Rapid flood mapping from aerial imagery using fine-tuned SAM and ResNet-backboned U-Net

Shokati, Hadi; Seufferheld, Kay D.; Fiener, Peter; Scholten, Thomas

doi:10.5194/hess-30-743-2026

Articles | Volume 30, issue 3

https://doi.org/10.5194/hess-30-743-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/hess-30-743-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 30, issue 3

Research article

|

09 Feb 2026

Research article |

| 09 Feb 2026

Rapid flood mapping from aerial imagery using fine-tuned SAM and ResNet-backboned U-Net

Hadi Shokati, Kay D. Seufferheld, Peter Fiener, and Thomas Scholten

Model code and software

hadi1994shokati/Flood-segmentation: Flood Segmentation (Version V01) Hadi Shokati https://doi.org/10.5281/zenodo.18485640

Short summary

Floods threaten lives and property and require rapid mapping. We compared two artificial intelligence approaches on aerial imagery: a fine‑tuned Segment Anything Model (SAM) guided by point or bounding box prompts, and a U‑Net network with ResNet‑50 and ResNet‑101 backbones. The point‑based SAM was the most accurate with precise boundaries. Faster and more reliable flood maps help rescue teams, insurers, and planners to act quickly.