pith. sign in

arxiv: 2512.23964 · v2 · submitted 2025-12-30 · 💻 cs.LG · cs.AI

DUALFloodGNN: Physics-informed Graph Neural Network for Operational Flood Modeling

Carlo Malapad Acosta , Herath Mudiyanselage Viraj Vidura Herath , Jia Yu Lim , Abhishek Saha , Sanka Rasnayaka , Lucy Marshall This is my paper

Pith reviewed 2026-05-16 19:21 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords flood modelinggraph neural networksphysics-informed learningautoregressive simulationhydrologic variablesoperational forecastingspatiotemporal hydrodynamicsGNN
0
0 comments X

The pith

DUALFloodGNN adds explicit global and local physical constraints to a graph neural network that jointly predicts water volume and flow for faster, more accurate flood simulations.

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

The paper introduces DUALFloodGNN, a graph neural network architecture that embeds physical constraints at both global and local scales through dedicated loss terms. It uses a shared message-passing structure to predict water volume at nodes and flow along edges simultaneously. Training incorporates a multi-step loss with dynamic curriculum learning to support stable autoregressive rollouts over time. The resulting model reports better accuracy on multiple hydrologic variables than standard GNNs and prior flood-specific GNNs while preserving computational speed suitable for operational use.

Core claim

DUALFloodGNN embeds physical constraints at both global and local scales through explicit loss terms in a shared message-passing GNN framework that jointly predicts water volume at nodes and flow along edges; training with multi-step loss and dynamic curriculum learning produces improved accuracy on water volume, flow, and depth while retaining high computational efficiency for operational flood modeling.

What carries the argument

Shared message-passing framework with dual global-local physical constraint loss terms plus multi-step curriculum training.

Load-bearing premise

Explicit global and local physical loss terms will enforce realistic long-term autoregressive behavior rather than only satisfying penalties on the training distribution.

What would settle it

Run the trained model autoregressively on a held-out flood sequence for several hundred time steps and measure whether total water mass stays conserved and depth errors remain bounded without retraining.

Figures

Figures reproduced from arXiv: 2512.23964 by Abhishek Saha, Carlo Malapad Acosta, Herath Mudiyanselage Viraj Vidura Herath, Jia Yu Lim, Lucy Marshall, Sanka Rasnayaka.

Figure 1
Figure 1. Figure 1: Discretization of a target catchment. A structured mesh [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The DUALFloodGNN architecture. To predict [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Task-specific RMSE of select benchmarked GNN models for each timestep in the case study event. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Maximum water depth map of flood-specific GNN models [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
read the original abstract

Flood models inform strategic disaster management by simulating the spatiotemporal hydrodynamics of flooding. While physics-based numerical flood models are accurate, their substantial computational cost limits their use in operational settings where rapid predictions are essential. Models designed with graph neural networks (GNNs) provide both speed and accuracy while having the ability to process unstructured spatial domains. Given its flexible input and architecture, GNNs can be leveraged alongside physics-informed techniques with ease, significantly improving interpretability and generalizability. We introduce a novel flood GNN architecture, DUALFloodGNN, which embeds physical constraints at both global and local scales through explicit loss terms. The model jointly predicts water volume at nodes and flow along edges through a shared message-passing framework. To improve performance for autoregressive inference, model training is conducted with a multi-step loss enhanced with dynamic curriculum learning. Compared with standard GNN architectures and state-of-the-art GNN flood models, DUALFloodGNN achieves substantial improvements in predicting multiple hydrologic variables (e.g., water volume, flow, and depth) while maintaining high computational efficiency. The model is open sourced at https://github.com/acostacos/dual_flood_gnn. The dataset is open sourced at https://hdl.handle.net/2123/35293 with the DOI 10.25910/9xav-0s86.

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

2 major / 2 minor

Summary. The manuscript introduces DUALFloodGNN, a physics-informed graph neural network for flood modeling. It jointly predicts water volume at nodes and flow along edges via shared message-passing, embeds explicit global and local physical constraint terms in the loss, and trains with multi-step losses plus dynamic curriculum learning to support autoregressive inference. The central claim is that this yields substantial improvements over standard GNN architectures and prior GNN flood models in accuracy for hydrologic variables (volume, flow, depth) while preserving computational efficiency; code and data are open-sourced.

Significance. If the quantitative gains and long-horizon stability hold, the work would offer a practical advance for operational flood forecasting by delivering fast, interpretable predictions that respect physical constraints, potentially enabling real-time use where traditional numerical solvers are too slow. Open-sourcing of both model and dataset supports reproducibility and benchmarking.

major comments (2)
  1. [§4.2 and §5.3] §4.2 (Physics-informed losses) and §5.3 (Autoregressive evaluation): the dual global/local constraint losses are shown to reduce training error, but no explicit metrics (e.g., mass-balance violation, cumulative volume drift) are reported for multi-step rollouts on held-out events beyond the training distribution; this directly bears on whether the penalties generalize or merely regularize the training support.
  2. [Table 2] Table 2 (comparison results): while mean errors are lower than baselines, the table lacks error bars, statistical significance tests, and separate columns for long-horizon autoregressive metrics; without these, the claim of 'substantial improvements' for operational use cannot be fully assessed.
minor comments (2)
  1. [§3.1] §3.1: the shared message-passing update equations would benefit from an explicit diagram or pseudocode to clarify how volume and flow predictions interact at each layer.
  2. [Figure 4] Figure 4: axis labels on the depth and flow time-series plots are too small for readability; increase font size and add units.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and will revise the manuscript to incorporate additional evaluation metrics and statistical analysis where needed.

read point-by-point responses

  1. Referee: [§4.2 and §5.3] §4.2 (Physics-informed losses) and §5.3 (Autoregressive evaluation): the dual global/local constraint losses are shown to reduce training error, but no explicit metrics (e.g., mass-balance violation, cumulative volume drift) are reported for multi-step rollouts on held-out events beyond the training distribution; this directly bears on whether the penalties generalize or merely regularize the training support.

    Authors: We acknowledge that while the dual global/local constraint losses demonstrably reduce training error, explicit reporting of physical consistency metrics during autoregressive multi-step rollouts on held-out events is important for assessing generalization. In the revised manuscript we will add mass-balance violation and cumulative volume drift metrics for long-horizon predictions on events outside the training distribution, both in §5.3 and the supplementary material. These additions will clarify whether the physics penalties generalize or primarily regularize within the training support. revision: yes

  2. Referee: [Table 2] Table 2 (comparison results): while mean errors are lower than baselines, the table lacks error bars, statistical significance tests, and separate columns for long-horizon autoregressive metrics; without these, the claim of 'substantial improvements' for operational use cannot be fully assessed.

    Authors: We agree that the current Table 2 would benefit from error bars, statistical tests, and explicit long-horizon metrics to strengthen the evidence for operational applicability. In the revision we will update Table 2 to report standard deviations across multiple random seeds, include p-values from paired statistical significance tests against baselines, and add a supplementary table (or expanded columns) with long-horizon autoregressive metrics such as 24-hour and 48-hour rollout errors. These changes will provide a more rigorous basis for the claimed improvements. revision: yes

Circularity Check

0 steps flagged

No circularity: model learns from data plus explicit physics penalties

full rationale

The derivation consists of a standard GNN trained with supervised losses on volume/flow targets plus added global/local physics constraint terms and multi-step curriculum training. No equation reduces a claimed prediction to its own inputs by construction, no parameter is fitted then renamed as a prediction, and no load-bearing step relies on self-citation or an imported uniqueness theorem. The architecture and losses are defined independently of the target performance metrics, and the open-sourced code/dataset make the claims externally falsifiable. This is the normal non-circular case for a physics-informed supervised model.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that physics can be adequately enforced through additive loss penalties rather than hard constraints, plus standard GNN message-passing assumptions; no new physical entities are introduced.

free parameters (1)
  • physics loss weights
    Balancing coefficients between data loss and global/local physical constraint terms are hyperparameters that must be chosen or tuned.
axioms (1)
  • domain assumption Physical conservation laws can be effectively approximated by differentiable penalty terms in the training loss
    Invoked when the paper states that explicit loss terms embed physical constraints at global and local scales.

pith-pipeline@v0.9.0 · 5568 in / 1239 out tokens · 36683 ms · 2026-05-16T19:21:21.286970+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages

  1. [1]

    [Ahmadlouet al., 2021 ] Mohammad Ahmadlou, A’kif Al- Fugara, Abdel Rahman Al-Shabeeb, Aman Arora, Rida Al-Adamat, Quoc Bao Pham, Nadhir Al-Ansari, Nguyen Thi Thuy Linh, and Hedieh Sajedi. Flood susceptibil- ity mapping and assessment using a novel deep learning model combining multilayer perceptron and autoencoder neural networks.Journal of Flood Risk Man...

    work page 2021

  2. [2]

    Physics-informed graph neural networks for water distribution systems.Proceed- ings of the AAAI Conference on Artificial Intelligence, 38(20):21905–21913, 3

    [Ashrafet al., 2024 ] Inaam Ashraf, Janine Strotherm, Luca Hermes, and Barbara Hammer. Physics-informed graph neural networks for water distribution systems.Proceed- ings of the AAAI Conference on Artificial Intelligence, 38(20):21905–21913, 3

    work page 2024

  3. [3]

    Battaglia, Jessica B

    [Battagliaet al., 2018 ] Peter W. Battaglia, Jessica B. Ham- rick, Victor Bapst, Alvaro Sanchez-Gonzalez, Vinicius Zambaldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner, Caglar Gulcehre, Francis Song, Andrew Ballard, Justin Gilmer, George Dahl, Ashish Vaswani, Kelsey Allen, Charles Nash, Vic- toria Langston, Chris Dyer,...

    work page 2018

  4. [4]

    Bentivoglio, E

    [Bentivoglioet al., 2023 ] R. Bentivoglio, E. Isufi, S. N. Jonkman, and R. Taormina. Rapid spatio-temporal flood modelling via hydraulics-based graph neural networks. Hydrology and Earth System Sciences, 27(23):4227–4246,

    work page 2023

  5. [5]

    Bentivoglio, E

    [Bentivoglioet al., 2024 ] R. Bentivoglio, E. Isufi, S. N. Jonkman, and R. Taormina. Multi-scale hydraulic graph neural networks for flood modelling.EGUsphere, 2024:1– 28,

    work page 2024

  6. [6]

    Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst

    [Bronsteinet al., 2017 ] Michael M. Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst. Geometric deep learning: Going beyond euclidean data. IEEE Signal Processing Magazine, 34(4):18–42, July

    work page 2017

  7. [7]

    Brunner.HEC-RAS Hydraulic Ref- erence Manual

    [Brunner, 2025] Gary W. Brunner.HEC-RAS Hydraulic Ref- erence Manual. Hydrologic Engineering Center, Insti- tute for Water Resources, U.S. Army Corps of Engineers,

    work page 2025

  8. [8]

    [Costabileet al., 2017 ] Pierfranco Costabile, Carmelina Costanzo, and Francesco Macchione

    Accessed: 2025-12-16. [Costabileet al., 2017 ] Pierfranco Costabile, Carmelina Costanzo, and Francesco Macchione. Performances and limitations of the diffusive approximation of the 2-d shallow water equations for flood simulation in urban and rural areas.Applied Numerical Mathematics, 116:141– 156,

    work page 2025

  9. [9]

    [Deviaet al., 2015 ] Gayathri K

    New Trends in Numerical Analysis: Theory, Methods, Algorithms and Applications (NETNA 2015). [Deviaet al., 2015 ] Gayathri K. Devia, B.P. Ganasri, and G.S. Dwarakish. A review on hydrological models. Aquatic Procedia, 4:1001–1007,

    work page 2015

  10. [10]

    [Fanget al., 2021 ] Zhice Fang, Yi Wang, Ling Peng, and Haoyuan Hong

    International Con- ference on Water Resources, Costal AND Ocean Engineer- ing (ICWRCOE’15). [Fanget al., 2021 ] Zhice Fang, Yi Wang, Ling Peng, and Haoyuan Hong. Predicting flood susceptibility using lstm neural networks.Journal of Hydrology, 594:125734,

    work page 2021

  11. [11]

    A spatial–temporal graph deep learning model for urban flood nowcasting leveraging heterogeneous community features.Scientific Reports, 13(1):6768, 4

    [Farahmandet al., 2023 ] Hamed Farahmand, Yuanchang Xu, and Ali Mostafavi. A spatial–temporal graph deep learning model for urban flood nowcasting leveraging heterogeneous community features.Scientific Reports, 13(1):6768, 4

    work page 2023

  12. [12]

    Fast graph representation learning with pytorch geometric,

    [Fey and Lenssen, 2019] Matthias Fey and Jan Eric Lenssen. Fast graph representation learning with pytorch geometric,

    work page 2019

  13. [13]

    Zahr, and Jian-Xun Wang

    [Gaoet al., 2022 ] Han Gao, Matthew J. Zahr, and Jian-Xun Wang. Physics-informed graph neural galerkin networks: A unified framework for solving pde-governed forward and inverse problems.Computer Methods in Applied Me- chanics and Engineering, 390:114502,

    work page 2022

  14. [14]

    Schoenholz, Patrick F

    [Gilmeret al., 2017 ] Justin Gilmer, Samuel S. Schoenholz, Patrick F. Riley, Oriol Vinyals, and George E. Dahl. Neural message passing for quantum chemistry,

    work page 2017

  15. [15]

    Hamilton, Rex Ying, and Jure Leskovec

    [Hamiltonet al., 2018 ] William L. Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large graphs,

    work page 2018

  16. [16]

    Subgrid informed neural net- works for high-resolution flood mapping.Journal of Hy- drology, 660, 10

    [Herathet al., 2025 ] Herath Mudiyanselage Viraj Vidura Herath, Lucy Marshall, Abhishek Saha, Sanka Rasnayaka, and Sachith Seneviratne. Subgrid informed neural net- works for high-resolution flood mapping.Journal of Hy- drology, 660, 10

    work page 2025

  17. [17]

    A deep learning model for predicting river flood depth and extent.Environmental Modelling & Software, 145:105186,

    [Hosseiny, 2021] Hossein Hosseiny. A deep learning model for predicting river flood depth and extent.Environmental Modelling & Software, 145:105186,

    work page 2021

  18. [18]

    Strategies for pre-training graph neural net- works,

    [Huet al., 2020 ] Weihua Hu, Bowen Liu, Joseph Gomes, Marinka Zitnik, Percy Liang, Vijay Pande, and Jure Leskovec. Strategies for pre-training graph neural net- works,

    work page 2020

  19. [19]

    Censnet: Convolution with edge-node switch- ing in graph neural networks

    [Jianget al., 2019 ] Xiaodong Jiang, Pengsheng Ji, and Sheng Li. Censnet: Convolution with edge-node switch- ing in graph neural networks. InProceedings of the Twenty-Eighth International Joint Conference on Artifi- cial Intelligence, IJCAI-19, pages 2656–2662. Interna- tional Joint Conferences on Artificial Intelligence Orga- nization, 7

    work page 2019

  20. [20]

    [Jianget al., 2024 ] Jiange Jiang, Chen Chen, Yang Zhou, Stefano Berretti, Lei Liu, Qingqi Pei, Jianming Zhou, and Shaohua Wan. Heterogeneous dynamic graph convolu- tional networks for enhanced spatiotemporal flood fore- casting by remote sensing.IEEE Journal of Selected Top- ics in Applied Earth Observations and Remote Sensing, 17:3108–3122,

    work page 2024

  21. [21]

    Floodgnn-gru: a spatio-temporal graph neural network for flood prediction

    [Kazadiet al., 2024 ] Arnold Kazadi, James Doss-Gollin, Antonia Sebastian, and Arlei Silva. Floodgnn-gru: a spatio-temporal graph neural network for flood prediction. Environmental Data Science, 3:e21,

    work page 2024

  22. [22]

    Kipf and Max Welling

    [Kipf and Welling, 2017] Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks,

    work page 2017

  23. [23]

    Mehta, and Karan Singh

    [Kumaret al., 2023 ] Vijendra Kumar, Kul Vaibhav Sharma, Tommaso Caloiero, Darshan J. Mehta, and Karan Singh. Comprehensive overview of flood modeling approaches: A review of recent advances.Hydrology, 10(7),

    work page 2023

  24. [24]

    Exploring a spatiotemporal hetero graph-based long short-term mem- ory model for multi-step-ahead flood forecasting.Journal of Hydrology, 633:130937,

    [Luoet al., 2024 ] Yuxuan Luo, Yanlai Zhou, Hua Chen, Li- hua Xiong, Shenglian Guo, and Fi-John Chang. Exploring a spatiotemporal hetero graph-based long short-term mem- ory model for multi-step-ahead flood forecasting.Journal of Hydrology, 633:130937,

    work page 2024

  25. [25]

    Nash and J.V

    [Nash and Sutcliffe, 1970] J.E. Nash and J.V . Sutcliffe. River flow forecasting through conceptual models part i — a discussion of principles.Journal of Hydrology, 10(3):282–290,

    work page 1970

  26. [26]

    Flood- ldm: Generalizable latent diffusion models for rapid and accurate zero-shot high-resolution flood mapping,

    [Neoet al., 2025 ] Sun Han Neo, Sachith Seneviratne, Herath Mudiyanselage Viraj Vidura Herath, Abhishek Saha, Sanka Rasnayaka, and Lucy Amanda Marshall. Flood- ldm: Generalizable latent diffusion models for rapid and accurate zero-shot high-resolution flood mapping,

    work page 2025

  27. [27]

    A new graph-based deep learning model to predict flooding with validation on a case study on the humber river.Water, 15(10),

    [Oliveira Santoset al., 2023 ] Victor Oliveira Santos, Paulo Alexandre Costa Rocha, John Scott, Jesse Van Griensven Th ´e, and Bahram Gharabaghi. A new graph-based deep learning model to predict flooding with validation on a case study on the humber river.Water, 15(10),

    work page 2023

  28. [28]

    Rahimzad, A

    [Rahimzadet al., 2021 ] M. Rahimzad, A. Moghaddam Nia, H. Zolfonoon, et al. Performance comparison of an lstm- based deep learning model versus conventional machine learning algorithms for streamflow forecasting.Water Re- sources Management, 35(12):4167–4187,

    work page 2021

  29. [29]

    Flood exposure and poverty in 188 countries.Nature Communications, 13(1):3527, 6

    [Rentschleret al., 2022 ] Jun Rentschler, Melda Salhab, and Bramka Arga Jafino. Flood exposure and poverty in 188 countries.Nature Communications, 13(1):3527, 6

    work page 2022

  30. [30]

    From data to action in flood forecasting leveraging graph neural networks and digi- tal twin visualization.Scientific Reports, 14(1):18571, 8

    [Roudbariet al., 2024 ] Naghmeh Shafiee Roudbari, Shub- ham Rajeev Punekar, Zachary Patterson, Ursula Eicker, and Charalambos Poullis. From data to action in flood forecasting leveraging graph neural networks and digi- tal twin visualization.Scientific Reports, 14(1):18571, 8

    work page 2024

  31. [31]

    A review of physics- informed machine learning in fluid mechanics.Energies, 16(5),

    [Sharmaet al., 2023 ] Pushan Sharma, Wai Tong Chung, Bassem Akoush, and Matthias Ihme. A review of physics- informed machine learning in fluid mechanics.Energies, 16(5),

    work page 2023

  32. [32]

    Sun, Zhi Li, Wonhyun Lee, Qixing Huang, Bridget R

    [Sunet al., 2023 ] Alexander Y . Sun, Zhi Li, Wonhyun Lee, Qixing Huang, Bridget R. Scanlon, and Clint Dawson. Rapid flood inundation forecast using fourier neural op- erator,

    work page 2023

  33. [33]

    Climate change impact on flood and extreme precipitation increases with water avail- ability.Scientific Reports, 10(1):13768,

    [Tabari, 2020] Hossein Tabari. Climate change impact on flood and extreme precipitation increases with water avail- ability.Scientific Reports, 10(1):13768,

    work page 2020

  34. [34]

    Interpretable physics- informed graph neural networks for flood forecasting

    [Taghizadehet al., 2025 ] Mohammad Taghizadeh, Zahra Zandsalimi, Mohammad Amin Nabian, Majid Shafiee- Jood, and Negin Alemazkoor. Interpretable physics- informed graph neural networks for flood forecasting. Computer-Aided Civil and Infrastructure Engineering, pages 1–21,

    work page 2025

  35. [35]

    Flipped classroom: Effective teaching for time series forecasting,

    [Teutsch and M¨ader, 2022] Philipp Teutsch and Patrick M¨ader. Flipped classroom: Effective teaching for time series forecasting,

    work page 2022

  36. [36]

    Unravelling the performance of physics-informed graph neural networks for dynamical systems,

    [Thangamuthuet al., 2023 ] Abishek Thangamuthu, Gunjan Kumar, Suresh Bishnoi, Ravinder Bhattoo, N M Anoop Krishnan, and Sayan Ranu. Unravelling the performance of physics-informed graph neural networks for dynamical systems,

    work page 2023

  37. [37]

    The Hu- man Cost of Disasters: An Overview of the Last 20 Years (2000-2019), 10

    [UNDRR and CRED, 2020] UNDRR and CRED. The Hu- man Cost of Disasters: An Overview of the Last 20 Years (2000-2019), 10

    work page 2020

  38. [38]

    Graph attention networks,

    [Veliˇckovi´cet al., 2018 ] Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Li `o, and Yoshua Bengio. Graph attention networks,

    work page 2018

  39. [39]

    Wang, Rory Nathan, and Yuefei Huang

    [Xieet al., 2021 ] Shuai Xie, Wenyan Wu, Sebastian Mooser, Q.J. Wang, Rory Nathan, and Yuefei Huang. Artificial neural network based hybrid modeling approach for flood inundation modeling.Journal of Hydrology, 592:125605,

    work page 2021

  40. [40]

    How powerful are graph neural net- works?,

    [Xuet al., 2019 ] Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural net- works?,

    work page 2019

  41. [41]

    Nenn: Incorporate node and edge features in graph neural net- works

    [Yang and Li, 2020] Yulei Yang and Dongsheng Li. Nenn: Incorporate node and edge features in graph neural net- works. In Sinno Jialin Pan and Masashi Sugiyama, edi- tors,Proceedings of The 12th Asian Conference on Ma- chine Learning, volume 129 ofProceedings of Machine Learning Research, pages 593–608. PMLR, 11

    work page 2020

  42. [42]

    Graph neural network-based surrogate modelling for real-time hydraulic prediction of urban drainage networks

    [Zhanget al., 2024 ] Zhiyu Zhang, Wenchong Tian, Chenkaixiang Lu, Zhenliang Liao, and Zhiguo Yuan. Graph neural network-based surrogate modelling for real-time hydraulic prediction of urban drainage networks. Water Research, 263:122142,

    work page 2024

  43. [43]

    Joint spatial and temporal modeling for hydrological prediction

    [Zhaoet al., 2020 ] Qun Zhao, Yuelong Zhu, Kai Shu, Ding- sheng Wan, Yufeng Yu, Xudong Zhou, and Huan Liu. Joint spatial and temporal modeling for hydrological prediction. IEEE Access, 8:78492–78503, 2020

    work page 2020