ISARD (v1.0) : A Reproducible Geostatistical Framework for Daily Precipitation Ensemble in Mountainous Terrain

Dura, Valentin; Evin, Guillaume; Favre, Anne-Catherine; Penot, David

doi:10.5194/egusphere-2025-5679

Preprints

https://doi.org/10.5194/egusphere-2025-5679

Preprints

25 Nov 2025

| 25 Nov 2025

ISARD (v1.0) : A Reproducible Geostatistical Framework for Daily Precipitation Ensemble in Mountainous Terrain

Valentin Dura, Guillaume Evin, Anne-Catherine Favre, and David Penot

Abstract. Gridded precipitation datasets are essential for hydrological and climate applications. However, commonly used products suffer from systematic biases such as seasonal total underestimations in mountainous regions and excessive smoothing of the spatial variability of extremes. Here, we present a reproducible workflow for generating a daily precipitation ensemble, conditioned on rain gauges, at 1 km resolution for mountainous regions. The approach leverages climatological information and spatial variability from Convection-Permitting Regional Climate Model (CP-RCM) simulations. The workflow corrects raingauge undercatch, incorporates CP-RCM-based climatology to improve seasonal totals, and estimates anisotropic variograms from CP-RCM daily fields to capture directional precipitation structures. Finally, Sequential Trans-Gaussian Simulations generate the daily ensemble of 100 members. We evaluate commonly used gridded precipitation products and the proposed approach using independent evaluation data, including in-situ measurements in mountainous areas (snow water equivalent, glacier mass balances, streamflow), regional catchment-scale water balance models, and hydrological models. Results demonstrate that our framework outperforms deterministic gridded products. First, it more accurately captures seasonal totals in highaltitude Snow Water Equivalent (SWE) and glacier observations, and reproduces both seasonal precipitation amounts and their interannual variability. Second, the daily ensemble captures fine-scale spatial variability and quantifies interpolation uncertainty, improving flood hydrological modelling. The workflow is fully reproducible via open-source code, transferable to regions with sparse rain-gauge networks or limited radar coverage. Beyond precipitation, it is adaptable to other climate variables simulated by weather models.

Received: 15 Nov 2025 – Discussion started: 25 Nov 2025

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Valentin Dura, Guillaume Evin, Anne-Catherine Favre, and David Penot

Status: final response (author comments only)

RC1:
'Comment on egusphere-2025-5679', Anonymous Referee #1, 21 Dec 2025
This study presents ISARD v1.0, a framework designed to generate daily precipitation datasets at 1 km resolution by integrating rain gauge correction, convection-permitting regional climate model (CP-RCM) simulations, and conditional geostatistical simulations. The topic is relevant to the scope of Geoscientific Model Development. However, the manuscript currently suffers from a number of issues that substantially hinder its readability and scientific clarity. These include typographical errors, missing or unclear figure captions, and, more importantly, insufficiently explained and sometimes confusing methodological descriptions. As a result, the framework is difficult to follow, even for readers with a general background in geoscientific modeling, and would require multiple rounds of careful reading to understand. I therefore recommend major revision. The authors should substantially reorganize and clarify the manuscript, especially the methodological sections, before it can be considered for publication.
Major comments
The abstract provides a detailed description of the proposed framework, but it lacks a concise and quantitative summary of the main intercomparison results. To improve its balance and informativeness, the abstract should also report key numerical outcomes (e.g., relative improvements or performance metrics) that summarize the advantages of ISARD over existing products. A reorganization of the abstract is recommended.

As a reader who is not a specialist in this specific field, I found it difficult to understand how the proposed framework actually works and what its main novelty is. The current workflow appears rather coarse and does not clearly reflect the role of the different reanalysis and model datasets used in the study. In particular, the specific functions of these datasets within the framework remain unclear. Moreover, the advantages of ISARD relative to existing approaches are not sufficiently highlighted from a methodological perspective. I strongly suggest that the authors reorganize and substantially expand the methodology section, explicitly linking each step of the workflow to the corresponding equations and clearly explaining how the different datasets contribute to the final product. This would greatly help readers understand both how the framework operates and why it represents an improvement over previous methods.

The study applies correction factors of 1.6 and 1.0 for solid and liquid precipitation, respectively. While this choice is briefly mentioned, the physical reasoning behind these values and the thresholds used for rain–snow partitioning are not sufficiently discussed. Simply stating that “the sensitivity to this choice was not evaluated in this study” is not satisfactory, given the importance of this assumption for the results. I believe that a simple sensitivity analysis with respect to the snowfall correction factor and/or rain–snow partitioning thresholds is necessary, or at least a more detailed justification and discussion of the potential impacts of this assumption should be provided.

In Section 3.1.2, Random Forest (RF) is used to interpolate the log-ratio term logR. However, key information about the RF implementation is missing. It is unclear how the RF model is trained and applied, which hyperparameters are used, and how predictions are generated. In addition, the notation in Eq. (2) is ambiguous. Please clarify whether the function f represents an RF regressor or classifier. A more detailed and transparent description of the RF setup is required to ensure reproducibility.

Minor comments
There are unnecessary empty lines before Eq. (2) and Eq. (5). Please check the formatting throughout the manuscript.

What is “subsectionSnow datasets”?

Line 126: (Gupta et al., 2009, KGE,) to (KGE) (Gupta et al., 2009)

Line 166: What is B(x,c)?

A summary table of the gridded precipitation products and models used in this study would be helpful. The table should at least highlight their main differences and roles within the manuscript.

Sections 2.2.4 and 2.2.5 describe instruments and glacier mass balance measurements, which are fundamentally different from reanalysis datasets such as ERA5-Land and CERRA-Land. Please clarify how these data are used in the study and their specific role in the evaluation.

eq(4) what is Z(x,t)? and eq(6) what is γ(h)? Please carefully check all equations and ensure that each term is properly explained.

Figure 5: Please clarify the meaning of panel (d), which appears substantially different from the other panels.

Figure 6: Both the left and right panels are labeled as “(b)”. Please correct this labeling.

Line 305: For example, the authors state that MORDOR-SD is used to estimate average precipitation, but I cannot identify the corresponding results in Figure 6. I am not sure whether there are any similar problems in the manuscript, please clarify and correct this statement.
Citation: https://doi.org/10.5194/egusphere-2025-5679-RC1
CEC1:
'Comment on egusphere-2025-5679 - No compliance with the policy of the journal', Juan Antonio Añel, 23 Dec 2025

Dear authors,
Unfortunately, after checking your manuscript, it has come to our attention that it does not comply with our "Code and Data Policy".
https://www.geoscientific-model-development.net/policies/code_and_data_policy.html
You have archived the AROME model in a site that is not a suitable repository for scientific publication. Medcordex.eu does not fulfil GMD’s requirements for a persistent data archive because:
*It does not appear to have a published policy for data preservation over many years or decades (some flexibility exists over the precise length of preservation, but the policy must exist).

*It does not appear to have a published mechanism for preventing authors from unilaterally removing material. Archives must have a policy which makes removal of materials only possible in exceptional circumstances and subject to an independent curatorial decision,

*It does not appear to issue a persistent identifier such as a DOI or Handle for each precise asset.
If we have missed a published policy which does in fact address this matter satisfactorily, please post a response linking to it. If you have any questions about this issue, please post them in a reply.
In addition, you do not provide the code for the machine learning model that you have used in your work, but simply link a paper for it, something we can not accept. You must provide the code of the machine learning model that you have used in an acceptable repository according to our policy, and a link and DOi o Handle for it.
Please, therefore, publish your code and data in one of the appropriate repositories and reply to this comment with the relevant information (link and a permanent identifier for it (e.g. DOI)) as soon as possible. We cannot have manuscripts under discussion that do not comply with our policy.
The 'Code and Data Availability’ section must also be modified to cite the new repository locations, and corresponding references added to the bibliography.
I must note that if you do not fix this problem, we cannot continue with the peer-review process or accept your manuscript for publication in GMD.
Juan A. Añel
Geosci. Model Dev. Executive Editor

Citation: https://doi.org/10.5194/egusphere-2025-5679-CEC1
- AC1:
  'Reply on CEC1', Valentin Dura, 17 Feb 2026
  
  Dear Juan A. Añel,
  To clarify, we are not the AROME data provider; Météo-France is responsible for the AROME simulations.
  We downloaded four years of daily AROME data and deposited them in our Zenodo archive (https://zenodo.org/records/17491114), with the associated code.
  Regarding the AROME references, we previously cited them incorrectly. The AROME simulations are currently temporarily archived at:
  
  https://thredds-su.ipsl.fr/thredds/catalog/CNRM-WCRP-Data/CORDEX-FPSCONV/output/ALP-3/CNRM/ECMWF-ERAINT/evaluation/r1i1p1/CNRM-AROME41t1/fpsconv-x2yn2-v1/day/pr/files/d20240327/catalog.html
  This temporary hosting is necessary because the CNRM-CERFACS ESGF node is currently offline. It is expected to become operational again during 2026.
  We provide the machine learning code in our Zenodo archive (https://zenodo.org/records/17491114), specifically in the script 2.1-build_climatological_fields.R.
  Please let us know whether this addresses your concerns.
  Valentin Dura
  
  Citation: https://doi.org/10.5194/egusphere-2025-5679-AC1
  - CEC2: 'Reply on AC1', Juan Antonio Añel, 17 Feb 2026
    
    Dear authors,
    First, I would like to bring to your attention that, despite my previous comment being posted last December, you have not replied until today, almost two months later. This is not how we expect authors to address the editorial comments on manuscripts or how Discussions is supposed to work. Your manuscript is in an irregular situation, as it should not be under revision or Discussions in GMD, given the non-compliance with the policy of the journal. You were supposed to address the mentioned issues with code and data promptly, to ensure a fair and complete review process, which can not be performed without full compliance with our policy, as reviewers could miss important parts of your manuscript (code and data) that can not be reviewed if they are not made public without restrictions.
    That said, you mention in your reply that you continue hosting simulation outputs on servers that we cannot accept: ipsl.fr and CNRM-CERFACS ESGF. The fact that they are temporarily unavailable or that you need a temporary host, as you acknowledge, is evidence that such servers are not acceptable. Therefore, you must store the information hosted in ipsl.fr in a long-term repository that we can accept. In this regard, we request that you address and resolve this issue promptly.
    Juan A. Añel
    Geosci. Model Dev. Executive Editor
    
    Citation: https://doi.org/10.5194/egusphere-2025-5679-CEC2
    
    AC2: 'Reply on CEC2', Valentin Dura, 20 Feb 2026
    
    Dear Juan A. Añel,
    We apologize for this delayed response, mainly due to the communications with the data provider. I would like to emphasize once again that we are not the producers of the CNRM-AROME climate simulations, and we are therefore dependent on their choices regarding data storage. If including four years of data on Zenodo is not sufficient, we see no other option than to withdraw our submission in GMD.
    Sinerely,
    Valentin Dura
    
    Citation: https://doi.org/10.5194/egusphere-2025-5679-AC2
    
    CEC3: 'Reply on AC2', Juan Antonio Añel, 23 Feb 2026
    
    Dear authors,
    Thanks for your reply. I think we need some additional clarification. First, my initial comment pointed out some critical issues in your manuscript, but it did not address all the concerns we need to clarify regarding compliance with the journal's code and data policy for your manuscript. It is my understanding that, beyond the AROME data and the R scripts that you have shared, for your work, you have other data too, such as ERA5, NRC data, or models such as MORDOR-SD. It must be clear that you must share in a permanent repository all the code and data that you have used for your work, and it is necessary to clarify that all these elements are publicly stored in a permanent repository that complies with the policy of the journal before Discussions and peer review.
    
    Regarding your reply on the AROME data. First, you mention that you are not the producers of the AROME data. By itself, this does not remove your obligation to be sure that, as you have used them, they are properly stored to ensure the provenance of the data and replicability of the manuscript submitted. If the original authors do not store the data properly, and you can store it, then you should do so, unless the size makes it unfeasible. In such cases, if the data amounts to, for example, several hundred GBs, we could make exceptions to this requirement. As you have already stored four years of data under the CC-BY 4.0 license, it seems clear that you can store the data in a public repository. Also, each year of data is about 700 MB. All this seems to make it clear that you can store the data and that it is feasible from a technical point of view; therefore, you must comply with the policy and ensure that the data used for your work is properly stored. If I am missing something that could lead to a different conclusion, let us know in the reply to this comment.
    
    Finally, if you feel that you are not in a position to comply with the code and data policy of the journal, we fully understand that you prefer to withdraw your manuscript.
    
    Juan A. Añel
    Geosci. Model Dev. Executive Editor
    
    Citation: https://doi.org/10.5194/egusphere-2025-5679-CEC3
    
    AC3: 'Reply on CEC3', Valentin Dura, 16 Mar 2026
    
    Dear Juan A. Añel,
    Unfortunately, we are unable to provide all the datasets publicly. We will therefore withdraw this article. I do not find the "withdraw" button, could you take care of that ?
    Sincerely,
    Valentin Dura
    
    Citation: https://doi.org/10.5194/egusphere-2025-5679-AC3
RC2: 'Comment on egusphere-2025-5679', Anonymous Referee #1, 22 Feb 2026

Publisher’s note: the content of this comment was removed on 23 February 2026 since the comment was posted by mistake.

Citation: https://doi.org/10.5194/egusphere-2025-5679-RC2
RC3:
'Comment on egusphere-2025-5679', Anonymous Referee #2, 10 Apr 2026
Major comments:
A framework or a dataset? Or a bit of both?
The title of the paper states ISARD as a framework / precipitation analysis system, but the paper is clearly also meant to be a description and evaluation of ISARD the gridded dataset. Perhaps the title of the paper needs to be reconsidered that this is about both the framework and the dataset.

Basic information to ISARD the dataset:
From what year and month that ISARD data is available?

1km horizontal resolution but on what map projection? Presumably it is some transverse Mercator northing-easting grid? This is related my question for equation D1 (as I presume the answer would be the same).

Make sure you restate the resolution in the Introduction; I think the 1km resolution is only stated in the abstract.

Observation comparison dataset
Comparisons with reanalyses precipitation is often seen as quite crude (especially with ERA). There are several gridded precipitation data products for France, and the authors themselves have mentioned them in the Introduction. Recent work by Verney et al (2025) has investigated this especially from the perspective from high altitude precipitation. It may be helpful to bring in at least 1 such dataset for comparison.

Such a comparison will be more challenging for the ISARD framework/dataset. This is also where “structural uncertainty” comes into play: two competitive datasets perhaps with different strengths and weakness and a fairer description of inter-dataset uncertainty.

Additional information needed to follow the methodology:
Weather types/regimes: How is this defined is not mentioned in the paper. Is it something akin to the DWD or UKMO weather types (Bissoli and Dittmann 2001, Neal et al 2016). Please clarify.

Equations 2, 3, and their associated discussion:
What is “buffer distance”? This is not defined in the paper.

About variable “c”: Why use season and month together? Isn’t just one of them being sufficient? The equation implies “c” is a vector (c = g( (1, 2, …, 12), (DJF, MAM, JJA, SON), (categorical variable for weather regime/type) ); am I correct?

What does it mean by M or log M as a covariate? Are you putting the entire AROME rainfall time series at the gauge grid point as training data? Please clarify. If anything, what is the logic of using both M and log M as covariate?

Equation 3 implies B = E(P^corr)(x_s, c)? Am I correct? In that case, you probably want to drop the use of “B”.

Equation D1: It implies some sort of map projection is used to compute h_x and h_y, and both variables are in some physical units (aka kilometres). Please state that. I am pretty sure you need to provide a proj4 or epsg to sf and sp in order to do this.

Figures 5 and 6:
Both plots are too small to be read easily.

Figure 5: The title labels are even smaller. I can barely make out that they are showing results for specific ensemble members (32, 41, 81). The colour bar legend the right 3 panels are not the same. It is hard to tell if panel b results are “spikier” than the 2 panels to the right. For the caption, I think you also mean “…, and (c, d) daily precipitation estimates from ERA5-Land…”

Figure 6: I can only see the ERA5-Land line by zooming into the plot. To highlight the differences between the lines, consider using log scale for the y-axes.

Future work and Lines 385-386:
Not all global regions have the necessary rain gauge observation density or convection-permitting model simulations. Are there any region that authors speculate that this framework immediately deployable? Are there plans in doing that?

Minor comment:
Line 102: “subsectionSnowdatasets” – LaTeX markup typo check.
Line 299, start of Section 4.3: I think you mean Figure 5 and not Figure 2.
Line 320: “… high quantiles” >>> “… 0.999 and 0.9995 quantiles”
Line 355: “… under present warming” >>> “… under global warming”
Citation: https://doi.org/10.5194/egusphere-2025-5679-RC3

Valentin Dura, Guillaume Evin, Anne-Catherine Favre, and David Penot

Model code and software

ISARD code and input data Valentin Dura, Guillaume Evin, Anne-Catherine Favre, David Penot https://doi.org/10.5281/zenodo.17491114

Valentin Dura, Guillaume Evin, Anne-Catherine Favre, and David Penot

Viewed

Total article views: 562 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
332	189	41	562	36	31

HTML: 332
PDF: 189
XML: 41
Total: 562
BibTeX: 36
EndNote: 31

Views and downloads (calculated since 25 Nov 2025)

Month	HTML	PDF	XML	Total
Nov 2025	67	9	4	80
Dec 2025	68	66	17	151
Jan 2026	31	29	2	62
Feb 2026	99	45	12	156
Mar 2026	59	36	5	100
Apr 2026	8	4	1	13

Cumulative views and downloads (calculated since 25 Nov 2025)

Month	HTML	PDF	XML	Total
Nov 2025	67	9	4	80
Dec 2025	68	66	17	151
Jan 2026	31	29	2	62
Feb 2026	99	45	12	156
Mar 2026	59	36	5	100
Apr 2026	8	4	1	13

Viewed (geographical distribution)

Total article views: 550 (including HTML, PDF, and XML) Thereof 550 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 15 Apr 2026

Short summary

Traditional precipitation analysis often misrepresent seasonal totals and spatial variability of intense rainfall in mountains. This study introduces a reproducible workflow to generate a daily precipitation ensembles, conditioned on rain gauges. It outperforms standard products by better capturing seasonal totals. It also quantifies interpolation uncertainty, improving flood modeling. The open-source workflow is transferable to regions with sparse rain-gauge networks or limited radar coverage.


Total:	0
HTML:	0
PDF:	0
XML:	0