AI system
- System
-
510 probabilistic typhoon impact model (XGBoost) predicting totally damaged houses, used as the CERF anticipatory action activation trigger
- Model
-
XGBoost within the 510 typhoon-impact model.
- Release
-
Exact release not established.
- Application developer
-
510, an initiative of the Netherlands Red Cross, with IFRC - Role
-
The model estimated how many houses a typhoon might destroy.
- Configuration
-
The 2021 revision replaced deterministic forecasts with probabilities derived from ECMWF ensemble tracks.
- Inputs and outputs
-
The model converted typhoon forecasts and data on exposed housing into estimates of houses likely to be destroyed.
- Autonomy
-
Not assessed
- Workflow
-
A pre-arranged CERF allocation with agency action plans ready before the typhoon season. ECMWF ensemble track, wind and rainfall forecasts fed to the 510 impact model every 6–12 hours. Historical house-damage data linking hazard, vulnerability and exposure to train the model. A threshold matrix of predicted totally damaged houses per region with a 1-in-3.5-year return period (2025).
- Human oversight
-
Not assessed
- Relation to AI
-
OCHA's 2021, 2023 and 2025 frameworks name the IFRC/510 impact model as the activation-trigger model. The 2021 framework describes it as partially a machine learning model. This is documented by the deployer, not only the maker. The model's forecasts were monitored during the December 2021 near activation (Typhoon Rai) without meeting criteria. The 2024 readiness phase for Pepito ended with CERF AA resources reallocated to response. OCHA's flash update attributes the 6 November 2025 activation for Uwan to the observational threshold (forecast peak 195 km/h against the 185 km/h trigger). It does not state that the 510 model's estimates reached the activation matrix. The AI's contribution to any payout is therefore not established. The maker's NHESS paper is the only source describing the algorithm (XGBoost).
- AI contribution
-
OCHA’s frameworks included the 510 model’s forecasts of damage to housing among the conditions for early action before a typhoon.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
OCHA frameworks (2021, 2023, 2025) use 510's XGBoost house-damage model as activation trigger; near-activation Dec 2021 (Rai); the Nov 2025 activation for Uwan ran on the 185 km/h wind trigger.
- 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
-
Not assessed
- Comparison
-
Not assessed
- Results
-
Not assessed
- Additional value from AI
-
Not assessed
- Limitations
-
The available accounts do not establish that the model triggered a payment. OCHA attributed the November 2025 activation for Typhoon Uwan to a wind-speed threshold, without saying that the model’s damage estimates met the activation criteria.
- 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:
510, an initiative of the Netherlands Red Cross, with IFRC. Operator: UN OCHA Philippines with the CERF AA Core Group (FAO, IOM, UNFPA, UNICEF, WFP); impact estimates generated by 510 (2021 framework) and by OCHA with its Centre for Humanitarian Data (2025 framework). - Jurisdictions
-
Benefit:
Philippines (Regions 5, 8 and 13). Developer: 
Netherlands. Operator: Philippines. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Forecast lead time of at least 72 hours; rapid intensification (Rai 2021) shortened lead time below the trigger window. Model impact estimates reaching the regional threshold matrix, which no read activation record documents. Agency delivery capacity (financial service providers, data-sharing agreements) in the 72 hours before landfall. PAGASA and ECMWF data availability.
- Replication attempts
-
3 follow-up entries. Benefit did not recur: 2. Outcome unresolved: 1.
- Later findings
-
6 November 2025. The framework stayed in force and CERF anticipatory action was activated for Super Typhoon Fung-Wong (Uwan) on 6 November 2025 with nearly US$6 million, but on the observational 185 km/h wind threshold rather than on the 510 model's output.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0354-source-0001–0009
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Operator, 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
-
Not assessed
- Reasons
-
People: No activation attributable to the model's output is documented, so no individual-level outcome can be assigned to the AI. Organizations: OCHA adopted the model as its trigger and computes estimates each cycle. The only documented activation (6 November 2025) was triggered by the observational wind-speed threshold. Society: Pre-landfall assistance in November 2025 (WFP reports 42,100 families) is documented, but not attributed to the AI model by any cited source.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
The deployer's account
Entry details
- Kind
-
Not assessed
- Related entries
-
Not assessed
- Field
-
Humanitarian anticipatory action / typhoon impact forecasting. UN-coordinated anticipatory financing (CERF).
- Purpose
-
Forecast the number of totally damaged houses from typhoon track, wind and rainfall forecasts so that CERF-funded early actions can be triggered before landfall. The benefit claimed (AI-triggered anticipatory payouts) is that purpose, but no activation on the model's output is documented.
- Dates
-
Date basis unassessed: 2021 (precision unassessed). Updated: 21 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: fb2e5c801030…
References
- Other identifiers
-
Not assessed
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
4 revisions. Latest: 21 October 2026.
Sources
-
Peer-reviewed study Developer
Kooshki Forooshani, M., van den Homberg, M., Kalimeri, K., Kaltenbrunner, A., Mejova, Y., Milano, L., et al. (2024). Towards a global impact-based forecasting model for tropical cyclones. Natural Hazards and Earth System Sciences 24, 309-329, 1 February 2024.
in 2021 as the trigger model for the UN OCHA AA pilot
-
Report Deployer
The Model includes a trigger mechanism based on the ECMWF forecast and IFRC/ Netherlands Red Cross 510 probabilistic typhoon impact prediction model developed for the Philippines
-
Report Deployer
Partially: Uses a machine learning model to determine relationship between wind speed and housing damage.
-
Report Deployer
While the typhoon at the end did not meet the AA pilot activation criteria, the CERF AA Core Group was closely monitoring the situation
-
Report Deployer
OCHA (2023). Philippines: CERF AA (Anticipatory Action) Framework, 2023 Revision. unocha.org, 3 October 2023 (PDF).
Tropical Cyclone with potential to reach category level 3 or higher
-
Report Deployer
This will be used to compare against 510 model outputs at 72 hours before landfall, in order to account for varying regional vulnerabilities.
-
Report Deployer
OCHA (2025). Philippines: Tropical Cyclones Kalmaegi/Tino and Fung-Wong/Uwan - Flash Update No. 1 (as of 07 November, 9 p.m. local time). ReliefWeb.
CERF Anticipatory Action (AA) was activated on 06 November after the observational threshold for Tropical Cyclone Fung-won (Uwan) was triggered
-
Report Deployer
OCHA (2025). Philippines: Multiple Tropical Cyclones: Humanitarian Needs and Priorities - Response Overview Report (November 2024 - April 2025). unocha.org, 27 June 2025.
CERF AA resources were reallocated to support the emergency response following Super Typhoon Pepito (Man-yi)
-
Press release Deployer
Paulose, H. (2026). How anticipatory action shaped the response to the Philippines' Twin Typhoons. OCHA Financing and Outreach Division, un.org, January 2026.
WFP alone reached 42,100 families through its anticipatory financing mechanism