Rapid flood scene extraction using deep learning for disaster emergency response.
Saved in:
| Title: | Rapid flood scene extraction using deep learning for disaster emergency response. |
|---|---|
| Authors: | Wei, Hsiao-Ping1 (AUTHOR), Chang, Deng-Lin2 (AUTHOR), Su, Yuan-Fong3 (AUTHOR) yuanfongsu@mail.ntou.edu.tw |
| Source: | Natural Hazards. Jun2026, Vol. 122 Issue 12, p1-20. 20p. |
| Abstract: | Floods are among the most destructive natural hazards worldwide, and accurately identifying inundation spatial distributions is critical for effective emergency response during typhoon events. However, physically-based hydrodynamic simulations are computationally demanding and fail to meet the time-critical requirements of typhoon emergency response, while existing hybrid deep learning approaches predominantly depend on data inputs—such as satellite imagery, multi-source hydrological observations, or compound real-time measurements—that are difficult to obtain within the critical pre-landfall time window, leaving decision-makers facing a critical information gap during the most crucial period before typhoon landfall. This study proposes a rapid flood scene extraction framework integrating SOBEK hydrodynamic modeling, Inception-v3 transfer learning classification, and similarity-based flood scene retrieval, using a radar rainfall composite as the sole system input to deliver township-level inundation scene predictions prior to typhoon landfall. The flood scene database encompasses 813 typhoon events, providing comprehensive and diverse scenario coverage for similarity-based retrieval. The Inception-v3 model achieved overall accuracies of 91.3% for the Toucheng Coastal Basin (TCB) and 90.7% for the Lanyang River Basin (LRB) on the test dataset, with per-class F1-scores consistently above 0.867. Validation against three historical typhoon events including Jangmi (2008), Megi (2010), and Saola (2012) yielded hit rates (HR) of 1.000 across all events, confirming no missed flooded townships, with critical success indices (CSI) of 0.778–1.000. The complete prediction workflow requires approximately 50 s, which is approximately 24 times faster than a single SOBEK scenario simulation. It effectively bridging the information gap in pre-landfall typhoon emergency response decision-making and providing an operationally deployable scientific basis for real-time evacuation order issuance and long-term flood risk management. [ABSTRACT FROM AUTHOR] |
| Database: | Energy & Power Source |
| FullText | Text: Availability: 0 |
|---|---|
| Header | DbId: enr DbLabel: Energy & Power Source An: 194600675 AccessLevel: 6 PubType: Academic Journal PubTypeId: academicJournal PreciseRelevancyScore: 0 |
| IllustrationInfo | |
| Items | – Name: Title Label: Title Group: Ti Data: Rapid flood scene extraction using deep learning for disaster emergency response. – Name: Author Label: Authors Group: Au Data: <searchLink fieldCode="AR" term="%22Wei%2C+Hsiao-Ping%22">Wei, Hsiao-Ping</searchLink><relatesTo>1</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Chang%2C+Deng-Lin%22">Chang, Deng-Lin</searchLink><relatesTo>2</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Su%2C+Yuan-Fong%22">Su, Yuan-Fong</searchLink><relatesTo>3</relatesTo> (AUTHOR)<i> yuanfongsu@mail.ntou.edu.tw</i> – Name: TitleSource Label: Source Group: Src Data: <searchLink fieldCode="JN" term="%22Natural+Hazards%22">Natural Hazards</searchLink>. Jun2026, Vol. 122 Issue 12, p1-20. 20p. – Name: Abstract Label: Abstract Group: Ab Data: Floods are among the most destructive natural hazards worldwide, and accurately identifying inundation spatial distributions is critical for effective emergency response during typhoon events. However, physically-based hydrodynamic simulations are computationally demanding and fail to meet the time-critical requirements of typhoon emergency response, while existing hybrid deep learning approaches predominantly depend on data inputs—such as satellite imagery, multi-source hydrological observations, or compound real-time measurements—that are difficult to obtain within the critical pre-landfall time window, leaving decision-makers facing a critical information gap during the most crucial period before typhoon landfall. This study proposes a rapid flood scene extraction framework integrating SOBEK hydrodynamic modeling, Inception-v3 transfer learning classification, and similarity-based flood scene retrieval, using a radar rainfall composite as the sole system input to deliver township-level inundation scene predictions prior to typhoon landfall. The flood scene database encompasses 813 typhoon events, providing comprehensive and diverse scenario coverage for similarity-based retrieval. The Inception-v3 model achieved overall accuracies of 91.3% for the Toucheng Coastal Basin (TCB) and 90.7% for the Lanyang River Basin (LRB) on the test dataset, with per-class F1-scores consistently above 0.867. Validation against three historical typhoon events including Jangmi (2008), Megi (2010), and Saola (2012) yielded hit rates (HR) of 1.000 across all events, confirming no missed flooded townships, with critical success indices (CSI) of 0.778–1.000. The complete prediction workflow requires approximately 50 s, which is approximately 24 times faster than a single SOBEK scenario simulation. It effectively bridging the information gap in pre-landfall typhoon emergency response decision-making and providing an operationally deployable scientific basis for real-time evacuation order issuance and long-term flood risk management. [ABSTRACT FROM AUTHOR] |
| PLink | https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=enr&AN=194600675 |
| RecordInfo | BibRecord: BibEntity: Identifiers: – Type: doi Value: 10.1007/s11069-026-08269-5 Languages: – Code: eng Text: English PhysicalDescription: Pagination: PageCount: 20 StartPage: 1 Titles: – TitleFull: Rapid flood scene extraction using deep learning for disaster emergency response. Type: main BibRelationships: HasContributorRelationships: – PersonEntity: Name: NameFull: Wei, Hsiao-Ping – PersonEntity: Name: NameFull: Chang, Deng-Lin – PersonEntity: Name: NameFull: Su, Yuan-Fong IsPartOfRelationships: – BibEntity: Dates: – D: 15 M: 06 Text: Jun2026 Type: published Y: 2026 Identifiers: – Type: issn-print Value: 0921030X Numbering: – Type: volume Value: 122 – Type: issue Value: 12 Titles: – TitleFull: Natural Hazards Type: main |
| ResultId | 1 |
