the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Jan 2026
Cartino2D: Scalable and Automated 2D Shallow Water Rainfall-Flood Inundation Modeling up to Very High Resolution for Large Domains
Abstract. While 2D shallow water models with rainfall and infiltration provide a physically consistent framework for flood inundation modeling, their automated application at large scales remains constrained by challenges related to unstructured mesh generation, parameter specifications, and the integration of heterogeneous geospatial and hydrological datasets. This study presents Cartino 2D (C2D), a novel automated framework that enables the large-scale deployment of the well-established Telemac2D model, for solving the complete 2D shallow water equations, with flexible, spatially distributed hydrological forcing—either from rainfall fields or discharge hydrographs. C2D features topography-aware unstructured mesh generation, optional automated handling of hydraulic structures, and spatial parameter estimation from diverse datasets, including land use. It supports multi-resolution simulations up to very high (metric) resolutions and includes optional automated flow analysis at user-defined transects. The framework also features an automatic subdomain sectorization step, based on preliminary simulations on a regular grid, to delineate hydrologically-hydraulically consistent regions and inform targeted unstructured meshing procedures. The framework is successfully applied at the national scale across France, using 100-year return rainfall and discharge values from the SHYREG database, as well as at very high resolution in the complex metropolitan area such as the Aix-Marseille Provence or Grabels City, demonstrating both scalability and robustness. Model outputs are evaluated using flood marks and firefighter intervention records, showing encouraging hydrological and hydraulic consistency. This advancement opens new opportunities for large-scale flood hazard pre-assessment in France and can be transposed to other countries using global and/or national data. Future work will focus on improving culvert representation, testing alternative infiltration models, and extending the framework for model parameter optimization, coastal flooding and real-time applications.
- Preprint
(27778 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 08 Aug 2026)
-
CEC1: 'Comment on egusphere-2025-3333 - No compliance with the policy of the journal', Juan Antonio Añel, 11 Feb 2026
reply
-
CC1: 'Reply on CEC1', Frédéric Pons, 13 Feb 2026
reply
Dear Juan A. Añel,
Thank you for your message and for bringing these important compliance points to our attention. We have carefully reviewed your comments and are committed to addressing them in full to ensure our manuscript aligns with Geoscientific Model Development’s Code and Data Policy. Below, we provide detailed responses and propose specific corrective actions.
A complete version of the Cartino2D source code is already archived in the Zenodo dataset associated with this paper at https://doi.org/10.5281/zenodo.17778609, specifically in the folder "Cartino2D_Github.zip". This reference is cited on line 370 (Pons and Hocini, 2025) and in the “Code and Data Availability” section.
Following your feedback, we propose updating the text in the “Code and Data Availability” section to explicitly mention the inclusion of the source code (which was already available in the Zenodo archive and has been successfully downloaded and used based on recent feedback):
The current text in the Code and Data Availability section reads:
“To facilitate reproducibility, a dataset is provided to run Cartino2D v1.0, available in the GitHub repository above. This dataset includes all input data required to execute Cartino2D (see Section 3), as well as reference outputs for the two cases described in the article (see Figure 3 and Figure 12). These cases cover both a ‘coarse’ 25m resolution and a finer resolution, with an overlap on Grabels areas. The dataset is permanently archived on Zenodo and accessible at https://doi.org/10.5281/zenodo.17778608 (Pons and Hocini, 2025).”
We propose clarifying this to state that the code archive of Cartino2D is available in Zenodo, along with the required data to run it:
“To facilitate reproducibility, the source code and a dataset are provided to run Cartino2D v1.0 in the Zenodo repository (https://doi.org/10.5281/zenodo.17778609). This source code and dataset include all input data required to execute Cartino2D (see Section 3), as well as reference outputs for the two cases described in the article (see Figure 3 and Figure 12). These cases cover both a ‘coarse’ 25m resolution and a finer resolution, with an overlap on Grabels areas. The dataset is permanently archived on Zenodo and accessible at https://doi.org/10.5281/zenodo.17778608 (Pons and Hocini, 2025).”
As already stated in the manuscript:
“The ANTILOPE J+1 product by Météo-France (Champeaux et al., 2009) is a commercial product, and access is restricted to authorized users. An alternative open dataset is available at https://www.data.gouv.fr/datasets/reanalyses-comephore/.”
The link on data.gouv.fr provides equivalent spatial and temporal coverage but does not include the exact commercial data used in our study. For the 6–7 October 2014 event (Grabels case), we have already included one input file for the Telemac2D code in the Zenodo dataset (10.5281/zenodo.17778608). The file name is: Cartino2D_Github/France_Pluie/Grabels/spatial_C0005_861143km_X765440Y6284136_Evts201410061700_08h15min_AMC2_ProjMF_Evt_Anti_15min_J1_MFCerema_2014.txt.
We tested the SHYREG webpage on 13 February 2026 and confirm that it is accessible, though it requires authentication. To ensure full reproducibility, we have already included one input file containing processed statistical SHYREG rainfall data for the Telemac2D code in the Zenodo dataset (10.5281/zenodo.17778608) for the coarse case study: Cartino2D_Github/France_Pluie/EAIM/C6147_733929km_X789149Y6302130/shyreg_spPB_C6147_733929km_X789149Y6302130_T0100_D24_PIC08.txt.
Please let us know if any further clarification or adjustments are needed.
Best regards,
Frédéric Pons and Pierre-André Garambois
Citation: https://doi.org/10.5194/egusphere-2025-3333-CC1 -
CEC2: 'Reply on CC1', Juan Antonio Añel, 13 Feb 2026
reply
Dear authors,
Thanks for the clarifications. The rewording of the Code and Data Availability section that you propose makes it easier to assess the compliance with the the policy of the journal. It is unfortunate that your work depends on proprietary data that are not widely available and properly stored, but we understand it, and we do not consider a violation of the current policy of the journal. That said, we can consider the current version of your manuscript in compliance with the policy of the journal.
Juan A. Añel
Geosci. Model Dev. Executive Editor
Citation: https://doi.org/10.5194/egusphere-2025-3333-CEC2
-
CEC2: 'Reply on CC1', Juan Antonio Añel, 13 Feb 2026
reply
-
CC1: 'Reply on CEC1', Frédéric Pons, 13 Feb 2026
reply
-
RC1: 'Comment on egusphere-2025-3333', Brett Sanders, 25 Jun 2026
reply
Summary
This article describes a framework to carry out flood hazard mapping at the national scale with a 2D hydrodynamic flood inundation model as the final step of a modeling chain. The framework is presented for France, it includes a 25 m resolution application at the national scale and a more refined model down to 3 m resolution covering a domain 800 km2 in size.
National scale flood hazard modeling at fine resolution is a significant undertaking. Considerable work has to go into planning the representation of hazard drivers (e.g., rainfall, streamflow), accessing the data sources needed to represent rainfall, infiltration and flood infrastructure, and building the modeling stack that pulls all of this information together in a systematic manner. The demands on computing and data management are also substantial. Alas, new efforts to pull together available models and create frameworks capable of flood hazard mapping are a good topic for research and development.
This paper builds out a system that relies on the unstructured grid model Telemac 2D for inundation modeling, which has been available for decades and used extensively for 2D hydraulic modeling. Telemac relies on an unstructured grid of triangular cells, and thus the modeling framework presented her includes a mesh generation element whereby topographic features constrain mesh edge locations and resolution with the aim of optimizing accuracy relative to computational effort.
Critique
This paper is long and reads as though it was somewhat hastily pulled together. The writing is fragmented, lacking flow and cohesiveness of ideas. Sentences are incomplete. Ideas are not fully formed. In terms of purpose, it’s almost as if the authors are caught between an effort to write a user manual and an effort to report on nationwide flood risks.
The work can be viewed as a major accomplishment based upon the scale of hydrodynamic flood modeling, but I don’t see it as an elegant solution. In particular, the nationwide application of the model relies on a structured grid whereby every Cartesian grid cell is divided into two triangles. If the national scale flood model is best achieved with a structured grid, then why not use a Cartesian grid of rectangles? The paper could potentially be far shorter and more concise without all of details on mesh generation. Another factor to consider is that the use of two triangles in place of every square or rectangular cell leads to a doubling of the number of compute cells, and a significantly higher compute cost. Kim et al. (2014) compared various types of unstructured grids including grids with all triangular and all quadrilateral and combinations of the two, and a takeway from that paper is that grids with only triangular cells are not the most efficient and should be avoided. That can be tough pill to swallow because meshing with triangular cells is very easy, and meshing with quadrilaterals is more difficult to automate. In our lab, we’ve gone away from unstructured grids and adopted Cartesian grids for the sake of simplified data management and high-performance computing with GPUs (Sanders and Schubert 2019, Sanders et al. 2025).
The city-scale application in this paper relies on unstructured grids and thus it manifests the modeling framework (data/meshing/modeling) in its totality. But by the time the reader arrives to this section, his or her concentration will likely be exhausted.
So to summarize, I applaud the authors on this accomplishment – it’s no simple task to pull together a large scale, fine-resolution flood inundation modeling framework and apply it both national and city scales. However, I encourage the authors to prepare a manuscript with scope and content that is more appropriate for digestion by scholars in the field: a clear purpose, a straightforward presentation, and a reasonable length.
My sense is that the authors could go two different directions for a more concise contribution: a) focus on the nationwide hydraulic flood inundation model without concern for the meshing, since regular grids are preferred, and feature the comparisons to observec claims and so forth, or b) focus on regional scale modeling (say ~1000 km2) where the unstructured grid is demonstrated and may be advantageous – including test results where these advantages are shown. Indeed, it would be a valuable contribution to compare the unstructured grid with variable resolution against the regular grid approach at the regional scale. And if the advantages are not shown, then that would be a valuable result as well.
Kim, B., Sanders, B. F., Schubert, J. E., & Famiglietti, J. S. (2014). Mesh type tradeoffs in 2D hydrodynamic modeling of flooding with a Godunov-based flow solver. Advances in Water Resources, 68, 42-61.
Sanders, B. F., & Schubert, J. E. (2019). PRIMo: Parallel raster inundation model. Advances in Water Resources, 126, 79-95.
Sanders, B. F., Schubert, J. E., Martin, E. M. H., Wang, S., Sukop, M. C., & Mach, K. J. (2025). A fast flood inundation model with groundwater interactions and hydraulic structures. Advances in water resources, 105057.
Citation: https://doi.org/10.5194/egusphere-2025-3333-RC1 -
AC1: 'Reply on RC1', Frédéric Pons, 03 Jul 2026
reply
Dear Dr. Sanders,
Thank you for your careful reading of the manuscript and for the constructive, detailed feedback. We are grateful that you recognize this work as a major accomplishment in bringing together data and models for flood hazard mapping over large areas, and we welcome the suggestions for improving the clarity and focus of the paper. Below we address each of your comments in turn and describe the changes we propose for the revised manuscript, which we expect to submit within two months.
The proposed revisions summarize as follows:
- Rescope the manuscript to give more weight to the national-scale and local-scale applications presented concisely and with additional quantitative results.
- Move the technical details of automatic mesh generation and model setup to an appendix, keeping the main text focused on the pre-/post-processing framework itself; the core contribution of this work.
- Clarify and strengthen the justification for the meshing choice, and better emphasize the generalizability of the framework to other hydraulic solvers and data assimilation approaches.
Kind regards,
Frederic Pons, Pierre-André Garambois and Nabil Hocini
In details
RC: This paper is long and reads as though it was somewhat hastily pulled together. The writing is fragmented, lacking flow and cohesiveness of ideas. Sentences are incomplete. Ideas are not fully formed. In terms of purpose, it’s almost as if the authors are caught between an effort to write a user manual and an effort to report on nationwide flood risks.
AR: Regarding the length of the paper and its objectives, we initially targeted Geoscientific Model Development (GMD) due to its unique focus on methodological descriptions and new open-source tools. However, we take your point that the current structure affects readability. In the revised manuscript, we will heavily streamline the main narrative by shifting the dense technical meshing algorithms to the appendices. This allows us to keep the main text sharply focused on the core framework, while expanding the national and local sections with the quantitative results you suggested. This restructuring will ensure the paper reads as a cohesive research contribution rather than a hybrid technical manual.
RC: The work can be viewed as a major accomplishment based upon the scale of hydrodynamic flood modeling, but I don’t see it as an elegant solution. In particular, the nationwide application of the model relies on a structured grid whereby every Cartesian grid cell is divided into two triangles. If the national scale flood model is best achieved with a structured grid, then why not use a Cartesian grid of rectangles? The paper could potentially be far shorter and more concise without all of details on mesh generation. Another factor to consider is that the use of two triangles in place of every square or rectangular cell leads to a doubling of the number of compute cells, and a significantly higher compute cost. Kim et al. (2014) compared various types of unstructured grids including grids with all triangular and all quadrilateral and combinations of the two, and a takeway from that paper is that grids with only triangular cells are not the most efficient and should be avoided. That can be tough pill to swallow because meshing with triangular cells is very easy, and meshing with quadrilaterals is more difficult to automate. In our lab, we’ve gone away from unstructured grids and adopted Cartesian grids for the sake of simplified data management and high-performance computing with GPUs (Sanders and Schubert 2019, Sanders et al. 2025).
AR: Thank you for recognizing the national scale of this hydrodynamic flood modeling as a major accomplishment, and for raising these excellent points regarding grid efficiency.
We completely agree with the reviewer's assessment: splitting a Cartesian grid into triangles doubles the element count, increases computational overhead, and is less inherently efficient than a native Cartesian or quadrilateral solver. However, our choice of discretization was strictly dictated by our target hydraulic engine, TELEMAC-2D, whose finite-element formulation is natively hardcoded for unstructured triangular meshes and cannot operate on quadrilateral or Cartesian rectangular grids. To achieve an automated, national-scale application within the TELEMAC ecosystem, splitting the structured Cartesian cells into triangles was the most robust operational path.
We will clarify this solver constraint in the revised manuscript and explicitly emphasize that Cartino2D is designed as a solver-agnostic pre- and post-processing framework, rather than an advocate for a specific mesh type. While this initial implementation targets TELEMAC's triangular structure, the underlying pipeline is highly flexible. As evidence of this flexibility, elements of our pipeline are already utilized by DasFlow2D for inversion of infiltration parameters within a shallow water model at the basin scale (Pujol et al., 2025) and are being adapted for a GPU-based, differentiable physics-AI finite-volume solver operating on diverse mesh types (Colleoni et al., in prep).
Following your excellent suggestions, we will significantly revise the manuscript to:
- Streamline the main text by reducing the dense algorithmic details of the meshing process, moving technical specifics to an appendix.
- Incorporate a dedicated discussion on grid trade-offs, explicitly citing Kim et al. (2014) regarding triangular vs. quadrilateral efficiency, and Sanders & Schubert (2019) / Sanders et al. (2025) regarding the clear data-management and high-performance benefits of modern GPU-accelerated Cartesian grids.
This framing shifts the focus of our paper away from defending triangular meshes, positioning it instead as an adaptable pipeline that bridges data management and large-scale hydraulic modeling across different discretization choices.
RC: The city-scale application in this paper relies on unstructured grids and thus it manifests the modeling framework (data/meshing/modeling) in its totality. But by the time the reader arrives to this section, his or her concentration will likely be exhausted.
AR: We agree that keeping the reader engaged is essential. Because the city-scale application demonstrates the framework in its totality, we want to ensure it stands out. To prevent reader fatigue, we will heavily streamline the preceding sections and move the technical meshing details to an appendix. This will make the manuscript significantly more concise, allowing the reader to reach this final application section much more quickly.
RC: So to summarize, I applaud the authors on this accomplishment – it’s no simple task to pull together a large scale, fine-resolution flood inundation modeling framework and apply it both national and city scales. However, I encourage the authors to prepare a manuscript with scope and content that is more appropriate for digestion by scholars in the field: a clear purpose, a straightforward presentation, and a reasonable length.
AR: Thank you for your encouraging remarks and for recognizing the scale of this accomplishment.
In the revised version, we will heavily streamline the text to achieve a more concise length and a straightforward presentation. Specifically, we will:
- Refocus the narrative on our core contribution: a scale-adaptable framework that bridges national and local applications via hydrological sectorization. This approach enables an iterative workflow where coarser national-scale models logically partition the domain to guide high-resolution, local-scale simulations.
- Remove dense algorithmic descriptions from the main text, moving specific meshing details to an appendix to significantly improve readability and flow.
We believe these adjustments will deliver a much more digestible manuscript that clearly highlights the practical value of our framework for scholars in the field.
RC: My sense is that the authors could go two different directions for a more concise contribution: a) focus on the nationwide hydraulic flood inundation model without concern for the meshing, since regular grids are preferred, and feature the comparisons to observec claims and so forth, or b) focus on regional scale modeling (say ~1000 km2) where the unstructured grid is demonstrated and may be advantageous – including test results where these advantages are shown. Indeed, it would be a valuable contribution to compare the unstructured grid with variable resolution against the regular grid approach at the regional scale. And if the advantages are not shown, then that would be a valuable result as well.
AR: We thank the reviewer for proposing these two clear directions for a more concise manuscript.
Regarding the paper's scope, our goal is to present the Cartino2D framework itself rather than just a single model application. Retaining both the national and local applications is essential to demonstrate the scalable nature of the tool, which aligns directly with the scope of Geoscientific Model Development (GMD). To ensure a more concise contribution as requested, we will shift the dense algorithmic meshing details to an appendix, allowing the main text to focus strictly on this multi-scale versatility. We plan to publish deep-dive, separate papers on the individual hydraulic findings at each scale in future journal articles.
Regarding your specific suggestions for Directions A and B:
- On Direction A (National-Scale Validation): We agree that more comprehensive comparison adds value. While insurance claim data is strictly confidential and unavailable in France, we will strengthen the national-scale section by incorporating a comparison of simulated discharges across approximately 5,000 control sections (leveraging measurement gauges and 100-year return period flows from national hydrological models), alongside our existing spatial comparisons with historic flood markers.
- On Direction B (Unstructured vs. Structured Grid Discrepancies): The reviewer makes an excellent point regarding grid comparisons. While structured grids at 25m or 5m resolution adequately preserve macro-scale flood patterns for extreme events, they deviate significantly from unstructured meshes in complex environments. Unstructured grids allow us to explicitly account for micro-topography, urban layouts, and small hydraulic structures.
- On the Challenge of Local Data: We will add a discussion acknowledging that the bottleneck for high-resolution, unstructured local modeling is often data availability. For example, at a departmental regional scale of approximately 5,000 km2, the dimensions of roughly 5,000 hydraulic structures remain unknown in public databases. Conversely, where highly detailed datasets are available, such as in our work within the Montpellier urban area, where over 200 buried structures greater than 800mm and low-lying road crossings were explicitly integrated, the advantages of an unstructured mesh become indispensable.
By restructuring the paper to highlight this continuum, using national-scale structured setups to identify regions and local unstructured meshes to capture complex structural hydraulics, we believe the manuscript will offer a straightforward, balanced contribution that addresses both of the reviewer's core insights.
Citation: https://doi.org/10.5194/egusphere-2025-3333-AC1
-
AC1: 'Reply on RC1', Frédéric Pons, 03 Jul 2026
reply
Data sets
Dataset for Cartino2D paper Frédéric Pons and Nabil Hocini https://doi.org/10.5281/zenodo.17778609
Model code and software
Cartino2D Frédéric Pons and Nabil Hocini https://github.com/CEREMA/cartino2d
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,404 | 2,041 | 120 | 3,565 | 128 | 167 |
- HTML: 1,404
- PDF: 2,041
- XML: 120
- Total: 3,565
- BibTeX: 128
- EndNote: 167
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
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
First, you have archived almost all of the assets necessary to replicate your work in sites that are not acceptable according to the policy of the journal. They include GitHub and GitLab sites, and several others such as data.gouv.fr and inrae.fr. In some cases, such as for the ANTILOPE J+1 product you state that access is restricted, and when checking it, the information in the web page linked states that the data are open. The https://shyreg.pluie.recover.inrae.fr/ web page does not even work.
Therefore, the current situation with your manuscript is highly irregular. It should have never been accepted for peer-review or Discussions in the journal given these issues. The GMD review and publication process depends on reviewers and community commentators being able to access, during the discussion phase, the code and data on which a manuscript depends, and on ensuring the provenance of replicability of the published papers for years after their publication. Please, therefore, publish all the code and data necessary to replicate your work 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