SKAI — WFP and Google Research's Open-Source AI for Rapid Earthquake Damage Mapping in Türkiye

SKAI is an open-source machine-learning tool built by the World Food Programme's Innovation Accelerator with Google Research to speed up post-disaster building-damage assessment from satellite and aerial imagery. Released as open source in June 2022, it was designed to be context-agnostic, usable across earthquake, flood, hurricane and cyclone responses without retraining from scratch for each disaster. It was put to its highest-profile test after the February 2023 Türkiye-Syria earthquake.

The tool aims to give humanitarian and disaster-response agencies a fast, scalable way to identify which buildings have been destroyed or heavily damaged, so that response resources can be routed to where they are needed most, without waiting for slow manual assessment.

A five-person joint WFP/Google Research team assessed roughly 600,000 buildings across southeast Türkiye in under a month, including a first pass of over 265,000 buildings across three major cities within a week, covering an area affected by roughly 5 million people. The output was shared to support the response coordinated by Türkiye's national disaster agency AFAD, alongside separate satellite-mapping work by Microsoft's AI for Good Lab.

The Türkiye deployment reached over 81% accuracy and flagged more than 28,000 heavily damaged or destroyed structures. WFP has since redeployed SKAI for Hurricane Ian in the United States (410,000 buildings, ~70% precision/recall), the 2022 Pakistan floods (850,000 households, 80–85% precision) and flooding in Durban, South Africa (250,000 buildings, 88% precision), with the tool assessing more than 2.1 million buildings cumulatively since release.

WFP's own account of the Türkiye deployment is candid about its limits, noting that the real-world impact of the AI-generated damage maps was difficult to measure and not intensively verified against on-the-ground outcomes. SKAI is a genuine, repeatedly reused open-source deployment, but its downstream humanitarian impact has not been independently evaluated.