in figure 5. I recommend performing relative Root mean Square Error to explore relative goodness of each dataset for both temperature and precipitation
• Fig 7. The panels are labelled a,b,c,b,c.
• Fig 7. It is very hard to see the distributions at typical printing size. Recommend altering the color scheme to better stand out against the background.
• L311: not clear what the author is trying to say here
• *L325-326: similar comments for precipitation and temperature. I suggest the author adds average comparison as well as relative comparison among the datasets to provide better perception on how their new dataset is better than existing datasets, if it is.
• L328: no need to write the full form again, be consistent
• L330: “it is not possible to obtain cross-validation values for L15 or Daymet”. This is a reason to use some other metric for intercomparison.
• L335: “neither L15 nor Daymet were capable of showing the temperature extremes that were obtained through the MexHiResClimDB”. This is not apparent from figure 9. The maximum values seem similar to L15.
• Figure 9: it is not clear what the author is trying to illustrate from these plots. It is already clear that their data has longer temporal coverage across Mexico than other mentioned datasets. It does not provide information on how better their dataset is compared to others.
• Figure 11: describe what climate stripes bars represent and what  additional information they provide other than the anomaly plots. Otherwise remove
• L393-L395: this has only been demonstrated for a few specific events, so the statement should likely be qualified as such.

The author develops a complete daily gridded dataset of precipitation and temperature for Mexico at a very high resolution considering the extension of the spatial domain. The research is mostly correct; however, I have some concerns about the method used to develop the daily grid and the validation process.

The choice of an interpolation method is not easy, especially for precipitation, since it can yield very different results depending on the method and the parameters. The KED is a trustworthy option when used for a single timestep, for example to show the daily precipitation/temperature in one day/event (as shown in the examples of extreme events). However, when you apply the same method for a long-term climate time series without further corrections, you are introducing temporal biases that can lead to unwanted inhomogeneities both in temporal and spatial dimensions. This is not new and it is a basic assumption in gridded datasets creation as shown in the wide (not cited) scientific literature (e.g. https://doi.org/10.1002/wat2.1555, https://doi.org/10.1002/joc.1322,https://doi.org/10.1029/2008JD010100, https://doi.org/10.1559/152304085783914686 ). I recommend the author to make a deeper review on the requirements for creating reliable grids, starting by some of the datasets that are cited in the article. The actual problem with creating a single grid for each day, independently from the previous and following, is that the number and location of neighboring observations change with time, and that lead to biases that have a significant impact in, for example the analysis of trends or even in the aggregation at coarser temporal scales. In this regard, my second main concern is the validation approach.

It is ok to check the errors by series with the proposed statistical tests, however, you are comparing the complete series of observations with their corresponding predictions without considering the number of missing values, the differences by elevation ranges, by months, seasons, etc. It is difficult to see biases that are useful to interpret how reliable are the results. In addition, the kriging process usually provides a variance dataset, measuring the uncertainty associated with the prediction at every specific location. It would be useful to see an associated gridded dataset with the error/uncertainty for each day and variable, as done in many other datasets, to account for the reliability of each prediction and let the user decide how to use the information.

Here are the minor and specific comments, line by line:

Introduction:

L30: “along the migratory route of the Monarch butterfly” Maybe this is too specific.

L44: Terraclimate is regularly updated and now it is available until 2024.

L66: For CONUS, I think that PRISM deserves to be mentioned since it was one of the first and still one of the more reliable gridded datasets (https://prism.oregonstate.edu)

Methods:

L101-102: how many stations was the final number?

L108: regarding the outliers, wasn't any additional quality control performed? There's a lot of scientific literature on this.

L108-109: how many stations were discarded?

L110: As mentioned before about the use of KED independently for each day, it can generate further problems in long-term trends and temporal aggregations. Also, why 30 nearest stations and a 140km radius? The number of observations can greatly vary under these conditions. Lastly, was the internal coherence of temperature (TMAX>TMIN) checked after the interpolation, for each day? The above problems are especially important if a proper quality control was not performed to control the spatial coherence of the data (and I read nothing in this regard).

L120-121: Is the code publicily available?

Validation:

I expected a more complete validation since here only daily data considering the whole series was checked. For example, how the interpolation worked at different elevation ranges? or in different months? Did the method correctly predict the number of dry/wet days? Are monthly (or other) averages and standard deviation fit between predicted and observed values? These are the basic checks for any gridded dataset.

Figure 5: I am not sure how to interpret these graphics since, for example, R2 needs complete series of predictions and observations to be compared but here you have one value per day/month/year

L242: why not comparing monthly or annual aggregates? or even trends? that would be more useful than comparing extreme events, which are not common (by definition) and the users may need a more regular use of the dataset.

L274-277: This comparison is not fair since you’re comparing predictions with observations but only in the case of your dataset you know that the observation does not participate in the interpolation, but not in the rest of datasets. Furthermore, not all of them were built with the same observations, so it is hard to justify better results on your dataset.

L298: what this function does?

L300: what was the threshold for considering a dry day? 0 mm / 0.1 mm / 0.001 mm?

L309: I dont see tendency in that table

L326-327: again, this is not a validation, just a comparison with other datasets that were not constructed with the same procedure. The only validation must be with the observations.

L332-335: a visual comparison does not guarantee a correct validation.

L347: Fig 10 shows absolute values, but this is not trends. If you want to show trends you should calculate some statistics (Mann Kendall, Sen’s slope) with their corresponding reliability value (p-value).

L350-351: this is not an acceptable way to indicate that there is a trend. Without statistical validation, this complete section must be removed.