pith. machine review for the scientific record. sign in

arxiv: 2604.21028 · v1 · submitted 2026-04-22 · 💻 cs.LG · cs.AI· cs.CV

Recognition: unknown

A Deep U-Net Framework for Flood Hazard Mapping Using Hydraulic Simulations of the Wupper Catchment

Christian Lammers , Fernando Ar\'evalo , Leonie M\"arker-Neuhaus , Daniel Heinenberg , Christian F\"orster , Karl-Heinz Spies
Authors on Pith no claims yet

Pith reviewed 2026-05-10 00:47 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.CV
keywords U-Netflood hazard mappingsurrogate modelhydraulic simulationsdeep learningwater level predictionWupper catchment
0
0 comments X

The pith

A U-Net model trained on hydraulic simulations predicts maximum water levels on the Wupper catchment with accuracy close to full simulations but far lower computation time.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to prove that a deep learning surrogate can stand in for traditional hydraulic models when mapping flood hazards. It trains and tunes a U-Net on simulation outputs from the Wupper catchment to forecast the highest water level at every grid point. If the approach works, flood maps could be produced in seconds instead of the hours or days required by physics-based solvers. The authors run targeted experiments on architecture choices, how to split the domain into patches, and how to prepare the training data. The result is presented as a practical, faster route to reliable hazard information for the tested region.

Core claim

A U-Net surrogate model, after optimization of its architecture, patch generation, and data handling, approximates the hydraulic simulation's output and delivers maximum water level grids for the Wupper catchment that are comparable in accuracy to the original simulations while requiring substantially less computation.

What carries the argument

The U-Net neural network configured as a surrogate that learns to map input conditions directly to the grid of maximum water levels produced by the hydraulic model.

If this is right

  • Flood hazard maps can be generated on demand rather than only for a small number of pre-selected events.
  • Emergency planners gain the ability to evaluate many more rainfall scenarios within the same computing budget.
  • The surrogate approach reduces the barrier to updating hazard assessments when new topographic or land-use data become available.
  • Similar models could be retrained for other catchments once a modest number of full simulations exist for those areas.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same patch-based U-Net training could be applied to predict additional variables such as flow velocity or inundation duration if those outputs are added to the training data.
  • Coupling the model with live rainfall forecasts would allow operational flood warnings to be issued faster than physics-only systems permit.
  • Performance on catchments with very different topography or land cover would likely require new training runs rather than zero-shot transfer.

Load-bearing premise

A U-Net trained on a limited collection of hydraulic simulations for the Wupper catchment will keep its accuracy when it encounters rainfall patterns or boundary conditions not present in the training set.

What would settle it

Produce a fresh set of hydraulic simulations for the Wupper catchment that use rainfall or boundary conditions absent from the training data, then measure the difference between the U-Net's predicted water level grid and the simulation results.

Figures

Figures reproduced from arXiv: 2604.21028 by Christian F\"orster, Christian Lammers, Daniel Heinenberg, Fernando Ar\'evalo, Karl-Heinz Spies, Leonie M\"arker-Neuhaus.

Figure 1
Figure 1. Figure 1: Study areas: Beyenburg (BEY), Kluserbrucke (KLU), and Kohlfurth (KOL) ¨ TABLE I: Geographical characteristics of the three study areas. Location Area (km2 ) Elevation (m NHN) Image size Avg. Sim. Time (min) Beyenburg (BEY) 1.73 183.25 – 226.39 2041 × 849 15 Kluserbrucke (KLU) 6.88 117.28 – 194.51 ¨ 10402 × 6617 20 Kohlfurth (KOL) 3.77 91.83 – 175.91 3470 × 10877 17 TABLE II: Hydrological characteristics of… view at source ↗
Figure 2
Figure 2. Figure 2: Test RMSE’s for the cross-validation experiment. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visual comparison of inference methods for a discharge of [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Training and validation loss per epoch of the final surrogate model. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative performance of the surrogate model’s ability to approximate the hydraulic model. Figure (a) shows the [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Spatial distribution of prediction error for the test discharge of 65 m³/s in the BEY domain. Red areas indicate over [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Spatial distribution of prediction error near a bridge structure. The figure shows a cluster of large positive errors (red), [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Model performance on validation set for different patch amounts per training image. [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Final performance per patch amount on Beyenburg (BEY) test set. [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
read the original abstract

The increasing frequency and severity of global flood events highlights the need for the development of rapid and reliable flood prediction tools. This process traditionally relies on computationally expensive hydraulic simulations. This research presents a prediction tool by developing a deep-learning based surrogate model to accurately and efficiently predict the maximum water level across a grid. This was achieved by conducting a series of experiments to optimize a U-Net architecture, patch generation, and data handling for approximating a hydraulic model. This research demonstrates that a deep learning surrogate model can serve as a computationally efficient alternative to traditional hydraulic simulations. The framework was tested using hydraulic simulations of the Wupper catchment in the North-Rhein Westphalia region (Germany), obtaining comparable results.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper develops a U-Net-based deep learning surrogate model trained on hydraulic simulations of the Wupper catchment (North-Rhine Westphalia, Germany) to predict maximum water levels on a grid for flood hazard mapping. Through experiments optimizing U-Net architecture, patch generation, and data handling, it claims the surrogate provides comparable accuracy to traditional hydraulic models while being computationally efficient.

Significance. A validated surrogate could enable rapid flood predictions for operational use, addressing the computational cost of traditional simulations amid increasing flood frequency. However, without quantitative metrics or generalization tests, the work does not yet demonstrate this efficiency advantage or reliability under new conditions.

major comments (3)
  1. [Abstract] Abstract and results sections: the assertion of 'comparable results' is unsupported by any reported quantitative error metrics (e.g., RMSE, MAE on water levels), validation splits, or baseline comparisons against the hydraulic model, preventing evaluation of the central accuracy claim.
  2. [Methods] Methods and experiments: no description or results are given for test cases with rainfall intensities, durations, or boundary conditions outside the training distribution, leaving the generalization assumption (required for operational surrogate use) untested.
  3. [Results] Results: the optimization of patch handling and architecture is performed on a finite set of simulations, but no ablation or sensitivity analysis quantifies how performance varies with input distribution shifts, undermining the efficiency-alternative claim.
minor comments (2)
  1. [Methods] Clarify the exact loss function and any physics-informed components (if present) in the U-Net training description.
  2. [Results] Add a table or figure with explicit numerical performance values and comparison to the hydraulic baseline.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback, which identifies key areas where the manuscript can be strengthened through additional quantitative support and clearer scoping of the experiments. We address each major comment point by point below, indicating planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract and results sections: the assertion of 'comparable results' is unsupported by any reported quantitative error metrics (e.g., RMSE, MAE on water levels), validation splits, or baseline comparisons against the hydraulic model, preventing evaluation of the central accuracy claim.

    Authors: We agree that the claim of 'comparable results' would be more robust with explicit quantitative metrics. The manuscript includes visual side-by-side comparisons of predicted versus simulated maximum water levels in the results section, but does not report numerical error metrics such as RMSE or MAE, nor details on validation splits or direct baseline error comparisons. In the revised manuscript, we will add these quantitative evaluations computed on a held-out test portion of the simulations, along with explicit baseline comparisons to the original hydraulic model. revision: yes

  2. Referee: [Methods] Methods and experiments: no description or results are given for test cases with rainfall intensities, durations, or boundary conditions outside the training distribution, leaving the generalization assumption (required for operational surrogate use) untested.

    Authors: The current work optimizes and evaluates the U-Net surrogate strictly on the distribution of the available hydraulic simulations for the Wupper catchment. No out-of-distribution test cases (e.g., unseen rainfall intensities, durations, or boundary conditions) are described or evaluated, as the primary objective was to demonstrate efficient approximation within the provided data regime. We acknowledge this limits claims about operational generalization. In the revision, we will add an explicit description of the training data distribution and a dedicated limitations subsection discussing generalization, while noting out-of-distribution testing as important future work. revision: partial

  3. Referee: [Results] Results: the optimization of patch handling and architecture is performed on a finite set of simulations, but no ablation or sensitivity analysis quantifies how performance varies with input distribution shifts, undermining the efficiency-alternative claim.

    Authors: The architecture and patch-handling ablations were performed on the finite set of available simulations, showing improvements in approximation quality and speed within that set. No sensitivity analysis for input distribution shifts was included. The efficiency advantage is demonstrated via inference-time comparisons to the full hydraulic model under matching conditions. In the revised results section, we will clarify the scope of the ablations, add any feasible sensitivity notes based on the existing data, and explicitly state that the efficiency claim applies to similar input regimes. revision: partial

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper trains a U-Net surrogate on finite hydraulic simulation outputs for the Wupper catchment and evaluates it on held-out simulation cases. This is standard supervised learning with no equations that reduce predictions to fitted parameters by construction, no self-definitional loops, and no load-bearing self-citations or imported uniqueness theorems. The central claim (DL as efficient alternative) rests on empirical performance metrics rather than tautological renaming or ansatz smuggling. Generalization limits are an assumption risk, not a circularity issue.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the standard assumption that a sufficiently large and representative set of hydraulic simulation runs exists and that the U-Net can learn the underlying mapping without explicit physics constraints.

axioms (2)
  • domain assumption Hydraulic simulation outputs constitute a reliable ground-truth distribution for training
    Invoked when the authors state that the model approximates the hydraulic model
  • domain assumption Patch-based training on grid data preserves spatial coherence needed for water-level prediction
    Mentioned in the description of data handling and patch generation

pith-pipeline@v0.9.0 · 5443 in / 1295 out tokens · 29256 ms · 2026-05-10T00:47:00.628092+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 5 canonical work pages · 2 internal anchors

  1. [1]

    Weather and climate extreme events in a changing climate,

    S. I. Seneviratne, X. Zhang, M. Adnan, W. Badi, C. Dereczynski, A. D. Luca, S. Ghosh, I. Iskandar, J. Kossin, S. Lewiset al., “Weather and climate extreme events in a changing climate,” 2021

    2021

  2. [2]

    Application of machine learning-based surrogate models for urban flood depth modeling in ho chi minh city, vietnam,

    T. Q. Dang, B. H. Tran, Q. N. Le, T. D. Dang, A. H. Tanim, Q. B. Pham, V . H. Bui, S. T. Mai, P. N. Thanh, and D. T. Anh, “Application of machine learning-based surrogate models for urban flood depth modeling in ho chi minh city, vietnam,”Applied Soft Computing, vol. 150, p. 111031, 2024

    2024

  3. [3]

    Fatal in- cidents during the flood of july 2021 in north rhine- westphalia, germany: what can be learnt for future flood risk management?

    A. Thieken, P. Bubeck, and M.-L. Zenker, “Fatal in- cidents during the flood of july 2021 in north rhine- westphalia, germany: what can be learnt for future flood risk management?”Journal of Coastal and Riverine Flood Risk, vol. 2, 2023

    2021

  4. [4]

    Informationen zum hochwasserereignis (14./15. juli 2021) in wuppertal,

    StadtWuppertal, “Informationen zum hochwasserereignis (14./15. juli 2021) in wuppertal,” 2025. [Online]. Available: https://www.wuppertal.de/microsite/ klimaschutz/klimawandel-und-klimaanpassung/ weitere-inhalte-klimawandel-klimaanpassung/ starkregenereignis-in-wuppertal-am-15.07.2021.php

    2021

  5. [5]

    A spatiotemporal deep learning approach for urban pluvial flood forecasting with multi- source data,

    B. Burrichter, J. Hofmann, J. K. da Silva, A. Niemann, and M. Quirmbach, “A spatiotemporal deep learning approach for urban pluvial flood forecasting with multi- source data,”Water (Switzerland), vol. 15, 2023

    2023

  6. [6]

    Spatiotemporal flood depth and velocity dynamics using a convolutional neural network within a sequential deep-learning framework,

    M. M. Fathi, Z. Liu, A. M. Fernandes, M. T. Hren, D. O. Terry, C. Nataraj, and V . Smith, “Spatiotemporal flood depth and velocity dynamics using a convolutional neural network within a sequential deep-learning framework,” Environmental Modelling and Software, vol. 185, 2025

    2025

  7. [7]

    Bergisches hochwassermeldesystem 4.0 (hws 4.0),

    KI.NRW, “Bergisches hochwassermeldesystem 4.0 (hws 4.0),” 2025. [Online]. Available: https://bergisches-hws. de/

    2025

  8. [8]

    Bergisches hochwassermeldesystem 4.0,

    BHWS4.0, “Bergisches hochwassermeldesystem 4.0,”

  9. [9]

    Available: https://bergisches-hws.de/

    [Online]. Available: https://bergisches-hws.de/

  10. [10]

    As- sessment of surrogate models for flood inundation: The physics-guided lsg model vs. state-of-the-art machine learning models,

    N. Fraehr, Q. J. Wang, W. Wu, and R. Nathan, “As- sessment of surrogate models for flood inundation: The physics-guided lsg model vs. state-of-the-art machine learning models,”Water Research, vol. 252, 2024

    2024

  11. [11]

    Training machine learning surro- gate models from a high-fidelity physics-based model: Application for real-time street-scale flood prediction in an urban coastal community,

    F. T. Zahura, J. L. Goodall, J. M. Sadler, Y . Shen, M. M. Morsy, and M. Behl, “Training machine learning surro- gate models from a high-fidelity physics-based model: Application for real-time street-scale flood prediction in an urban coastal community,”Water Resources Research, vol. 56, no. 10, p. e2019WR027038, 2020

    2020

  12. [12]

    Deep learning methods for flood mapping: a review of existing applications and future research directions,

    R. Bentivoglio, E. Isufi, S. N. Jonkman, and R. Taormina, “Deep learning methods for flood mapping: a review of existing applications and future research directions,” Hydrology and Earth System Sciences Discussions, vol. 2022, pp. 1–50, 2022

    2022

  13. [13]

    Flo-sr: Deep learning-based urban flood super-resolution model,

    H. Choi, H. Woo, M. Kim, H. Ryu, J. H. Lee, S. Lee, and S. J. Noh, “Flo-sr: Deep learning-based urban flood super-resolution model,”Journal of Hydrology, vol. 661, 2025

    2025

  14. [14]

    A rapid flood inundation model for hazard mapping based on least squares support vector machine regression,

    M. Berm ´udez, L. Cea, and J. Puertas, “A rapid flood inundation model for hazard mapping based on least squares support vector machine regression,”Journal of Flood Risk Management, vol. 12, p. e12522, 2019. [Online]. Available: https://onlinelibrary.wiley.com/doi/ abs/10.1111/jfr3.12522

    work page doi:10.1111/jfr3.12522 2019

  15. [15]

    Genetic programming for hydrological applications: to model or to forecast that is the question,

    H. M. V . V . Herath, J. Chadalawada, and V . Babovic, “Genetic programming for hydrological applications: to model or to forecast that is the question,”Journal of Hydroinformatics, vol. 23, no. 4, pp. 740–763, 2021

    2021

  16. [16]

    A deep convolutional neural network model for rapid prediction of fluvial flood inundation,

    S. Kabir, S. Patidar, X. Xia, Q. Liang, J. Neal, and G. Pender, “A deep convolutional neural network model for rapid prediction of fluvial flood inundation,”Journal of Hydrology, vol. 590, p. 125481, 2020

    2020

  17. [17]

    Data-driven rapid flood prediction mapping with catchment generalizabil- ity,

    Z. Guo, V . Moosavi, and J. P. Leit ˜ao, “Data-driven rapid flood prediction mapping with catchment generalizabil- ity,”Journal of Hydrology, vol. 609, p. 127726, 2022

    2022

  18. [18]

    A deep learning model for predicting river flood depth and extent,

    H. Hosseiny, “A deep learning model for predicting river flood depth and extent,”Environmental Modelling & Software, vol. 145, p. 105186, 2021

    2021

  19. [19]

    A deep learning model for generalized surface water flooding across multiple return periods,

    S. Kabir, D. Wood, and S. Waller, “A deep learning model for generalized surface water flooding across multiple return periods,”Engineering Proceedings, vol. 39, no. 1, p. 94, 2023

    2023

  20. [20]

    Enhancing generalizability of data-driven urban flood models by incorporating contex- tual information,

    T. Cache, M. S. Gomez, T. Beucler, J. Blagojevic, J. P. Leitao, and N. Peleg, “Enhancing generalizability of data-driven urban flood models by incorporating contex- tual information,”Hydrology and Earth System Sciences, vol. 28, no. 24, pp. 5443–5458, 2024

    2024

  21. [21]

    U-Net: Convolutional Networks for Biomedical Image Segmentation

    O. Ronneberger, P. Fischer, and T. Brox, “U- Net: Convolutional Networks for Biomedical Image Segmentation,” May 2015, arXiv:1505.04597 [cs]. [Online]. Available: http://arxiv.org/abs/1505.04597

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  22. [22]

    Geodesic convolutional neural networks on riemannian manifolds,

    J. Masci, D. Boscaini, M. Bronstein, and P. Van- dergheynst, “Geodesic convolutional neural networks on riemannian manifolds,” inProceedings of the IEEE inter- national conference on computer vision workshops, 2015, pp. 37–45

    2015

  23. [23]

    Singha, S

    C. Singha, S. Sahoo, A. B. Mahtaj, A. Moghimi, M. Welzel, and A. Govind, “Advancing flood risk as- sessment: Multitemporal sar-based flood inventory gen- eration using transfer learning and hybrid fuzzy-ahp- machine learning for flood susceptibility mapping in the mahananda river basin,”Journal of Environmental Management, vol. 380, p. 124972, 2025

    2025

  24. [24]

    Digitales gel ¨andemodell,

    BezirksregierungNRW, “Digitales gel ¨andemodell,” 2025. [Online]. Available: https://www.bezreg-koeln.nrw. de/geobasis-nrw/produkte-und-dienste/hoehenmodelle/ digitale-gelaendemodelle/digitales-gelaendemodell

    2025

  25. [25]

    Data-driven flood emulation: Speeding up urban flood predictions by deep convolutional neural networks,

    Z. Guo, J. P. Leitao, N. E. Sim ˜oes, and V . Moosavi, “Data-driven flood emulation: Speeding up urban flood predictions by deep convolutional neural networks,” Journal of Flood Risk Management, vol. 14, no. 1, p. e12684, 2021

    2021

  26. [26]

    Integrating net rainfall calculation in deep learning- based surrogate modeling frameworks for 2d flood pre- diction,

    J. F. Farf ´an-Dur´an, C. Montalvo, L. Cea, and J. P. Leit˜ao, “Integrating net rainfall calculation in deep learning- based surrogate modeling frameworks for 2d flood pre- diction,”Journal of Hydrology, vol. 661, 2025

    2025

  27. [27]

    Real-time flood maps forecasting for dam-break sce- narios with a transformer-based deep learning model,

    M. Pianforini, S. Dazzi, A. Pilzer, and R. Vacondio, “Real-time flood maps forecasting for dam-break sce- narios with a transformer-based deep learning model,” Journal of Hydrology, vol. 635, 2024

    2024

  28. [28]

    A hybrid deep learning approach using vision transformer and u-net for flood segmentation,

    C. Dofitas, Y .-W. Kim, and Y .-C. Byun, “A hybrid deep learning approach using vision transformer and u-net for flood segmentation,”Computers, Materials and Continua, vol. 86, pp. 1–19, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/ pii/S1546221825012512

    2025

  29. [29]

    arXiv preprint arXiv:1805.12219

    B. Huang, D. Reichman, L. M. Collins, K. Bradbury, and J. M. Malof, “Tiling and stitching segmentation output for remote sensing: Basic challenges and recommendations,” 2019. [Online]. Available: https: //arxiv.org/abs/1805.12219

    work page arXiv 2019

  30. [30]

    River flow forecasting through conceptual models part i—a discussion of prin- ciples,

    J. E. Nash and J. V . Sutcliffe, “River flow forecasting through conceptual models part i—a discussion of prin- ciples,”Journal of hydrology, vol. 10, no. 3, pp. 282–290, 1970

    1970

  31. [31]

    Urban flood modeling and forecasting with deep neural operator and transfer learning,

    Q. Xu, L. F. D. V os, Y . Shi, N. R¨uther, A. Bronstert, and X. X. Zhu, “Urban flood modeling and forecasting with deep neural operator and transfer learning,”Journal of Hydrology, vol. 661, 2025

    2025

  32. [32]

    Surrogate modeling for rapid estimation of spatially-resolved flood damage: Application to a coastal region,

    S. Li, C. Ding, and J. Yang, “Surrogate modeling for rapid estimation of spatially-resolved flood damage: Application to a coastal region,”Journal of Hydrology, vol. 666, p. 134763, 2026. [Online]. Available: https://www.sciencedirect.com/science/article/ pii/S0022169425021031

    2026

  33. [33]

    Hydroas,

    Hydrotec, “Hydroas,” 2025. [Online]. Available: https: //www.hydrotec.de/software/hydroas/

    2025

  34. [34]

    Pytorch 2: Faster machine learning through dynamic python bytecode transformation and graph compilation,

    J. Ansel, E. Yang, H. He, N. Gimelshein, A. Jain, M. V oznesensky, B. Bao, P. Bell, D. Berard, E. Burovski, G. Chauhan, A. Chourdia, W. Constable, A. Desmaison, Z. Devito, E. Ellison, W. Feng, J. Gong, M. Gschwind, B. Hirsh, S. Huang, K. Kalambarkar, L. Kirsch, M. La- zos, M. Lezcano, Y . Liang, J. Liang, Y . Lu, C. K. Luk, B. Maher, Y . Pan, C. Puhrsch, ...

    2024

  35. [35]

    R., Millman, K

    C. R. Harris, K. J. Millman, S. J. van der Walt, R. Gommers, P. Virtanen, D. Cournapeau, E. Wieser, J. Taylor, S. Berg, N. J. Smith, R. Kern, M. Picus, S. Hoyer, M. H. van Kerkwijk, M. Brett, A. Haldane, J. F. del R ´ıo, M. Wiebe, P. Peterson, P. G ´erard-Marchant, K. Sheppard, T. Reddy, W. Weckesser, H. Abbasi, C. Gohlke, and T. E. Oliphant, “Array progr...

    work page doi:10.1038/s41586-020-2649-2 2020

  36. [36]

    Optuna: A next-generation hyperparameter optimization framework,

    T. Akiba, S. Sano, T. Yanase, T. Ohta, and M. Koyama, “Optuna: A next-generation hyperparameter optimization framework,” inProceedings of the ACM SIGKDD Inter- national Conference on Knowledge Discovery and Data Mining, 2019

    2019

  37. [37]

    Scikit-learn: Machine learning in python,

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vanderplas, A. Passos, D. Cour- napeau, M. Brucher, M. Perrot, and E. Duchesnay, “Scikit-learn: Machine learning in python,”Journal of Machine Learning Research, vol. 12, pp. 2825–2830, 2011

    2011

  38. [38]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization,” Jan. 2017, arXiv:1412.6980 [cs]. [Online]. Available: http://arxiv.org/abs/1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  39. [39]

    Satellite video remote sensing for flood model validation,

    C. Masafu and R. Williams, “Satellite video remote sensing for flood model validation,”Water Resources Research, vol. 60, 2024

    2024

  40. [40]

    U-flood–topographic deep learning for predicting urban pluvial flood water depth,

    R. L ¨owe, J. B ¨ohm, D. G. Jensen, J. Leandro, and S. H. Rasmussen, “U-flood–topographic deep learning for predicting urban pluvial flood water depth,”Journal of Hydrology, vol. 603, p. 126898, 2021

    2021

  41. [41]

    Deep transfer learning based on transformer for flood forecasting in data-sparse basins,

    Y . Xu, K. Lin, C. Hu, S. Wang, Q. Wu, L. Zhang, and G. Ran, “Deep transfer learning based on transformer for flood forecasting in data-sparse basins,”Journal of Hydrology, vol. 625, p. 129956, 2023

    2023

  42. [42]

    Accelerating urban flood inundation simulation under spatio-temporally varying rainstorms using convl- stm deep learning model,

    Y . Liao, Z. Wang, H. Yu, W. Gao, Z. Zeng, X. Li, and C. Lai, “Accelerating urban flood inundation simulation under spatio-temporally varying rainstorms using convl- stm deep learning model,”Water Resources Research, vol. 61, 2025

    2025

  43. [43]

    Swinflood: A hybrid cnn- swin transformer model for rapid spatiotemporal flood simulation,

    W. Song, M. Guan, and D. Yu, “Swinflood: A hybrid cnn- swin transformer model for rapid spatiotemporal flood simulation,”Journal of Hydrology, vol. 660, 2025

    2025

  44. [44]

    Pinn framework for urban flood depth prediction: integrating data-driven insights with physical constraints,

    W. Liu, Q. Wu, Z. Liu, X. Zheng, T. Liu, and H. Gao, “Pinn framework for urban flood depth prediction: integrating data-driven insights with physical constraints,”Journal of Hydrology, vol. 665, p. 134693,

  45. [45]

    Available: https://www.sciencedirect

    [Online]. Available: https://www.sciencedirect. com/science/article/pii/S0022169425020335

  46. [46]

    Physics-informed neural networks for solving flow problems modeled by the 2d shallow water equations without labeled data,

    X. Qi, G. A. M. de Almeida, and S. Maldonado, “Physics-informed neural networks for solving flow problems modeled by the 2d shallow water equations without labeled data,”Journal of Hydrology, vol. 636, p. 131263, 2024. [Online]. Available: https://www.sciencedirect.com/science/article/ pii/S0022169424006589

    2024

  47. [47]

    Physwrnet: A physics-guided deep learning framework for flood inundation mapping with sar and hydrodynamic simulations,

    L. Li, Y . Zhao, S. Qin, D. Pan, J. Zhang, A. Li, and Q. Hu, “Physwrnet: A physics-guided deep learning framework for flood inundation mapping with sar and hydrodynamic simulations,”Journal of Hydrology, vol. 665, p. 134662, 2026. [Online]. Available: https://www.sciencedirect.com/science/article/ pii/S0022169425020025

    2026