A nonparametric framework for quantifying temporal trends in the seasonality of forest fire risk


Lightning-caused fires account for approximately 45% of ignitions and 80% of area burned annually in Canada.  Investigating the seasonality of these fires and how this is changing over time is of interest to fire managers and researchers.  In this project, we develop flexible models for describing the temporal variation in the risk of lightning-caused ignitions and fit these models to historical forest fire records from Alberta and Ontario, Canada. The generalized additive models we utilize provide smooth estimates of fire risk by day of year for each year.  Inverse calculations are used to obtain interval estimates of the start and end of the fire season annually; these are defined by the crossing of a risk threshold.  Permutation-based methods are employed to test for significant trends. Finally, trends from this complex approach are compared to those of simple empirical estimates.  Results suggest changes to the timing of the fire season in Alberta, but not Ontario.