On the distribution of wildfire sizes

A variety of models for the wildfire size distribution are examined using data on Los Angeles County wildfires greater than 100 acres between 1950 and 2000. In addition to graphs and likelihood criteria, Kolmogorov-Smirnoff and Cramer-von Mises statistics are used to compare the models. The tapered Pareto distribution appears to fit the data quite well and offers some advantages over the untapered Pareto distribution, while alternatives including the lognormal, half-normal, exponential and extremal distributions fit poorly. The size distribution appears to be quite stable over the examination period, though inspection of the transformed wildfire sizes for the tapered Pareto reveals some limited trend in the residuals, indicating a very slight gradual decrease in the average fire size in Los Angeles County over these 50 years. Copyright (C) 2003 John Wiley Sons, Ltd.