Fire risk is characterized by sustained periods of high fire risk due to dry and hot conditions, which can be puncuated by shorter periods of extreme risk due to high winds. To understand the how climate change has affected fire risk, it is important to assess the risk over the course of the entire fire season. And it is important to model how accurately capture the temporal dependence of fire risk at high and extreme levels.
We propose using transformed-linear regularly varying time series models for this purpose. These models have a dependence structure which is tied to a widely-used framework for extremes, but also have a construction method similar to traditional ARMA models. We fit the models to only the extreme values of the time series in order to capture dependence in the tail.
By fitting these models to the Fire Weather Index (FWI) for both a reference climate and a current climate, we can compare the probabilities of observing fire seasons under these two different climates. Preliminary results show that the weather conditions observed in Colorado in 2020 are about twice as likely to occur in the current climate than in the climate of the mid-20th century.
|