What happened
UNOSAT, the United Nations Satellite Centre within UNITAR, runs an emergency service that maps disasters and crises from satellite imagery. With research partners on CERN's OpenLab platform, it developed Flood AI, a model for mapping floods. UNITAR's evaluation unit issued a midterm evaluation of the Norway-funded service on 1 May 2024, carried out by an independent consultant. The evaluator relays figures from the project's first annual report, for 2022 to 2023. By that report, Flood AI was deployed in seven activations, and 86 percent of the images it processed needed no correction by analysts. These are the deployer's own figures, and the evaluator did not check them. The evaluator writes that the project exceeded its expected use of Flood AI in the first year, against a target of four activations. The evaluator also writes that the potential for AI models to speed up satellite image analysis is large. The evaluation gives no time saved and no comparison with manual mapping. It also found an apparent reduction in research on other AI models, as funding fell short of the request.
Sources
-
Descriptive report, by an evaluator that the makers or users commissioned.
The project's first annual report states that the Flood AI model was deployed as a response to seven activations and that 86 per cent of the images processed by the model did not require correction or adjustment by the analysts.
Details
AI system
- Workflow
-
UNOSAT ran Flood AI on satellite images in seven emergency activations in the project's first year. Analysts validated the model's outputs through a human-in-the-loop protocol before products were released. UNOSAT developed the model with research partners on CERN's OpenLab platform.
Follow-up
- Conditions for replication
-
Analysts must validate the model's outputs before products are released. Research on AI models needs funding, which fell short of the project's request.
Entry details
- Sector
-
Humanitarian disaster mapping.
- Purpose
-
UNOSAT built Flood AI to map flooded areas in satellite imagery faster during emergencies.
31 more questions not yet checked
model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). UNOSAT reports that analysts left 86% of Flood AI images uncorrected (AIBD-1167). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1167.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)