Development of a climate-driven stochastic event catalogue for Wildfire in Europe

Recent years have seen wildfires causing widespread environmental and economic damage as well as numerous fatalities globally. With record breaking yearly burnt areas, longer fire seasons, and more extreme events, wildfire is emerging as a growing concern for populations, governments and the private sector alike. In Europe, destruction and disruption have been historically more prominent in southern countries where key sectors of the economy like tourism, forestry, and agriculture can remain severely affected for years in the aftermath of catastrophic events.  

Over the last 30 years, catastrophe modelling solutions have been crucial in aiding the understanding of the economic impacts of natural risks like wildfire, making them essential tools for the (re)insurance industry for managing their exposure and quantifying potential losses. Such solutions typically involve the development of large scale and physically-based probabilistic models. 

We present here a climate-driven stochastic event catalogue for wildfire in Europe. The model allows us to expand on the limited historical records by generating millions of synthetic event footprints. For this, we first consider how climate conditions drive spatio-temporal patterns of wildfire activity in terms of yearly burnt area (fire activity module). In a second step, events are sampled via an ignition module that leverages machine learning algorithms and draws correlations between anthropogenic and bio-climate factors, and historical events. Finally, a propagation module generates event footprints given the local topography, fuel data, and meteorological conditions. The stochastic catalogue consists of 50K synthetic years and about 25M unique footprints at 100m resolution. This allows us to estimate hazard metrics like event frequency, event size, and tail risk over the whole continent as well as performing impact analyses. Lastly, we present an evaluation of structures at risk in France by intersecting our catalogue with a representative dataset of buildings.