Skip to content

Activations

When a disaster hits, fAIr is used for rapid response: local models are trained on imagery mapped during the activation, then used to speed up feature and damage mapping while volunteers validate the results. This page tracks fAIr disaster activations.

Nepal floods 2026

In August 2026, flash floods and mudslides swept down the Lhende Khola valley into the Bhote Koshi in Rasuwa district. Within 24 hours of the flood, HOT, NAXA and NDRRMA opened the first mapping project. Volunteers mapped from Vantor before-and-after imagery, and two fAIr models were trained on that mapping: one found the buildings, the other scored the damage level per building. Volunteers then validated the AI results.

How the Nepal flood response used fAIr: imagery released, volunteers map, validators check, fAIr trains, models predict, MapSwipe validates.

  • 338 experienced volunteer mappers added 20,100 buildings to OpenStreetMap (campaign totals on osmsg, 13 September 2026).
  • 4,735 MapSwipe contributors, counted once per project, validated the AI results in nine projects, with 468,340 results (as of 30 October 2026).
  • The damage model, Nepal Flood Damage, can be tried on Try fAIr.
  • Open data on HDX: flood-affected buildings, the river corridor, and the fAIr damage assessment.

The workflow PDF reports its figures as of 13 September 2026.

  Download the Nepal floods workflow (PDF)

Colombia earthquake 2026

On 10 August 2026, a magnitude 7.4 earthquake centred in the Chocó department struck western Colombia, causing widespread destruction in Cali, Pereira, Manizales, Quibdó, and other cities. HOT worked with OSM Colombia, UNGRD, UN Maps, UNDP, and IGAC to define priority areas. Volunteers mapped buildings and roads in the Tasking Manager, fAIr produced building and damage predictions, and MapSwipe volunteers validated the damage predictions.

  • More than 450 contributors mapped buildings and roads in OpenStreetMap (as of 8 September 2026, per the HDX dataset).
  • 717 MapSwipe contributors, counted once per project, validated the building-damage predictions in two projects, for Viterbo (Caldas) and La Tebaida, with 49,058 results (as of 30 October 2026).
  • Open data: OpenStreetMap data on HDX, and fAIr's building and damage predictions for the Pereira area on Hugging Face.

Venezuela earthquake 2026

On 24 June 2026, two earthquakes struck northern Venezuela seconds apart, a magnitude 7.5 shock following a 7.2 foreshock, damaging buildings across Caracas, La Guaira, and the coastal Vargas area. fAIr delivered building density, AI-detected building footprints, and building-damage predictions from post-event Vantor imagery to early responders. Volunteers then mapped buildings in OpenStreetMap and validated the damage predictions through MapSwipe.

  • 2,681 MapSwipe contributors, counted once per project, validated damage predictions grouped into hexagonal grid cells in 23 projects (13 of them also with detections from Microsoft's AI for Good Lab), with 180,385 results (as of 30 October 2026), producing human-validated damage assessments for Caraballeda, La Guaira, and Caracas.
  • See the results of one project, Catia La Mar, in the Results Analyzer; more on Validation.
  • Open data on HDX: OSM and Overture data with fAIr building predictions and the damage assessment areas.