AI system
- System
-
xView2 building damage assessment models (semantic segmentation on satellite imagery, trained on the xBD dataset)
- Model
-
Not assessed
- Release
-
Not assessed
- Application developer
-
US Defense Innovation Unit (DIU) with Carnegie Mellon University Software Engineering Institute; research partners
Microsoft AI for Good and UC Berkeley - Role
-
Not assessed
- Configuration
-
Not assessed
- Inputs and outputs
-
Not assessed
- Autonomy
-
Not assessed
- Workflow
-
Usable high-resolution post-event satellite imagery (first available 9 February 2023, three days after the quake). A pre-trained open-source damage model (xBD/xView2) run by the developers on request of responding organizations. Direct contact between DIU/Microsoft researchers and responders (UN INSARAG teams, AFAD).
- Human oversight
-
Not assessed
- Relation to AI
-
The AI produced the damage classifications that responders are said to have used. Which decisions changed is not documented. Sources: the developer (DIU's Gupta) and the World Bank specialist, quoted in one news report, and Microsoft's own report of delivering estimates to AFAD. World Bank's GRADE report and DLR's account do not corroborate operational use of AI products.
- AI contribution
-
The AI produced the damage classifications that responders are said to have used. Which decisions changed is not documented. Sources: the developer (DIU's Gupta) and the World Bank specialist, quoted in one news report, and Microsoft's own report of delivering estimates to AFAD. World Bank's GRADE report and DLR's account do not corroborate operational use of AI products.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
estimating 3,849 buildings damaged or destroyed
- Category
-
Improved Knowledge Discovery
- Status
-
Not assessed
- Beneficiary groups
-
People. Organizations. Society.
- Beneficiaries
-
Not assessed
- Reach
-
Not assessed
- Distribution and access
-
Not assessed
- Duration
-
Not assessed
- Expected?
-
Yes
- Costs and harms
-
Not assessed
Measurement
- Studies
-
Not assessed
- Study design
-
Not assessed
- Measurement
-
Not assessed
- Setting and sample
-
Not assessed
- Measures
-
estimating 3,849 buildings damaged or destroyed
- Comparison
-
Assumed alternative: conventional damage assessment
- Results
-
Assumed comparison. Reference: weeks. Result: hours or minutes.
- Additional value from AI
-
Not assessed
- Limitations
-
Not assessed
- Denominators
-
Not assessed
- Materials for replication
-
Not assessed
Money
- Kind of value
-
Not assessed
- Amount
-
Not assessed
- Currency and prices
-
Not assessed
- Period and unit
-
Not assessed
- Realization
-
Not assessed
- Recipients and payers
-
Not assessed
- Gross or net
-
Not assessed
- Included costs
-
Not assessed
- Comparison
-
Not assessed
- Method
-
Not assessed
- Return and payback
-
Not assessed
People and places
- Participants
-
Developer: US Defense Innovation Unit (DIU) with Carnegie Mellon University Software Engineering Institute; research partners
Microsoft AI for Good and UC Berkeley. Operator: UN INSARAG (search and rescue ground teams in Adıyaman); AFAD, World Bank, IFRC and WFP (per the developer's account); Microsoft AI for Good delivered estimates to AFAD (per Microsoft). - Jurisdictions
-
Benefit: Adıyaman and four cities in southeast Türkiye including Kahramanmaraş (Marash). Developer:
United States. Operator: 
Switzerland. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Cloud-free daytime imagery with a satellite overhead. The article notes far fewer images in remote or less developed areas such as Syria. Responder trust in an off-site AI product, which the developer describes as hard to obtain. Documentation by the responding agency of what the maps changed, which no cited source provides.
- Replication attempts
-
2 follow-up entries. Outcome unresolved: 2.
- Later findings
-
Direct Relief's 10 December 2025 article describes a retrospective research study of the Microsoft and Google 2023 models that found they disagreed in significant ways; no later operational use of xView2 was found in the cited sources.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0346-source-0001–0008
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Independent, Developer
- Original evidence
-
Not assessed
- Access and versions
-
Not assessed
- Supported findings
-
Not assessed
- Evidence links
-
Not assessed
- Evidence strength
-
Not assessed
- Missing evidence
-
Not assessed
Grades
- Assessed benefits
-
People. Organizations. Society.
- Depth
-
People: Not assessed. Organizations: 2. Society: 1.
- Reasons
-
People: No individual rescue or beneficiary is attributed to the maps in any cited source. Organizations: Responders are reported to have received building-level damage maps in hours instead of weeks and to have found damaged areas they were unaware of. The reports come from the developer and a World Bank specialist, not the responding agencies. Society: No downstream effect on rescues, shelter siting or reconstruction is documented. The maps covered part of the affected zone after imagery became usable on 9 February.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
Independently reported
Entry details
- Kind
-
Not assessed
- Related entries
-
Not assessed
- Field
-
Disaster response / post-earthquake damage assessment. Humanitarian and civil-protection organizations.
- Purpose
-
Rapid building-damage assessment from pre- and post-event satellite imagery for humanitarian assistance and disaster response. This use was that purpose.
- Dates
-
Date basis unassessed: 2023 (precision unassessed). Updated: 22 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: 0552f28b520c…
References
- Other identifiers
-
Not assessed
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
9 revisions. Latest: 22 October 2026.
Sources
-
Report Developer
Robinson, C., Gupta, R., Fobi Nsutezo, S., Pound, E., Ortiz, A., Rosa, M., White, K., Dodhia, R., Zolli, A., Birge, C., Lavista Ferres, J. M. (2023). Turkey Earthquake Report. Microsoft AI for Good Research Lab, February 2023.
partnered with Turkey's Ministry of Interior Disaster and Emergency Management Presidency (AFAD) to deliver building level damage estimates over four cities in southeast Turkey
3,849 buildings were damaged/destroyed across the four cities
-
Preprint Developer
Gupta, R., Hosfelt, R., Sajeev, S., Patel, N., Goodman, B., Doshi, J., Heim, E., Choset, H., Gaston, M. (2019). xBD: A Dataset for Assessing Building Damage from Satellite Imagery. arXiv:1911.09296.
Massive potential exists for using aerial imagery combined with computer vision algorithms to assess damage and reduce the potential danger to human life.
-
News report Independent author
Ryan-Mosley, T. (2023). How AI can actually be helpful in disaster response. MIT Technology Review, 20 Feb 2023.
xView2 has been used by at least two different ground teams of search and rescue personnel from the UN’s International Search and Rescue Advisory Group in Adiyaman
noting Turkey’s Disaster and Emergency Management Presidency, the World Bank, the International Federation of the Red Cross, and the United Nations World Food Programme have all used the platform
Atishay Abbhi, a disaster risk management specialist at the World Bank, tells me that this same degree of assessment would typically take weeks and now takes hours or minutes.
It uses machine-learning algorithms in conjunction with satellite imagery from other providers to identify building and infrastructure damage in the disaster area and categorize its severity
-
News report Independent author
Direct Relief (2025). After Disaster, AI Models Support Rapid Understanding of Health Facility Damage. 10 Dec 2025 (describes a Direct Relief/Stanford research study in PLOS Digital Health; does not mention xView2).
Two of the most prominent models in use during the 2023 earthquake response came from Microsoft AI for Good and from Google.