Large scale atmospheric cross-peril stochastic catastrophe models

Impact Forecasting, a catastrophe model development branch of Aon, develops catastrophe models for various countries and perils, including floods, windstorms, earthquakes, wildfires, hurricanes, and typhoons. These models are crucial for the insurance and reinsurance industry to estimate losses in terms of severity and frequency.

To address the increasing demand for evaluating losses across multiple countries and perils, Impact Forecasting has started using large ensembles of global climate models (GCM) and regional climate models (RCM). These models serve as a common forcing input for catastrophe models related to atmospheric perils such as flooding (fluvial, pluvial, and coastal), summer storms, windstorms, and wildfires.

The use of GCM/RCM as common forcing input offers two main advantages:

• Spatial Consistency: The data are spatially consistent at a global or continental scale, which helps in addressing the issue of cross-country correlations.
• Variability: The large number of available ensembles provides sufficient variability to build a representative stochastic catalogue of potential catastrophes.

We will present several examples of this approach (Pan-European flood model, the Canadian flood and wildfire models), where common GCM/RCM inputs are used to provide a consistent view of losses across large regions and various perils. We will also show how we adress the issue of low resolution of GCM/RCM models using machine learning.