Catastrophic wildfires in California have become more frequent in past decades, while insured losses per event have been rising substantially. On average, California ranks the highest among states in the U.S. in the number of fires as well as the number of acres burned each year. The study of catastrophic wildfire models plays an important role in the prevention and mitigation of such disasters. Accurate forecasts of potential large fires assist fire managers in preparing resources and strategic planning for fire suppression. Furthermore, fire forecasting can a priori inform insurers on potential financial losses due to large fires. This paper describes a probabilistic model for predicting wildland fire risks using the two-stage Heckman procedure. Using 37 years of spatial and temporal information on weather and fire records in Southern California, this model measures the probability of a fire occurring and estimates the expected size of the fire on a given day and location, offering a technique to predict and forecast wildfire occurrences based on weather information that is readily available at low cost.
biased sampling forest fires fire occurrence probabilities fire weather