Top 22%
in this catalogue (1041 scored practices)
Scores cluster high, so position within the catalogue is often more telling than the number alone.
Evidence of impact / measured public value2/3
UNOSAT's AI-assisted assessment of Carriacou was produced and released within three days of landfall, but the reported 85-98% match to expert judgment and the $218m/16.5%-of-GDP damage estimate come from the nine-disaster evaluation and a separate World Bank GRADE report rather than a Grenada-specific accuracy study.
Transparency, fairness & accountability3/3
The SKAI model is open-sourced on GitHub and its human-review workflow is publicly documented, and the resulting damage maps were published openly via ReliefWeb and the Humanitarian Data Exchange.
Transferability / demonstrated replication3/3
The same AI pipeline was applied and evaluated across nine separate disasters in multiple countries (Syria, Libya, Tunisia, Rwanda, the Philippines, Tonga, Bangladesh and the Caribbean), demonstrating real cross-context replication.
Scalability beyond pilot2/3
The workflow already runs as a standing service across multiple disasters and agencies (UNOSAT, WFP, Google Research), but each release still requires expert human correction before publication, capping how far it can scale without added review capacity.
Governance, capability & sustainability1/3
Roles are split between Google Research (model), the WFP Innovation Accelerator (fine-tuning) and UNOSAT analysts (validation and release), but no Grenada-specific AI governance or oversight body is documented for how the outputs are used domestically.
Evidoria. AI-Assisted Satellite Damage Assessment for Grenada After Hurricane Beryl. Persistent ID: b3fe365c-333d-4ebc-8252-c411a5d64bc3.