A parametric model of the fire-size distribution

This paper developes statistical models of the size distribution of lightning-caused wildfires in the boreal mixedwood forests of Alberta, Canada, for the intervals 1980-1998 and 1961-1998. Above any minimum threshold size >=3 ha, the logarithm of fire size is approximately exponentially distributed. However, computer simulations using the best-fit distribution would over predict the frequency of large fires, and thus the mean rate of disturbance. A truncated exponential distribution, which places an upper bound on fire size, is more suitable and, according to probability plots, provides an excellent fit to the data. I estimate the maximum fire size in the study area to be approximately 650 000 ha. This estimate is insensitive to the choice of lower bound for fire sizes (between 3 and 1000 ha) and to the choice of sampling interval. Parametric modelling of fire sizes using covariates derived from forest inventory data shows that the expected size of a fire is positively related to the abundance of pine forest in the vicinity of the point of detection and negatively related to the abundance of recently logged or burnt areas. This implies that variation in forest structure and disturbance history impose marked spatial variability on the fire size distribution. Other covariates, such as periodic indices of fire weather, could readily be evaluated in this framework.