Articles | Volume 26, issue 5
https://doi.org/10.5194/nhess-26-2089-2026
© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
https://doi.org/10.5194/nhess-26-2089-2026
© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Research article
| 
08 May 2026
Research article |
| 08 May 2026
| 08 May 2026
Probabilistic flood hazard mapping for dike-breach floods via graph neural networks
Roberto Bentivoglio
CORRESPONDING AUTHOR
Department of Water Management, Faculty of Civil Engineering and Geosciences, Delft University of Technology, Delft, the Netherlands
Sebastiaan Nicolas Jonkman
Department of Hydraulic Engineering, Faculty of Civil Engineering and Geosciences, Delft University of Technology, Delft, the Netherlands
Elvin Isufi
Department of Intelligent Systems, Faculty of Electrical Engineering, Mathematics and Computer Science, Delft University of Technology, Delft, the Netherlands
Riccardo Taormina
Department of Water Management, Faculty of Civil Engineering and Geosciences, Delft University of Technology, Delft, the Netherlands
Related authors
Roberto Bentivoglio, Elvin Isufi, Sebastiaan Nicolas Jonkman, and Riccardo Taormina
Nat. Hazards Earth Syst. Sci., 25, 335–351, https://doi.org/10.5194/nhess-25-335-2025, https://doi.org/10.5194/nhess-25-335-2025, 2025
Short summary
Short summary
Deep learning methods are increasingly used as surrogates for spatio-temporal flood models but struggle with generalization and speed. Here, we propose a multi-resolution approach using graph neural networks that predicts dike breach floods across different meshes, topographies, and boundary conditions with high accuracy and up to 1000× speed-ups. The model also generalizes to larger more complex case studies with just one additional simulation for fine-tuning.
Roberto Bentivoglio, Elvin Isufi, Sebastiaan Nicolas Jonkman, and Riccardo Taormina
Hydrol. Earth Syst. Sci., 27, 4227–4246, https://doi.org/10.5194/hess-27-4227-2023, https://doi.org/10.5194/hess-27-4227-2023, 2023
Short summary
Short summary
To overcome the computational cost of numerical models, we propose a deep-learning approach inspired by hydraulic models that can simulate the spatio-temporal evolution of floods. We show that the model can rapidly predict dike breach floods over different topographies and breach locations, with limited use of ground-truth data.
Roberto Bentivoglio, Elvin Isufi, Sebastian Nicolaas Jonkman, and Riccardo Taormina
Hydrol. Earth Syst. Sci., 26, 4345–4378, https://doi.org/10.5194/hess-26-4345-2022, https://doi.org/10.5194/hess-26-4345-2022, 2022
Short summary
Short summary
Deep learning methods have been increasingly used in flood management to improve traditional techniques. While promising results have been obtained, our review shows significant challenges in building deep learning models that can (i) generalize across multiple scenarios, (ii) account for complex interactions, and (iii) perform probabilistic predictions. We argue that these shortcomings could be addressed by transferring recent fundamental advancements in deep learning to flood mapping.
Roberto Bentivoglio, Elvin Isufi, Sebastian Nicolaas Jonkman, and Riccardo Taormina
Hydrol. Earth Syst. Sci. Discuss., https://doi.org/10.5194/hess-2021-614, https://doi.org/10.5194/hess-2021-614, 2021
Manuscript not accepted for further review
Short summary
Short summary
Deep Learning methods have been increasingly used in flood mapping as an alternative to traditional modeling techniques. While promising results have been obtained, our review shows significant challenges in building Deep Learning models that can generalize across multiple scenarios, account for complex interactions, and provide probabilistic predictions. We argue that these shortcomings could be addressed by transferring recent fundamental advancements in Deep Learning.
Laurie van Gijzen, Alexander M. Bakker, and Sebastiaan N. Jonkman
EGUsphere, https://doi.org/10.5194/egusphere-2025-5801, https://doi.org/10.5194/egusphere-2025-5801, 2025
Short summary
Short summary
Future sea level rise will likely challenge many low-lying deltas in the world. A possible mitigation strategy discharging the incoming river discharge with large pump-sluice stations. In this study we determine the required pump and sluice capacity by including the new concept of operational discharge capacity. We found that including operational discharge capacity led to an significant increase in flood frequency, making it crucial to account for in the design of pump-sluice stations.
Roberto Bentivoglio, Elvin Isufi, Sebastiaan Nicolas Jonkman, and Riccardo Taormina
Nat. Hazards Earth Syst. Sci., 25, 335–351, https://doi.org/10.5194/nhess-25-335-2025, https://doi.org/10.5194/nhess-25-335-2025, 2025
Short summary
Short summary
Deep learning methods are increasingly used as surrogates for spatio-temporal flood models but struggle with generalization and speed. Here, we propose a multi-resolution approach using graph neural networks that predicts dike breach floods across different meshes, topographies, and boundary conditions with high accuracy and up to 1000× speed-ups. The model also generalizes to larger more complex case studies with just one additional simulation for fine-tuning.
Nienke Tempel, Laurène Bouaziz, Riccardo Taormina, Ellis van Noppen, Jasper Stam, Eric Sprokkereef, and Markus Hrachowitz
Hydrol. Earth Syst. Sci., 28, 4577–4597, https://doi.org/10.5194/hess-28-4577-2024, https://doi.org/10.5194/hess-28-4577-2024, 2024
Short summary
Short summary
This study explores the impact of climatic variability on root zone water storage capacities and, thus, on hydrological predictions. Analysing data from 286 areas in Europe and the US, we found that, despite some variations in root zone storage capacity due to changing climatic conditions over multiple decades, these changes are generally minor and have a limited effect on water storage and river flow predictions.
Tim Hans Martin van Emmerik, Tim Willem Janssen, Tianlong Jia, Thank-Khiet L. Bui, Riccardo Taormina, Hong-Q. Nguyen, and Louise Jeanne Schreyers
EGUsphere, https://doi.org/10.5194/egusphere-2024-2270, https://doi.org/10.5194/egusphere-2024-2270, 2024
Preprint archived
Short summary
Short summary
Plastic pollution has negative effects on the environment. Tropical rivers around the world suffer from both plastic pollution and invasive water hyacinths. Water hyacinths grow rapidly and form dense floating mats. Using drones, cameras and AI, we show that along the Saigon river, Vietnam, the majority of floating plastic pollution is carried by water hyacinths. Better understanding water hyacinth-plastic trapping supports improving pollution monitoring and management strategies.
Hanqing Xu, Elisa Ragno, Sebastiaan N. Jonkman, Jun Wang, Jeremy D. Bricker, Zhan Tian, and Laixiang Sun
Hydrol. Earth Syst. Sci., 28, 3919–3930, https://doi.org/10.5194/hess-28-3919-2024, https://doi.org/10.5194/hess-28-3919-2024, 2024
Short summary
Short summary
A coupled statistical–hydrodynamic model framework is employed to quantitatively evaluate the sensitivity of compound flood hazards to the relative timing of peak storm surges and rainfall. The findings reveal that the timing difference between these two factors significantly affects flood inundation depth and extent. The most severe inundation occurs when rainfall precedes the storm surge peak by 2 h.
Roberto Bentivoglio, Elvin Isufi, Sebastiaan Nicolas Jonkman, and Riccardo Taormina
Hydrol. Earth Syst. Sci., 27, 4227–4246, https://doi.org/10.5194/hess-27-4227-2023, https://doi.org/10.5194/hess-27-4227-2023, 2023
Short summary
Short summary
To overcome the computational cost of numerical models, we propose a deep-learning approach inspired by hydraulic models that can simulate the spatio-temporal evolution of floods. We show that the model can rapidly predict dike breach floods over different topographies and breach locations, with limited use of ground-truth data.
Roberto Bentivoglio, Elvin Isufi, Sebastian Nicolaas Jonkman, and Riccardo Taormina
Hydrol. Earth Syst. Sci., 26, 4345–4378, https://doi.org/10.5194/hess-26-4345-2022, https://doi.org/10.5194/hess-26-4345-2022, 2022
Short summary
Short summary
Deep learning methods have been increasingly used in flood management to improve traditional techniques. While promising results have been obtained, our review shows significant challenges in building deep learning models that can (i) generalize across multiple scenarios, (ii) account for complex interactions, and (iii) perform probabilistic predictions. We argue that these shortcomings could be addressed by transferring recent fundamental advancements in deep learning to flood mapping.
Julius Schlumberger, Christian Ferrarin, Sebastiaan N. Jonkman, Manuel Andres Diaz Loaiza, Alessandro Antonini, and Sandra Fatorić
Nat. Hazards Earth Syst. Sci., 22, 2381–2400, https://doi.org/10.5194/nhess-22-2381-2022, https://doi.org/10.5194/nhess-22-2381-2022, 2022
Short summary
Short summary
Flooding has serious impacts on the old town of Venice. This paper presents a framework combining a flood model with a flood-impact model to support improving protection against future floods in Venice despite the recently built MOSE barrier. Applying the framework to seven plausible flood scenarios, it was found that individual protection has a significant damage-mediating effect if the MOSE barrier does not operate as anticipated. Contingency planning thus remains important in Venice.
Manuel Andres Diaz Loaiza, Jeremy D. Bricker, Remi Meynadier, Trang Minh Duong, Rosh Ranasinghe, and Sebastiaan N. Jonkman
Nat. Hazards Earth Syst. Sci., 22, 345–360, https://doi.org/10.5194/nhess-22-345-2022, https://doi.org/10.5194/nhess-22-345-2022, 2022
Short summary
Short summary
Extratropical cyclones are one of the major causes of coastal floods in Europe and the world. Understanding the development process and the flooding of storm Xynthia, together with the damages that occurred during the storm, can help to forecast future losses due to other similar storms. In the present paper, an analysis of shallow water variables (flood depth, velocity, etc.) or coastal variables (significant wave height, energy flux, etc.) is done in order to develop damage curves.
Christopher H. Lashley, Sebastiaan N. Jonkman, Jentsje van der Meer, Jeremy D. Bricker, and Vincent Vuik
Nat. Hazards Earth Syst. Sci., 22, 1–22, https://doi.org/10.5194/nhess-22-1-2022, https://doi.org/10.5194/nhess-22-1-2022, 2022
Short summary
Short summary
Many coastlines around the world have shallow foreshores (e.g. salt marshes and mudflats) that reduce storm waves and the risk of coastal flooding. However, most of the studies that tried to quantify this effect have excluded the influence of very long waves, which often dominate in shallow water. Our newly developed framework addresses this oversight and suggests that safety along these coastlines may be overestimated, since these very long waves are largely neglected in flood risk assessments.
Roberto Bentivoglio, Elvin Isufi, Sebastian Nicolaas Jonkman, and Riccardo Taormina
Hydrol. Earth Syst. Sci. Discuss., https://doi.org/10.5194/hess-2021-614, https://doi.org/10.5194/hess-2021-614, 2021
Manuscript not accepted for further review
Short summary
Short summary
Deep Learning methods have been increasingly used in flood mapping as an alternative to traditional modeling techniques. While promising results have been obtained, our review shows significant challenges in building Deep Learning models that can generalize across multiple scenarios, account for complex interactions, and provide probabilistic predictions. We argue that these shortcomings could be addressed by transferring recent fundamental advancements in Deep Learning.
Cited articles
Apel, H., Thieken, A. H., Merz, B., and Blöschl, G.: A probabilistic modelling system for assessing flood risks, Nat. Hazards, 38, 79–100, https://doi.org/10.1007/s11069-005-8603-7, 2006. a, b
Bentivoglio, R.: mSWE-GNN: version for paper “Probabilistic flood hazard mapping for dike-breach floods via graph neural networks” In Natural Hazards and Earth System Sciences, Zenodo [code and data set], https://doi.org/10.5281/zenodo.20053958, 2026. a, b
Bentivoglio, R., Isufi, E., Jonkman, S. N., and Taormina, R.: Deep learning methods for flood mapping: a review of existing applications and future research directions, Hydrol. Earth Syst. Sci., 26, 4345–4378, https://doi.org/10.5194/hess-26-4345-2022, 2022. a
Besseling, L. S., Bomers, A., Warmink, J. J., and Hulscher, S. J.: A conceptual model to quantify probabilistic dike breach outflow, Nat. Hazards, 1–29, https://doi.org/10.1007/s11069-025-07500-z, 2025. a, b
Brufau, P., Vázquez-Cendón, M., and García-Navarro, P.: A numerical model for the flooding and drying of irregular domains, Int. J. Numer. Meth. Fl., 39, 247–275, https://doi.org/10.1002/fld.285, 2002. a
Burrichter, B., Hofmann, J., Koltermann da Silva, J., Niemann, A., and Quirmbach, M.: A Spatiotemporal Deep Learning Approach for Urban Pluvial Flood Forecasting with Multi-Source Data, Water, 15, 1760, https://doi.org/10.3390/w15091760, 2023. a
Cache, T., Gomez, M. S., Beucler, T., Blagojevic, J., Leitao, J. P., and Peleg, N.: Enhancing generalizability of data-driven urban flood models by incorporating contextual information, Hydrol. Earth Syst. Sci., 28, 5443–-5458, https://doi.org/10.5194/hess-28-5443-2024, 2024. a
Cao, X., Wang, B., Yao, Y., Zhang, L., Xing, Y., Mao, J., Zhang, R., Fu, G., Borthwick, A. G. L., and Qin, H.: U-Rnn High-Resolution Spatiotemporal Nowcasting of Urban Flooding, SSRN [preprint], https://doi.org/10.2139/ssrn.4935234, 2024. a
de Moel, H., Bouwer, L. M., and Aerts, J. C.: Uncertainty and sensitivity of flood risk calculations for a dike ring in the south of the Netherlands, Sci. Total Environ., 473, 224–234, https://doi.org/10.1016/j.scitotenv.2013.12.015, 2014. a, b, c, d
Deltares: Delft3D-FM User Manual, https://content.oss.deltares.nl/delft3d/D-Flow_FM_User_Manual.pdf (last access: 28 January 2026), 2022. a
den Heijer, F. and Kok, M.: Assessment of ductile dike behavior as a novel flood risk reduction measure, Risk Anal., 43, 1779–1794, https://doi.org/10.1111/risa.14071, 2023. a
do Lago, C. A., Giacomoni, M. H., Bentivoglio, R., Taormina, R., Gomes, M. N., and Mendiondo, E. M.: Generalizing rapid flood predictions to unseen urban catchments with conditional generative adversarial networks, J. Hydrol., 129276, https://doi.org/10.1016/j.jhydrol.2023.129276, 2023. a
D'Oria, M. and Maranzoni, A.: Probabilistic assessment of flood hazard due to levee breaches using fragility functions, Water Resour. Res., 55, 8740–8764, https://doi.org/10.1029/2019WR025369, 2019. a, b, c
Dottori, F., Alfieri, L., Bianchi, A., Skoien, J., and Salamon, P.: A new dataset of river flood hazard maps for Europe and the Mediterranean Basin, Earth Syst. Sci. Data, 14, 1549–1569, https://doi.org/10.5194/essd-14-1549-2022, 2022. a
EuroHPC: Large Unified Modern Infrastructure (LUMI), https://www.lumi-supercomputer.eu/, last access: 5 August 2025. a
Ferrari, A., Dazzi, S., Vacondio, R., and Mignosa, P.: Enhancing the resilience to flooding induced by levee breaches in lowland areas: a methodology based on numerical modelling, Nat. Hazards Earth Syst. Sci., 20, 59–72, https://doi.org/10.5194/nhess-20-59-2020, 2020. a
Fey, M. and Lenssen, J. E.: Fast graph representation learning with PyTorch Geometric, arXiv [preprint], https://doi.org/10.48550/arXiv.1903.02428, 2019. a
Gao, W., Liao, Y., Chen, Y., Lai, C., He, S., and Wang, Z.: Enhancing transparency in data-driven urban pluvial flood prediction using an explainable CNN model, J. Hydrol., 645, 132228, https://doi.org/10.1016/j.jhydrol.2024.132228, 2024. a
Gibbons, S. J., Lorito, S., Macías, J., Løvholt, F., Selva, J., Volpe, M., Sánchez-Linares, C., Babeyko, A., Brizuela, B., Cirella, A., Castro, M. J., de la Asunción, M., Lanucara, P., Glimsdal, S., Lorenzino, M. C., Nazaria, M., Pizzimenti, L., Romano, F., Scala, A., Tonini, R., Manuel González Vida, J., and Vöge, M.: Probabilistic Tsunami Hazard Analysis: High Performance Computing for Massive Scale Inundation Simulations, Front. Earth Sci., 8, https://doi.org/10.3389/feart.2020.591549, 2020. a
Guo, Z., Leitão, J. P., Simões, N. E., and Moosavi, V.: Data-driven flood emulation: Speeding up urban flood predictions by deep convolutional neural networks, J. Flood Risk Manag., 1–14, https://doi.org/10.1111/jfr3.12684, 2020. a
Guo, Z., Moosavi, V., and Leitão, J. P.: Data-driven rapid flood prediction mapping with catchment generalizability, J. Hydrol., 609, 127726, https://doi.org/10.1016/J.JHYDROL.2022.127726, 2022. a
Hall, J. and Solomatine, D.: A framework for uncertainty analysis in flood risk management decisions, International Journal of River Basin Management, 6, 85–98, https://doi.org/10.1080/15715124.2008.9635339, 2008. a, b
Hall, J. W., Manning, L. J., and Hankin, R. K.: Bayesian calibration of a flood inundation model using spatial data, Water Resour. Res., 47, https://doi.org/10.1029/2009WR008541, 2011. a
Hegnauer, M., Beersma, J., Van den Boogaard, H., Buishand, T., and Passchier, R.: Generator of Rainfall and Discharge Extremes (GRADE) for the Rhine and Meuse basins, Final report of GRADE 2.0, https://publications.deltares.nl/1209424_004_0018.pdf (last access: 28 January 2026), 2014. a
Huang, T. and Merwade, V.: Beyond a Fixed Number: Investigating Uncertainty in Popular Evaluation Metrics of Ensemble Flood Modeling Using Bootstrapping Analysis, J. Flood Risk Manag., 17, e12982, https://doi.org/10.1111/jfr3.12982, 2024. a
Jongejan, R. and Maaskant, B.: Quantifying flood risks in the Netherlands, Risk Anal., 35, 252–264, https://doi.org/10.1111/risa.12285, 2015. a
Kazadi, A., Doss-Gollin, J., and da Silva, A. L.: Pluvial Flood Emulation with Hydraulics-informed Message Passing, in: Forty-first International Conference on Machine Learning, https://openreview.net/forum?id=kIHIA6Lr0B (last access: 1 November 2025), 2024. a
Kingma, D. P. and Ba, J.: Adam: A method for stochastic optimization, arXiv [preprint], https://doi.org/10.48550/arXiv.1412.6980, 2014. a
Lhomme, J., Sayers, P., Gouldby, B., Wills, M., and Mulet-Marti, J.: Recent development and application of a rapid flood spreading method, Flood Risk Management: Research and Practice, 15–24, https://doi.org/10.1201/9780203883020.ch2, 2008. a
Liao, Y., Wang, Z., Chen, X., and Lai, C.: Fast simulation and prediction of urban pluvial floods using a deep convolutional neural network model, J. Hydrol., 624, 129945, https://doi.org/10.1016/j.jhydrol.2023.129945, 2023. a
Mazzoleni, M., Bacchi, B., Barontini, S., Di Baldassarre, G., Pilotti, M., and Ranzi, R.: Flooding hazard mapping in floodplain areas affected by piping breaches in the Po River, Italy, J. Hydrol. Eng., 19, 717–731, https://doi.org/10.1061/(ASCE)HE.1943-5584.0000840, 2014. a, b
Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., Desmaison, A., Kopf, A., Yang, E., DeVito, Z., Raison, M., Tejani, A., Chilamkurthy, S., Steiner, B., Fang, L., Bai, J., and Chintala, S.: PyTorch: An Imperative Style, High-Performance Deep Learning Library, Advances in neural information processing systems, 32, https://proceedings.neurips.cc/paper_files/paper/2019/file/bdbca288fee7f92f2bfa9f7012727740-Paper.pdf (last access: 1 November 2025), 2019. a
Pianforini, M., Dazzi, S., Pilzer, A., and Vacondio, R.: FloodSformer: A transformer-based data-driven model for predicting the 2-D dynamics of fluvial floods, Environ. Modell. Softw., 193, 106599, https://doi.org/10.1016/j.envsoft.2025.106599, 2025. a, b
Savage, J. T. S., Bates, P., Freer, J., Neal, J., and Aronica, G.: When does spatial resolution become spurious in probabilistic flood inundation predictions?, Hydrol. Process., 30, 2014–2032, https://doi.org/10.1002/hyp.10749, 2016. a, b, c
Shao, Y., Chen, J., Zhang, T., Yu, T., and Chu, S.: Advancing rapid urban flood prediction: a spatiotemporal deep learning approach with uneven rainfall and attention mechanism, J. Hydroinform., 26, 1409–1424, https://doi.org/10.2166/hydro.2024.024, 2024. a
Song, W., Guan, M., and Yu, D.: SwinFlood: A hybrid CNN-Swin Transformer model for rapid spatiotemporal flood simulation, J. Hydrol., 660, https://doi.org/10.1016/j.jhydrol.2025.133280, 2025. a
Wang, Z., Lyu, H., Fu, G., and Zhang, C.: Time-guided convolutional neural networks for spatiotemporal urban flood modelling, J. Hydrol., 645, 132250, https://doi.org/10.1016/j.jhydrol.2024.132250, 2024. a
Wei, G., Xia, W., He, B., and Shoemaker, C.: Quick large-scale spatiotemporal flood inundation computation using integrated Encoder–Decoder LSTM with time distributed spatial output models, J. Hydrol., 130993, https://doi.org/10.1016/j.jhydrol.2024.130993, 2024. a
Westerhof, S. G., Booij, M. J., Van den Berg, M. C., Huting, R. J., and Warmink, J. J.: Uncertainty analysis of risk-based flood safety standards in the Netherlands through a scenario-based approach, International Journal of River Basin Management, 21, 559–574, https://doi.org/10.1080/15715124.2022.2060243, 2023. a
Xu, L. and Gao, L.: A hybrid surrogate model for real-time coastal urban flood prediction: An application to Macao, J. Hydrol., 642, 131863, https://doi.org/10.1016/j.jhydrol.2024.131863, 2024. a
Short summary
Obtaining probabilistic flood maps with numerical models is very time-consuming. Deep learning models can speed this up, but their predictions are difficult to verify without reference data, and they ignore structures like dikes or canals. This work introduces a mass-based validation measure to assess prediction plausibility and adapts a graph-based model to include hydraulic structures, enabling realistic, large-scale probabilistic flood mapping in the Netherlands.
Obtaining probabilistic flood maps with numerical models is very time-consuming. Deep learning...
Altmetrics
Final-revised paper
Preprint