Fund prevention first, and rank it by the loss it avoids.

Most disaster money arrives after the disaster

UNDRR's 2025 report finds that most disaster financing still goes to response and recovery rather than prevention. It notes that only about a quarter of climate-related catastrophe losses in the EU are insured, and that around half of the National Adaptation Plans in one survey had not costed what they would take to carry out. Warning coverage is uneven too. According to WMO, 119 countries, 60% of the total, now report having a multi-hazard early warning system, against 43% of small island developing states.

We think that order should be reversed.

What we would model, and with which data

A disaster risk model has three parts: the hazard, the people and assets exposed to it, and how badly each asset is damaged at each hazard level. Open data is enough to start.

• Hazard data from open services, such as the flood, wildfire and drought services of the Copernicus Emergency Management Service, and from national weather and geological agencies.
• Exposure data on roads, buildings and population, from OpenStreetMap, census tables and free Copernicus satellite imagery.
• Damage curves, which give the expected damage to each type of asset at each hazard level.
• Asset registers from utilities, transport agencies and companies, added when they change the answer.
• Past loss and insurance claims records, to calibrate the model.

How the model is built and checked

A probabilistic risk model simulates thousands of possible events and adds up the losses each would cause, which gives an expected annual loss and the chance of very large losses. Adaptation options, such as raising a substation, restoring a wetland or tightening a building code, run through the same model and are compared on avoided loss against cost. Climate scenarios shift the hazard inputs, so the ranking is tested against more than one future. Back-testing replays a past event, such as a recorded flood, and checks that the model reproduces the reported damage. Every result carries its uncertainty.

What decision-makers get

• Finance and planning ministries get a costed ranking of adaptation options that a budget or a National Adaptation Plan can use.
• Disaster management agencies get impact estimates before an event, showing which roads, hospitals and neighborhoods a forecast flood is likely to reach.
• Utilities and transport operators get the assets whose failure would cause the largest knock-on losses.
• Insurers and companies get loss estimates for their sites and supply routes. UNDRR notes that risk reduction can earn financial returns for the private sector.

A first project proves the ranking on one hazard in one place, such as river flooding in a single basin. We would build the model from open hazard and exposure data, back-test it against a past event and hand over a scorecard with the ranked options. The public body or company then decides whether to continue.

How it runs safely

We disagree with ranking protection by money alone, because it favors areas with expensive assets. The model reports people at risk next to losses in money.

• The model ranks options. Engineers and officials review them, and a named official decides.
• Fairness checks compare protection across neighborhoods and income groups.
• Data on critical infrastructure stays in the owner's systems, and published outputs are aggregated.
• In the EU, the AI Act classes AI that dispatches emergency services or sets their priority as high-risk. This page is not legal advice.

Open-source building blocks

Open models let others rerun a government's ranking.

• CLIMADA, a free and open-source framework for climate risk assessment and adaptation option appraisal.
• OpenQuake Engine, from the Global Earthquake Model Foundation, for seismic hazard and risk.
• Oasis LMF, a toolkit for developing, testing and running catastrophe loss models.
• LISFLOOD, a spatially distributed water resources model developed by the European Commission's Joint Research Centre since 1997.
• Delft-FIAT, a Python package from Deltares for estimating damage from hazards.