the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 12 Jan 2026
Landslide-Tsurrogate v1.0: A computationally efficient framework for probabilistic tsunami hazard assessment applied to Mayotte (France)
Abstract. Landslide-Tsurrogate v1.0 is an open-source Python and MATLAB tool that helps scientists quickly estimate the tsunami hazards generated by submarine landslides. Instead of running thousands of heavy deterministic numerical simulations, the software builds surrogate models that reproduce the main results with a fraction of the computational cost. The method relies on a mathematical approach called generalized polynomial chaos expansion, which efficiently explores how uncertain landslide parameters affect tsunami generation. Users can perform sensitivity analyses, identify the most influential parameters, and quantify the variability of possible outcomes. The tool includes a Jupyter Notebook User Manual and interactive MATLAB and Jupyter Notebook interfaces, making it easy to understand the methodology, set up the surrogate simulations and visualize the results. The Landslide-Tsurrogate v1.0 model’s performance is demonstrated through a real-world test case involving five zones in Mayotte (France). For this application, the surrogate models achieve convergence with only 135 deterministic simulations per zone and produce probabilistic results in less than 2 seconds within the user-friendly interfaces used on a basic laptop, demonstrating the computational efficiency of the approach. Beyond this example, the framework can be applied to any coastal region prone to submarine landslides. By combining physical modeling, statistical analysis, and user-friendly design, Landslide-Tsurrogate v1.0 enables faster and more transparent probabilistic tsunami hazard assessments.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(88649 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(88649 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-5671', Anonymous Referee #1, 04 Feb 2026
-
AC1: 'Reply on RC1', Clea Denamiel, 27 Mar 2026
We would like to thank the reviewer for their careful reading of the manuscript and for their constructive and insightful comments. We appreciate the time and effort dedicated to improving the quality and clarity of our work.
In the joint document, we provide a detailed, point-by-point response to each comment. Where appropriate, we have highlighted (in blue) how the manuscript will be revised to incorporate the suggested changes, clarify methodological aspects, and improve the overall presentation.
We believe that these revisions will strengthen the manuscript and we hope that the responses provided satisfactorily address the reviewer’s concerns.
-
AC1: 'Reply on RC1', Clea Denamiel, 27 Mar 2026
-
RC2: 'Comment on egusphere-2025-5671', Anonymous Referee #2, 09 Mar 2026
The manuscript describes the development and implementation of Landslide-Tsurrogate v1.0, a computational framework that builds generalized Polynomial Chaos Expansion surrogate models for landslide-generated tsunamis. The computational tools and worklows in Landslide-
Tsurrogate v1.0 are implemented in both Python and MATLAB to support accessibility within the scientific community and are distributed in an open-source format. The manuscript describes the methodology and use of the workflow and the case study in Mayotte. Five zones are studied in Mayotte, each one 207 simulations of Multilayer HySEA with variable volume, distance and friction angle. The framework is then used to predict 1000 scenarios across 211 locations.
There is novelty in providing such a tool in an open access format and in using it for prediction in the region of Mayotte. I believe this work is useful for scientists for early warning and should be published however there is first a need to address/clarify some of the points noted below.
Major points:
1) Suitability and transferability of the framework need to thoroughly addressed/clarified. At the first point what makes this surrogate toolbox tailored for applying to landslide tsunamis and not any other type of tsunamis. The output parameters of the deterministic model for training the surrogates are tsunami characteristics but this could be valid for any generation mechanism. Are the stochastic variables specific to landslide physical/geometric characteristics? If that is so, what is the range of variables that can be used with respect to influence for tsunamigenesis, such as the initial depth of the submarine landslide, that are not being studied. In general, I would place the information on the experimental design earlier in the manuscript.
2) Following the point above, it needs to be clarified whether the framework is suited for subaerial landslide tsunamis (as the dynamics of such events differ significantly from their submarine counterparts) and how transferable it may be across different bathymetric domains.
3) The authors should improve the readability of the manuscript as it is currently limited, the current form would be more suited for a chapter. Consider moving some sub-sections (in sections 2 or 3) of the manuscript as appendix or supplementary material to highlight the novelty of the work, (plus more on experimental design) and increase readability.
4) The term PTHA appears many times within the manuscript. Current literature for PTHA usually refers to the scientific method that quantifies the probability of a specific location experiencing a tsunami of a certain intensity within a given timeframe. Under these terms the return period of the tsunamigenic landslides is also being considered, meaning we assess the probability of when a 1 in 500 or a 1 in 50 years event is expected. Although the hazard is studied probabilistically, the concept of time is not considered in this work I would refrain from using this term to be consistent with current terminology.
L255 User specification: this section should be further clarified in terms of what can be achieved within this framework (also point 1). There are also some important variables in tsunamigenesis that need to be highlighted here such as the depth of the landslide, mode of failure, initial acceleration, maximum velocity, runout etc. It needs to be clarified whether these are assessed.
Section 4 Mayotte test case: More information on the design of experiments is needed as to why the three stochastic variables selected take precedence over other important landslide properties for tsunami generation.
L523 1035 simulations carry an associated computational cost for training with 3 parameters, even though across five regions. Have any tests been performed to understand whether convergence can be achieved with a smaller of simulations?
L670-672 Basal friction influences the acceleration and maximum velocity of a landslide which are both important parameters on the magnitude of tsunamis. How do the findings of friction not playing an important role compared to studies of similar type?
L707 An important question here is how these 1000 hypothetical scenarios add to the knowledge of the hazard that could not be acquired with the 200 computational runs. Beyond speed what are the additional gains of the surrogate modelling.
Minor points:
Introduction There are several approaches for building surrogate model, it would be good to have an understanding upfront in the introduction why this method is attractive and how this work contributes to other works on machine learning for landslide tsunamis.
L78-80 I agree, however it also needs to be stressed that real-time forecasting is still cumbersome as often there is no information on the magnitude of the tsunamigenic source when it comes to landslides. Surrogates are useful to provide a quantification of the uncertainty in that sense.
Figure 1 needs to be improved as the regions are not clearly seen in the larger map.
L500 please specify these values refer to basal friction rather than internal friction?
L507 please clarify what is considered the northernmost point?
4.2.1 More information is needed on the computational cost and spatial resolution, bathymetry data sources of the deterministic simulations.
L570 There is some freedom across the values of paraboloid length, width and thickness that result in similar values, how were these determined?
L570 Please clarify why the value of 211 locations (not more or less) is chosen, is this associated to a spatial interval in the deterministic runs?
L595 as inundation is not studied in this work since the emulators are offshore, please remove the word inundation as this can be confusing for the reader.
Citation: https://doi.org/10.5194/egusphere-2025-5671-RC2 -
AC2: 'Reply on RC2', Clea Denamiel, 27 Mar 2026
We would like to thank the reviewer for their careful reading of the manuscript and for their constructive and insightful comments. We appreciate the time and effort dedicated to improving the quality and clarity of our work.
In the joint document, we provide a detailed, point-by-point response to each comment. Where appropriate, we have highlighted (in blue) how the manuscript will be revised to incorporate the suggested changes, clarify methodological aspects, and improve the overall presentation.
We believe that these revisions will strengthen the manuscript and we hope that the responses provided satisfactorily address the reviewer’s concerns.
-
AC2: 'Reply on RC2', Clea Denamiel, 27 Mar 2026
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-5671', Anonymous Referee #1, 04 Feb 2026
The present article presents the Landslide-Tsurrogate v1.0 framework, an open-source tool, based on the generalized polynomial chaos expansion approach, for probabilistic tsunami hazard assessment generated by submarine landslides. After the description of the technique, an application to the Mayotte (France) study case is presented and discussed.
The topic is of interest to readership of the journal. The manuscript is in general well written, although a reorganization of the manuscript is strongly recommended (see following comments). Nevertheless, the manuscript presents some important issues that should be addressed before recommending the publication. Specifically:
1) Insufficient description and discussion of a crucial aspect: the deterministic simulations. While the mathematical framework that forms the Landslide-Tsurrogate v1.0 framework is carefully presented, the description and discussion of the deterministic simulations are inadequately addressed, while these are crucial for building and applying the probabilistic framework. Reading the manuscript, it appears that the choice of the numerical model to be used for the deterministic simulations is largely inconsequential, as is the distinction between 2D and 3D simulations, regardless of the specific characteristics of the site at hand. In my view, this represents a severe oversimplification. Depending on the considered site, 3D (or 2D multi-layer) simulations or may represent the only option to have reliable results, while in other cases also very simple models can be safely used. The choice of the model and its accuracy in reproducing the complex physics of landslide-tsunami generation, propagation and interaction with coast play a crucial role. These aspects should be better discussed as in the current version of the manuscript are not properly addressed. Moreover, the deterministic simulations rely on several crucial (sometimes arbitrary) assumptions: i.e., landslide locations, volumes and composition, rheological parameters, shape, extension and geometry of the failure surfaces, etc. These are characterized by huge uncertainty. Since the surrogate models builds upon the deterministic simulations results, these assumptions, or, more importantly, the criteria used to select the input parameters, are crucial. These aspects should be better discussed in the manuscript as in the current version of the manuscript are not properly addressed. Finally, it is clear that the surrogate model is highly computationally efficient. Nevertheless, a very important information to be provided is the exact number, or a detailed discussion on the criteria to select the exact number, of the deterministic simulations to be carried out (in general and for the considered study case) and their computational costs. These aspects should be specifically quantified and discussed for the considered case study. In any case, it would be important to provide guidelines for the application of the framework.
2) In its current form, some parts of the manuscript read more like a user manual than a scientific paper. This is not merely a stylistic issue; rather, the presentation does not adequately convey the points of novelty, nor does it sufficiently emphasize the limitations and/or the assumptions of the proposed framework. Some sections (e.g., Section 3.2, Technical description) are not particularly relevant from a scientific perspective. Similar considerations apply to Figures 4, 5, and 13. In my view, both Section 3.2 and Figures 4, 5, and 13, should not be included in the main manuscript but rather provided as supplementary material. Therefore, I strongly recommend a reorganization of the manuscript aimed, on the one hand, at avoiding the “user manual” effect and, on the other hand, at emphasizing the novel aspects of the present work.
3) The limitations of the present framework are not adequately presented or discussed. This aspect also relates to point 1). While the proposed approach appears useful and promising, like any model it is affected by limitations and shortcomings. These should be more thoroughly discussed and clearly emphasized, particularly in view of applying the framework to generic sites rather than only the considered case study. I therefore recommend explicitly describing and discussing the limitations of the approach, possibly in a dedicated section or subsection.
For the above reasons, I recommend a major revision of the manuscript before recommending it for publication.
Specific points:
L79-80. “Since monitoring such landslide-generated tsunamis is not currently state-of-the-art, achieving this goal necessitates a numerical modeling approach.” It is not clear what the Authors meant with this sentence. Please, explain/reformulate/remove.
L80-81. “However, executing ensembles of landslide-tsunami simulations in real time is computationally prohibitive.” It is not clear to me why ensembles simulations should be performed in real-time. PTHA is based on carrying out many simulations before the potential event, aiming at exploring the uncertainty in the parameters space. Based on the results of the PTHA real-time simulations might not be computationally expensive. In fact, they are currently used for Tsunami Early Warning Systems (TEWS) purposes.
L308-310. “The balance between numerical cost and accuracy is thus constraint by the availability of computational resources which, practically, plays a crucial role on the choice of the numerical model and simulation setup.” See comment 1).
L663-665. “Across all zones and most locations, landslide volume emerges as the dominant contributor to output variance of the maximum elevation and speed, with its sensitivity index typically approaching or exceeding 0.75. This finding underscores the central role of landslide magnitude in determining tsunami hazard intensity. The landslide location shows variable importance depending on the region and metric.” These are expected and well-known aspects, not new findings of the present work. Please, reformulate/remove.
L692-693. “An important aspect of the Mayotte submarine landslide test case is the emphasis on user-friendly interfaces (Fig. 13) that allow both researchers and decision-makers to interact with probabilistic tsunami hazard data efficiently.” I am not sure about the usefulness of Figure 13 as it is not clear the scientific content of the figure itself. See comment 2). I suggest removing this figure and/or providing it as an appendix/supporting information. The same considerations apply to Figures 4, 5, Section 3.2 Technical description.
Citation: https://doi.org/10.5194/egusphere-2025-5671-RC1 -
AC1: 'Reply on RC1', Clea Denamiel, 27 Mar 2026
We would like to thank the reviewer for their careful reading of the manuscript and for their constructive and insightful comments. We appreciate the time and effort dedicated to improving the quality and clarity of our work.
In the joint document, we provide a detailed, point-by-point response to each comment. Where appropriate, we have highlighted (in blue) how the manuscript will be revised to incorporate the suggested changes, clarify methodological aspects, and improve the overall presentation.
We believe that these revisions will strengthen the manuscript and we hope that the responses provided satisfactorily address the reviewer’s concerns.
-
AC1: 'Reply on RC1', Clea Denamiel, 27 Mar 2026
-
RC2: 'Comment on egusphere-2025-5671', Anonymous Referee #2, 09 Mar 2026
The manuscript describes the development and implementation of Landslide-Tsurrogate v1.0, a computational framework that builds generalized Polynomial Chaos Expansion surrogate models for landslide-generated tsunamis. The computational tools and worklows in Landslide-
Tsurrogate v1.0 are implemented in both Python and MATLAB to support accessibility within the scientific community and are distributed in an open-source format. The manuscript describes the methodology and use of the workflow and the case study in Mayotte. Five zones are studied in Mayotte, each one 207 simulations of Multilayer HySEA with variable volume, distance and friction angle. The framework is then used to predict 1000 scenarios across 211 locations.
There is novelty in providing such a tool in an open access format and in using it for prediction in the region of Mayotte. I believe this work is useful for scientists for early warning and should be published however there is first a need to address/clarify some of the points noted below.
Major points:
1) Suitability and transferability of the framework need to thoroughly addressed/clarified. At the first point what makes this surrogate toolbox tailored for applying to landslide tsunamis and not any other type of tsunamis. The output parameters of the deterministic model for training the surrogates are tsunami characteristics but this could be valid for any generation mechanism. Are the stochastic variables specific to landslide physical/geometric characteristics? If that is so, what is the range of variables that can be used with respect to influence for tsunamigenesis, such as the initial depth of the submarine landslide, that are not being studied. In general, I would place the information on the experimental design earlier in the manuscript.
2) Following the point above, it needs to be clarified whether the framework is suited for subaerial landslide tsunamis (as the dynamics of such events differ significantly from their submarine counterparts) and how transferable it may be across different bathymetric domains.
3) The authors should improve the readability of the manuscript as it is currently limited, the current form would be more suited for a chapter. Consider moving some sub-sections (in sections 2 or 3) of the manuscript as appendix or supplementary material to highlight the novelty of the work, (plus more on experimental design) and increase readability.
4) The term PTHA appears many times within the manuscript. Current literature for PTHA usually refers to the scientific method that quantifies the probability of a specific location experiencing a tsunami of a certain intensity within a given timeframe. Under these terms the return period of the tsunamigenic landslides is also being considered, meaning we assess the probability of when a 1 in 500 or a 1 in 50 years event is expected. Although the hazard is studied probabilistically, the concept of time is not considered in this work I would refrain from using this term to be consistent with current terminology.
L255 User specification: this section should be further clarified in terms of what can be achieved within this framework (also point 1). There are also some important variables in tsunamigenesis that need to be highlighted here such as the depth of the landslide, mode of failure, initial acceleration, maximum velocity, runout etc. It needs to be clarified whether these are assessed.
Section 4 Mayotte test case: More information on the design of experiments is needed as to why the three stochastic variables selected take precedence over other important landslide properties for tsunami generation.
L523 1035 simulations carry an associated computational cost for training with 3 parameters, even though across five regions. Have any tests been performed to understand whether convergence can be achieved with a smaller of simulations?
L670-672 Basal friction influences the acceleration and maximum velocity of a landslide which are both important parameters on the magnitude of tsunamis. How do the findings of friction not playing an important role compared to studies of similar type?
L707 An important question here is how these 1000 hypothetical scenarios add to the knowledge of the hazard that could not be acquired with the 200 computational runs. Beyond speed what are the additional gains of the surrogate modelling.
Minor points:
Introduction There are several approaches for building surrogate model, it would be good to have an understanding upfront in the introduction why this method is attractive and how this work contributes to other works on machine learning for landslide tsunamis.
L78-80 I agree, however it also needs to be stressed that real-time forecasting is still cumbersome as often there is no information on the magnitude of the tsunamigenic source when it comes to landslides. Surrogates are useful to provide a quantification of the uncertainty in that sense.
Figure 1 needs to be improved as the regions are not clearly seen in the larger map.
L500 please specify these values refer to basal friction rather than internal friction?
L507 please clarify what is considered the northernmost point?
4.2.1 More information is needed on the computational cost and spatial resolution, bathymetry data sources of the deterministic simulations.
L570 There is some freedom across the values of paraboloid length, width and thickness that result in similar values, how were these determined?
L570 Please clarify why the value of 211 locations (not more or less) is chosen, is this associated to a spatial interval in the deterministic runs?
L595 as inundation is not studied in this work since the emulators are offshore, please remove the word inundation as this can be confusing for the reader.
Citation: https://doi.org/10.5194/egusphere-2025-5671-RC2 -
AC2: 'Reply on RC2', Clea Denamiel, 27 Mar 2026
We would like to thank the reviewer for their careful reading of the manuscript and for their constructive and insightful comments. We appreciate the time and effort dedicated to improving the quality and clarity of our work.
In the joint document, we provide a detailed, point-by-point response to each comment. Where appropriate, we have highlighted (in blue) how the manuscript will be revised to incorporate the suggested changes, clarify methodological aspects, and improve the overall presentation.
We believe that these revisions will strengthen the manuscript and we hope that the responses provided satisfactorily address the reviewer’s concerns.
-
AC2: 'Reply on RC2', Clea Denamiel, 27 Mar 2026
Peer review completion
Journal article(s) based on this preprint
Data sets
Landslide-Tsurrogate-V1.0-GMD Clea Denamiel https://zenodo.org/records/17519408
Model code and software
Landslide-Tsurrogate-V1.0-GMD Clea Denamiel https://zenodo.org/records/17519408
Interactive computing environment
Landslide-Tsurrogate-V1.0-GMD Clea Denamiel https://zenodo.org/records/17519408
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 288 | 404 | 32 | 724 | 20 | 39 |
- HTML: 288
- PDF: 404
- XML: 32
- Total: 724
- BibTeX: 20
- EndNote: 39
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Cléa Lumina Denamiel
Anne Mangeney
Anne Le Friant
Marc Peruzzetto
Antoine Lucas
Manuel J. Castro Díaz
Enrique Fernández-Nieto
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(88649 KB) - Metadata XML
The present article presents the Landslide-Tsurrogate v1.0 framework, an open-source tool, based on the generalized polynomial chaos expansion approach, for probabilistic tsunami hazard assessment generated by submarine landslides. After the description of the technique, an application to the Mayotte (France) study case is presented and discussed.
The topic is of interest to readership of the journal. The manuscript is in general well written, although a reorganization of the manuscript is strongly recommended (see following comments). Nevertheless, the manuscript presents some important issues that should be addressed before recommending the publication. Specifically:
1) Insufficient description and discussion of a crucial aspect: the deterministic simulations. While the mathematical framework that forms the Landslide-Tsurrogate v1.0 framework is carefully presented, the description and discussion of the deterministic simulations are inadequately addressed, while these are crucial for building and applying the probabilistic framework. Reading the manuscript, it appears that the choice of the numerical model to be used for the deterministic simulations is largely inconsequential, as is the distinction between 2D and 3D simulations, regardless of the specific characteristics of the site at hand. In my view, this represents a severe oversimplification. Depending on the considered site, 3D (or 2D multi-layer) simulations or may represent the only option to have reliable results, while in other cases also very simple models can be safely used. The choice of the model and its accuracy in reproducing the complex physics of landslide-tsunami generation, propagation and interaction with coast play a crucial role. These aspects should be better discussed as in the current version of the manuscript are not properly addressed. Moreover, the deterministic simulations rely on several crucial (sometimes arbitrary) assumptions: i.e., landslide locations, volumes and composition, rheological parameters, shape, extension and geometry of the failure surfaces, etc. These are characterized by huge uncertainty. Since the surrogate models builds upon the deterministic simulations results, these assumptions, or, more importantly, the criteria used to select the input parameters, are crucial. These aspects should be better discussed in the manuscript as in the current version of the manuscript are not properly addressed. Finally, it is clear that the surrogate model is highly computationally efficient. Nevertheless, a very important information to be provided is the exact number, or a detailed discussion on the criteria to select the exact number, of the deterministic simulations to be carried out (in general and for the considered study case) and their computational costs. These aspects should be specifically quantified and discussed for the considered case study. In any case, it would be important to provide guidelines for the application of the framework.
2) In its current form, some parts of the manuscript read more like a user manual than a scientific paper. This is not merely a stylistic issue; rather, the presentation does not adequately convey the points of novelty, nor does it sufficiently emphasize the limitations and/or the assumptions of the proposed framework. Some sections (e.g., Section 3.2, Technical description) are not particularly relevant from a scientific perspective. Similar considerations apply to Figures 4, 5, and 13. In my view, both Section 3.2 and Figures 4, 5, and 13, should not be included in the main manuscript but rather provided as supplementary material. Therefore, I strongly recommend a reorganization of the manuscript aimed, on the one hand, at avoiding the “user manual” effect and, on the other hand, at emphasizing the novel aspects of the present work.
3) The limitations of the present framework are not adequately presented or discussed. This aspect also relates to point 1). While the proposed approach appears useful and promising, like any model it is affected by limitations and shortcomings. These should be more thoroughly discussed and clearly emphasized, particularly in view of applying the framework to generic sites rather than only the considered case study. I therefore recommend explicitly describing and discussing the limitations of the approach, possibly in a dedicated section or subsection.
For the above reasons, I recommend a major revision of the manuscript before recommending it for publication.
Specific points:
L79-80. “Since monitoring such landslide-generated tsunamis is not currently state-of-the-art, achieving this goal necessitates a numerical modeling approach.” It is not clear what the Authors meant with this sentence. Please, explain/reformulate/remove.
L80-81. “However, executing ensembles of landslide-tsunami simulations in real time is computationally prohibitive.” It is not clear to me why ensembles simulations should be performed in real-time. PTHA is based on carrying out many simulations before the potential event, aiming at exploring the uncertainty in the parameters space. Based on the results of the PTHA real-time simulations might not be computationally expensive. In fact, they are currently used for Tsunami Early Warning Systems (TEWS) purposes.
L308-310. “The balance between numerical cost and accuracy is thus constraint by the availability of computational resources which, practically, plays a crucial role on the choice of the numerical model and simulation setup.” See comment 1).
L663-665. “Across all zones and most locations, landslide volume emerges as the dominant contributor to output variance of the maximum elevation and speed, with its sensitivity index typically approaching or exceeding 0.75. This finding underscores the central role of landslide magnitude in determining tsunami hazard intensity. The landslide location shows variable importance depending on the region and metric.” These are expected and well-known aspects, not new findings of the present work. Please, reformulate/remove.
L692-693. “An important aspect of the Mayotte submarine landslide test case is the emphasis on user-friendly interfaces (Fig. 13) that allow both researchers and decision-makers to interact with probabilistic tsunami hazard data efficiently.” I am not sure about the usefulness of Figure 13 as it is not clear the scientific content of the figure itself. See comment 2). I suggest removing this figure and/or providing it as an appendix/supporting information. The same considerations apply to Figures 4, 5, Section 3.2 Technical description.