The Philippines Anticipatory Action Framework for Tropical Cyclones is an operational programme coordinated by the United Nations Office for the Coordination of Humanitarian Affairs (OCHA) Philippines, in partnership with the Humanitarian Country Team (HCT), the Central Emergency Response Fund (CERF), and multiple UN agencies including FAO, IOM, WFP, UNICEF, and UNFPA. The framework uses an artificial intelligence-driven impact forecasting model developed by 510, an initiative of the Netherlands Red Cross (NLRC), to predict severe housing damage and population impact ahead of tropical cyclone landfall, thereby triggering the release of pre-arranged financing and enabling early humanitarian actions before a typhoon strikes (OCHA, 2024, AA Framework, pp. 1-2; Centre for Humdata, 2022).
The AI component at the core of this framework is the 510 typhoon impact model. This model uses machine learning applied to ensemble typhoon track forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) to provide a probabilistic estimate of the percentage of completely damaged houses per municipality in the Philippines. The model is built from several indicator categories including historical typhoon track data, weather forecasts (wind speed and rainfall), and housing damage data. It was originally developed by 510 as a deterministic model for use as a funding-release trigger for the IFRC's Disaster Response Emergency Fund (DREF) and has since been revamped into a probabilistic impact forecast using ECMWF data (OCHA, 2024, AA Framework, p. 8; Centre for Humdata, 2022). The model is described in the peer-reviewed literature as an XGBoost-based damage prediction system using ECMWF ensemble tracks, winds, and rainfall combined with exposure and vulnerability inputs at the municipality level (academic literature, specifically Teklesadik et al., 2024). The model is open source and free to use, and it is continually being improved upon (Centre for Humdata, 2022).
The framework operates through a two-stage trigger activation protocol. The first stage is a readiness trigger (pre-activation), activated 3-7 days prior to forecast landfall when a tropical cyclone has the potential to reach category 2 or higher (greater than 154 km/h maximum 1-minute sustained wind speed or 136 km/h maximum 10-minute sustained wind speed) and is projected to directly impact areas within the CERF pilot regions of Region 5 (Bicol), Region 8 (Eastern Visayas), and Region 13 (Caraga). The second stage is the activation trigger, reached on or before 72 hours (3 days) prior to forecast landfall: OCHA Philippines together with the Humanitarian Data Exchange (HDX) calculates the predicted total number of totally damaged buildings using the 510 model, produces an impact map, and updates it every 6-12 hours. The CERF anticipatory action is activated if the predicted number of houses to be totally damaged falls within the range of 50 percent probability that 50,000 houses or more will be totally damaged, or 85 percent probability that at least 8,000 houses will be totally damaged (OCHA, 2024, AA Framework, pp. 1-2, 10-11).
The HCT Triggers Group, comprising OCHA, HDX, and NLRC 510, validates model outputs against the predefined readiness and activation thresholds and provides a recommendation to the Resident Coordinator/Humanitarian Coordinator (RC/HC) on whether to activate the framework. Activation for anticipatory action will only be for the scenario where readiness was previously activated. The Triggers Group monitors the development of tropical cyclones and runs a parallel programme based on the 510 model throughout the typhoon season. If activation triggers are not met for a scenario that was activated for readiness 72 hours prior to landfall, a stand-down notice is issued to partners and to CERF (OCHA, 2024, AA Framework, pp. 11-13).
For the 2024 typhoon season, CERF allocated up to US$7.5 million for this pilot framework. Under Scenario 1, the pilot aims to reach approximately 372,700 people (74,900 of the most vulnerable households) in prioritized provinces of Region 5 and Region 8. Under Scenario 2, the pilot targets approximately 203,300 people (35,300 households) in Region 8 and Region 13 (OCHA, 2024, AA Framework, pp. 1, 6). Multi-sectoral assistance is delivered by UN agencies, NGOs, and the Red Cross/Red Crescent in close collaboration with local authorities. Key interventions include pre-emptive multi-purpose cash transfers of approximately US$60 per household delivered through over-the-counter remittance shops 72 hours before landfall (delivered by FAO, IOM, UNFPA, and WFP); cash for shelter grants of PhP 4,400 (approximately US$80) by IOM; unconditional top-up cash transfers of PhP 1,200 (approximately US$22) per household through Landbank to beneficiaries of the government's Pantawid Pamilya Pilipino Program (4Ps) by UNICEF; and community-based safe storage of livelihood assets by FAO with OXFAM and CARE (OCHA, 2024, AA Framework, pp. 2-3, 25).
The approach to targeting at-risk beneficiaries uses a composite vulnerability index constructed from Philippine Statistics Authority data on municipal poverty incidence (2015), exposure to typhoon impact, number of women, and people living in houses made of light material. Municipalities were prioritized and ranked across the three pilot regions, with selected areas overlaid on a map with existing presence and operational capabilities of UN agencies, Philippine Red Cross, and START Network partners (OCHA, 2024, AA Framework, p. 17). The segment-based approach ensures tailored actions for intersectional groups including women, children, elderly, LGBTQIA+, and persons with disabilities (OCHA, 2024, AA Framework, p. 1).
The peer review conducted by the Centre for Humanitarian Data in 2022 found that the 510 model has a clear use case with well-defined assumptions and limitations. However, it identified aspects requiring more explanation or justification, recommended additional documentation and model release tracking to improve reproducibility, and highlighted the critical issue of false negatives due to rapid intensification of typhoons. In the cases of Super Typhoons Goni (2020) and Rai/Odette (2021), the observed wind speed rapidly increased less than 24 hours before landfall, and the anticipatory action framework was not activated because rapid intensification was not forecasted in time. Super Typhoon Rai damaged 1.9 million homes, killed hundreds, and displaced millions. In response, NLRC/510 conducted a study to incorporate rapid intensification into their model, although results were not conclusive, and further study was recommended (Centre for Humdata, 2022; OCHA, 2024, AA Framework, pp. 7, 13-14).
Republic Act No. 12287, signed on 12 September 2025, established the legal concept of a 'State of Imminent Disaster' in Philippine law, enabling government agencies to access funds and implement anticipatory measures prior to a disaster event. This legislation complements the humanitarian anticipatory action frameworks by providing a formal legal basis for pre-emptive government action (Supreme Court of the Philippines, 2025). The bill originated from the advocacy work of the AA Technical Working Group with national government agencies, which has been working since 2023 to institutionalize anticipatory action in the Philippine preparedness and response structure (OCHA, 2024, AA Framework, p. 7).
An independent evaluation partnership with Kobo and the Harvard Humanitarian Initiative has been established to measure the impact of anticipatory action interventions on households and communities, using a mixed-methods, community-centred approach. An ad hoc M&E group (the Learning Team) was also established as part of the CERF Philippines Core Group to assess how collective and coordinated early action can work at scale for typhoon response (OCHA, 2024, AA Framework, pp. 3-4).