AI-Assisted Satellite Damage Assessment for Grenada After Hurricane Beryl

Hurricane Beryl made landfall on Carriacou, Grenada as a high-end Category 4 storm on 1 July 2024, causing widespread destruction to buildings, the island's airport terminal and the harbour at Tyrrel Bay.

UNOSAT applied Google Research's SKAI AI building-damage-detection model to satellite imagery to speed up post-disaster damage assessment and inform response coordination.

SKAI combines a pre-trained 'Open Buildings' segmentation model with a damage-classification model fine-tuned with the World Food Programme's Innovation Accelerator; imagery came via the International Charter Space and Major Disasters, Copernicus and Microsoft's AI for Good Lab, with human analysts reviewing and correcting AI output before publishing preliminary damage maps via ReliefWeb and the Humanitarian Data Exchange on 4 July 2024, three days after landfall.

Across nine disasters evaluated by UN Global Pulse and UNOSAT, including Beryl, the AI-assisted workflow let analysts cover seven times more area and cut assessment time sixfold to under a day, with outputs matching expert assessments on 85-98% of buildings in past events. The World Bank's GRADE methodology separately put Beryl's total economic damage to Grenada at US$218.0 million, about 16.5% of 2023 GDP.

The published performance statistics are aggregated across all nine disasters rather than broken out for Grenada alone, and the system still depends on expert human review before release, so it functions as a decision-support accelerant rather than a fully autonomous system.