Catastrophe modelling is the use of computer simulations to estimate how much a natural disaster, such as a cyclone, a flood or an earthquake, would cost an insurer. It is often shortened to cat modelling. The model does not predict when the next disaster will come. It estimates how large the losses could be, and how often losses of that size happen.
A catastrophe model has three parts. The first part is a long list of imagined events. For cyclones, the model creates tens of thousands of possible storms, each with a path, a wind speed and a chance of happening in any one year. The second part works out the damage. For each storm, it calculates the wind or the water at every building the insurer covers, and how badly that kind of building would be damaged. This uses property risk data, such as each building's age and roof. The third part adds up the money. It applies each policy's terms and totals the insurer's loss for each storm.
Consider an insurer that covers 2 lakh homes and shops along the Odisha coast. The model runs 10,000 simulated years of cyclones against these buildings. In most years the loss is small. But in the worst one year out of every 200, the loss is above ₹1,800 crore. The insurer does not keep that much spare money. So it buys reinsurance, which is insurance that an insurer itself buys to share its largest risks. In many countries, regulators also look at this number, to check that the insurer can survive a very bad year.
Running millions of storm-and-building combinations needs a lot of computing. The models often run on a distributed computing framework, which splits the work across many computers. After a real disaster, insurers compare the model with what actually happened. Synthetic aperture radar (SAR) satellites can map flooded areas even through cloud. Reading such images with AI models is part of geospatial AI.
Catastrophe modelling differs from ordinary underwriting. Underwriting prices one building at a time. A catastrophe model looks at everything the insurer covers at once, because one storm can damage thousands of buildings on the same day.