United StatesPublic sector2023 – ongoing

American Red Cross

Satellite-image damage assessment after the Rolling Fork tornado

American Red Cross received AI damage estimates for 1,347 buildings after the Rolling Fork tornado

WorkingCompany's own claim

Problem

Disaster teams needed rapid building damage estimates to inform ground operations. Models trained on other disasters could perform poorly on unfamiliar imagery.

Approach

Microsoft researchers and the American Red Cross tested a locally adapted image model after the March 2023 tornado. Researchers labeled post-disaster imagery and trained a segmentation model. They supplied building-level estimates to the Red Cross and later compared predictions with field assessments.

Results

1,347buildings with damage estimates supplied to the Red Cross by the project teamCompany's own claim[1]

0.86precision for detecting major damage or destruction against field assessments, reported by the project researchersCompany's own claim[1]

Takeaway

Local image labeling supported rapid adaptation to a specific disaster. Field assessments remained necessary to evaluate the predictions.

Sources

Rapid building damage assessment workflow: An implementation for the 2023 Rolling Fork, Mississippi tornado event
arXiv24 Aug 2023Research paper or report

Source summaries

[1]arXiv24 Aug 2023

Microsoft and Red Cross researchers described an operational disaster-response trial. They reported estimates for 1,347 buildings and precision of 0.86 against field assessments.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026