Microsoft's free HASTE tool maps disaster damage from satellite images within hours

Microsoft has set out how aid groups and governments are using a disaster tool built by its AI for Good Lab. HASTE, short for High-speed Assessment and Satellite Tracking for Emergencies, reads satellite images taken before and after a disaster and marks the areas that look most damaged, often within hours rather than weeks.[1]

The 22 September feature describes use after 23 major disasters in 16 countries since 2023, from floods in Libya to earthquakes in Venezuela this June. The code is free to download, which matters for flood-prone countries such as Malaysia.[1][2]

Microsoft says the system needs only a few hours of someone training it on a specific event. A person marks examples of damaged buildings in images captured before and after the disaster, and the model then applies those patterns across territory that would take analysts far longer to check by hand.[1]

During major emergencies the AI for Good team often works with Planet Labs and other imagery providers. Microsoft researcher Caleb Robinson says the approach helps most when damage is uneven, with some communities devastated and others barely touched. Every disaster leaves a different visual footprint, so the model is trained afresh for each one.[1]

After earthquakes struck Venezuela on 24 June 2026, Catholic Relief Services asked Microsoft for help and had a damage map by the next morning. Nora Lindstrom, who leads the charity's global digital programming team, says producing a similar assessment would have taken her group two weeks of largely manual work.[1]

At the UN humanitarian office, a data scientist built a viewer that placed HASTE results beside eight other mapping products so agencies could compare them. Microsoft's Kevin White puts the trade-off plainly: "If you have fast garbage, that's actually worse." Microsoft also says HASTE maps informed decisions within an hour of Colombia's August earthquake.[1]

The code sits on GitHub under the permissive MIT licence. The repository was created in April 2026 and was last updated on 24 September. Its guide says a full local copy runs with Docker Compose and needs no Azure subscription, although that setup is meant only for development and evaluation.[2]

A production system runs on Microsoft Azure through the project's deployment guide, so an agency would still pay for cloud computing and would need timely satellite imagery of the affected area. Catholic Relief Services has shown the code can be adapted, using it to check damaged routes and to map where tarpaulins were distributed.[2][1]

Malaysia is not starting from zero. The science ministry says the Malaysian Space Agency, MYSA, obtains and analyses satellite images of disaster areas before, during and after an event, and builds mapping systems used online by the National Disaster Management Agency, NADMA.[4]

In January 2025, NADMA held a technical session to connect MYSA satellite data with MyFloodRAS, the flood mapping system of the Department of Irrigation and Drainage, for early flood damage assessment. The last northeast monsoon began in November 2025 and was expected to bring five to seven episodes of continuous heavy rain.[3][5]