Nepal Flash Flood: Damage Assessment using Sentinel-1 Radar

Rapid screening of likely damage or unusual change using a 28 August 2026 Sentinel-1 acquisition, a one-year radar baseline, and a built-environment analysis domain.

Status
Reviewed
Provider
REACH Initiative
Resource type
Derived Product
Evidence class
Automated

Identity and coverage

Catalogue identifier
res-01de4451-c684-5e12-bbae-12624e0e455d
Persistent identifier
57723eed-8f89-4bec-8ab4-1acd12a34df1
Resource version
HDX metadata state observed 3 September 2026
Publication date
31 August 2026
Last updated
3 September 2026
Metadata reviewed
3 September 2026
Language
English
Geographic coverage
Provider-defined assessment area in the Nepal event corridor, with municipal summaries for the intersecting analysis area.
Bounding box (WGS 84)
[84.52739332, 27.78286814, 85.56596468, 28.3460786]
Temporal coverage
26 August 2025 to 28 August 2026; Mixed
Themes
Damage Proxy, Synthetic Aperture Radar, Buildings, Rapid Prioritization

Method and provenance

Production method
Sentinel-1 IW VV/VH backscatter observations from 26 August 2025 to 26 August 2026 defined a pre-event baseline. The 28 August 2026 acquisition was standardized by z-score; maximum absolute VV/VH anomaly was averaged with a normalized 50 m kernel and restricted to within 30 m of pre-event Overture building footprints. A K50 absolute z-score threshold of at least 1.60 defined the high-confidence layer, which was then visually reviewed and edited.
Provenance
Produced by REACH Initiative/IMPACT Initiatives from Sentinel-1 observations, Overture building footprints, and held-out Copernicus EMSR927 grading data; published through HDX.
Validation
Partially Validated
Uncertainty
Against held-out Copernicus damaged/destroyed observations, the provider reports 53.1% recall and 69.4% F1 for the automated threshold. The zero of 92 'possibly damaged' observations above threshold must not be interpreted as conventional precision or specificity because that class is not verified undamaged.
Limitations
The output is a backscatter-change proxy, not proof of structural damage. Signals can arise from debris, sediment, inundation, moisture, roughness, vegetation, clearance, or other change; damage can also be missed. Steep terrain and viewing geometry are important because dedicated radiometric terrain flattening was not applied.

Access and reuse

Access
Direct Open; open access route
Licence
Creative Commons Attribution (CC BY; version not specified); licence evidence
Redistribution
Permitted Under Stated Package Licence
Catalogue status
Reviewed

Assets

Authoritative metadata sources

Spatial extent

Loading map…

Recorded bounding box