Please see me detailed comments in the attachment.

The study presents a new open-source Python library for ensemble estimation of geospatial earth system variables. The new library (GPEP) is based on an existing one (GMET) that is programmed in Fortran. The authors aim to increase the flexibility that GMET provides, increasing the amount of variables that can be analyzed, the number of spatial interpolation schemes and other important characteristics. They apply the library to three demonstration experiments where they compare their results to those provided by GMET. They conclude by remarking the advantages of the new library and some of its drawbacks.

The main contribution of this work is to have translated a model in FORTRAN to Python. FORTRAN is an old-school programming language, fast and expressive, that has been extensively used for numerical model programming and other intensive tasks. It is less popular than it used to be and compiling it may be complicated. On the other hand, Python is an interpreted language that does not require to compile its codes, is very portable and with a plethora of libraries around it that automatically creates synergies with every new library, like it could be the case with GPEP. Indeed, the authors mention the possibility to interoperate GPEP with Scikit-Learn, a Python-based machine learning library, which may give access to many machine learning and artificial intelligence algorithms. Translating GMET to Python may make the software more accessible to many young researchers, so I believe that this work constitutes an interesting contribution to GMD.

I have, nonetheless, some comments and concerns about the current incarnation of the document that I would like to present.

1. One of the main drawbacks of the new library, with respect to GMET, is its efficiency. It is a well-known drawback of Python, so it could have been expected. However, I wonder why, GMET being open source, the authors did not use the existing GMET FORTRAN code and wrap it using Cython, for instance, to get the best of both worlds. Many performant Python libraries, like SciPy, are just wrapping legacy FORTRAN code -for linear algebra, for instance-, making fast code available to Pythonistas. This approach would have saved time, since most of the routines were already programmed, and the effort would have been directed toward making the fast code accessible through Python. The total effort would have been smaller and the result equal on functional terms but better from a performance perspective.

2. I miss some manual, or at the very least a small tutorial, to start to work with the library. Jupyter notebook is one of the standard ways to provide tutorials for Python libraries, which helps the user to familiarize with how to work with a new library in a graphical environment. I have checked the test cases in Zenodo, but they are quite a dry way to approach a new library. On top of showing its performance, I believe more developed examples would be very appreciated. Indeed, I believe the manual would be especially important to learn how to use the library in conjunction with sklearn since library interaction is not always an easy topic to grasp.

3. A comment similar to the previous one, but related to the problem being solved. Existing users of GMET may already know the problem that the library solves, but new users could benefit from a small schematic definition of the problem. A small sketch or diagram could help the users to understand what problem the library solves and how it is solved. I have seen in the GitHub repo that there are many links to different papers, but a user, which may not necessarily be a scientist, may really appreciate a small theoretical introduction to the problem. In a sense, it would be making the library accessible for those outside academia that may use it to solve more practical problems.

    With this comment in mind, I would say that section 2 could be extended so that readers may get a better idea of the library's inner workings. I believe that papers should provide enough detail to be self-contained, and I am not sure that all the most relevant details have been included in the document.

Some minor comments:

Line 117, equation 1. I believe it should be two equations. It is just a problem with the organization of the equations.

Line 149: I think the 's' should be removed

This paper described a new Gridded Meteorological Ensemble Tool (GMET) for probabilistic hydro-meteorological estimations. The new tool was developed with Python, ensuring the improved flexibility and wider application ranges. This work would be a big contribution to the community for ensemble studies. However, there are a few concerns need to be addressed before the paper can be published.

Input data intensity & model performance evaluation: The method relies heavily on available gauge data, and authors use US area as shown studies in this paper. I wonder how the method perform when applied to data-sparse area? LOO cross-validation method might not be enough, but author can evaluate for instance different proportion of calibration/validation gauges, to check the change in model performance.

L152, one of the largest advantages of translating GMET to GPEP is the easy extension of python libraries. For instance, the supervised learning method, however, authors didn’t show how this could be operated in the current model or how results are differing compared to conventional methods.

Authors state that the spatial interpolation method could be different from investigated variables, however, details about the method selection and explanation are missing. I would like to know which method is more ideal for different basic variables and the reason.

L315-316. Why using the standard deviation of KGE is better for temperature?

Figure 6. Why the uncertainty range is so large?

Figure 8. the scale of the two maps should be unified.

The manuscript touches upon an interesting and important subject, the creation of ensemble gridded data sets. The manuscripts presents a FORTRAN-to-Python translated, yet also improved, software package. In addition, the manuscript shows a number of example applications of the software. The manuscript could be a welcome addition to what is currently available to the scientific community.

The manuscript is written in a clear style, is interesting to read, and provides well-chosen examples.

MAJOR COMMENTS

1) Given the importance of creating ensemble grids, which is reflected by the title and introduction, it would be desirable that the authors spend more time on explaining the way in which the software generates the ensemble members. Currently, the discussion of the ensemble creation methodology is not detailed enough.

2) In the examples, there is much attention for the predictive accuracy of the ensemble mean. However, there is less emphasis on the accuracy of the ensemble dispersion. Given that one of the main pros for using this software is that it can create an ensemble data set, I would suggest that the authors show the accuracy of the gridded ensemble dispersion, for example by showing rank histograms.

3) The repository is somewhat minimal. Please consider including some example scripts, possibly in a Jupyter Notebook. Since the paper is essentially a presentation of new software, the repository could have some more features to introduce the software to new users.

MINOR COMMENTS

The text is clear and well-written. However, the figures require some more attention.

Fig. 1: Please make clear in the figure caption what LWR1, LWR2 and RF are. Also, in the main text, discuss why LWR2 is so much slower thatn LWR1.

Fig. 2: Please make sure to show axis lables and physical units in all figures.

Fig. 3: Please make sure to show axis lables and physical units in all figures. The axis numbers on the x-axis are unreadable. Also, please make the color range of (c) and (d) the same.

Fig. 4: Please make sure to show axis lables and physical units in all figures.

Fig. 5: Please make sure to show axis lables and correct physical units in all figures. The axis numbers on the x-axis are unreadable.

Fig. 6: It is unclear what is shown here; is this for a specific location, or is at a spatial average for a specific polygon? Please explain in figure caption.

Fig. 7: Please make sure to show axis lables and physical units in all figures.

Fig. 8: Please make color range of (b) and (c) correspond.