FloodTransformer: Efficient real-time high-resolution flood forecasting

Gu, Z; Kang, J; Jin, W; Tong, F; Guo, YJ; Jia, W

FloodTransformer: Efficient real-time high-resolution flood forecasting

Gu, Z

Kang, J Jin, W Tong, F Guo, YJ Jia, W

Permalink

Publisher:: ELSEVIER SCI LTD
Publication Type:: Journal Article
Citation:: Environmental Modelling and Software, 2026, 197
Issue Date:: 2026-02-01

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download Published versionAdobe PDF (3.99 MB)

View on publisher's site

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Gu, Z https://orcid.org/0000-0002-0628-5575
dc.contributor.author	Kang, J
dc.contributor.author	Jin, W
dc.contributor.author	Tong, F
dc.contributor.author	Guo, YJ
dc.contributor.author	Jia, W https://orcid.org/0000-0002-0940-3338
dc.date.accessioned	2026-03-07T21:07:59Z
dc.date.available	2026-03-07T21:07:59Z
dc.date.issued	2026-02-01
dc.identifier.citation	Environmental Modelling and Software, 2026, 197
dc.identifier.issn	1364-8152
dc.identifier.issn	1873-6726
dc.identifier.uri	http://hdl.handle.net/10453/193785
dc.description.abstract	Flood forecasting is crucial for disaster planning and risk management, yet conventional hydrodynamic-based approaches are often slow in response and computationally intensive. We present a hybrid framework leveraging traditional hydrodynamic modelling with a novel AI model to enable accurate, real-time, and high-resolution flood prediction. To address the computational challenges of large-scale, dense flood prediction, we develop an efficient flood prediction model, FloodTransformer, which possesses three key novelties: variable-size cell embedding, tokenised time-sequence encoding, and physics-informed multi-task optimisation. These components effectively capture complex spatiotemporal dependencies, allowing accurate sequential predictions in a single run. Comprehensive evaluations on both simulated and historical flood events demonstrate FloodTransformer's excellent accuracy and efficiency: NSE 0.9445, KGE 0.9759 for water-depth prediction, and IoU 0.8180, F1 0.8997 for inundation classification, outperforming all comparative models. With 3s inference enabling multiple horizons in one pass, FloodTransformer offers a robust and practical solution for operational flood risk management.
dc.language	English
dc.publisher	ELSEVIER SCI LTD
dc.relation	NSW Department of Planning, Housing, and Infrastructure DPHI (previously NSW Department of Planning, Industry and Environment)
dc.relation.ispartof	Environmental Modelling and Software
dc.relation.isbasedon	10.1016/j.envsoft.2025.106832
dc.rights	info:eu-repo/semantics/openAccess
dc.subject.classification	Environmental Engineering
dc.title	FloodTransformer: Efficient real-time high-resolution flood forecasting
dc.type	Journal Article
utslib.citation.volume	197
pubs.organisational-group	University of Technology Sydney
pubs.organisational-group	University of Technology Sydney/Faculty of Engineering and Information Technology
pubs.organisational-group	University of Technology Sydney/Faculty of Engineering and Information Technology/School of Electrical and Data Engineering
pubs.organisational-group	University of Technology Sydney/UTS Groups
pubs.organisational-group	University of Technology Sydney/UTS Groups/Global Big Data Technologies Centre (GBDTC)
utslib.copyright.status	open_access	*
dc.rights.license	This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0). To view a copy of this license, visit https://creativecommons.org/licenses/by-nc/4.0/
dc.date.updated	2026-03-07T21:07:50Z
pubs.publication-status	Published
pubs.volume	197

Abstract:

Flood forecasting is crucial for disaster planning and risk management, yet conventional hydrodynamic-based approaches are often slow in response and computationally intensive. We present a hybrid framework leveraging traditional hydrodynamic modelling with a novel AI model to enable accurate, real-time, and high-resolution flood prediction. To address the computational challenges of large-scale, dense flood prediction, we develop an efficient flood prediction model, FloodTransformer, which possesses three key novelties: variable-size cell embedding, tokenised time-sequence encoding, and physics-informed multi-task optimisation. These components effectively capture complex spatiotemporal dependencies, allowing accurate sequential predictions in a single run. Comprehensive evaluations on both simulated and historical flood events demonstrate FloodTransformer's excellent accuracy and efficiency: NSE 0.9445, KGE 0.9759 for water-depth prediction, and IoU 0.8180, F1 0.8997 for inundation classification, outperforming all comparative models. With 3s inference enabling multiple horizons in one pass, FloodTransformer offers a robust and practical solution for operational flood risk management.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/193785