DISHA — UNOSAT and Google Research's AI Damage-Assessment Workflow, Deployed for Colombia's 2026 Floods

Disaster responders need fast, reliable estimates of building damage after earthquakes, floods and storms, but manual review of satellite and radar imagery is slow and labour-intensive.

UNOSAT (the UN Satellite Centre), UN Global Pulse and Google Research built DISHA, an AI-assisted workflow to produce fast, directional damage assessments after disasters by cross-referencing AI-derived building maps with radar and optical satellite imagery.

A zero-shot AI model (SKAI) gives analysts a first pass at which areas show the most building damage, letting UNOSAT experts focus detailed review where it matters most. The workflow has been used in 11 real activations for earthquakes, floods and cyclones, including Hurricane Melissa in Jamaica (October 2025) and Colombia's February 2026 floods, where UNOSAT cross-referenced AI-derived building maps with radar imagery to inform both the Colombian national government's and UN agencies' response planning.

Across nine tested emergencies, the workflow let analysts cover roughly seven times more area and cut the time needed to produce directional damage findings by a factor of six, to under a day. After Hurricane Melissa it assigned preliminary damage scores to more than 385,000 buildings; the February 2026 Colombian floods affected more than 580,000 people, destroyed over 4,000 houses and damaged roughly 23,000 more (UNITAR figures as of 6 March 2026).

The tool's own operators describe it as producing preliminary, directional guidance for expert analysts rather than a finished assessment, and the performance figures come from Google's and the UN's own reporting rather than an independent third-party audit.