A State Space Model for predicting Wildland Fire Risk

Wildland fire managers have long desired to know the risks of severe fire events well in advance of its happening. A number of actions are available to address severe fire seasons. However, contingency actions have associated costs and timeliness issues. These issues require information about the likelihood of fire occurrence. In this talk we describe a probability model for predicting wildland fire risks based on non-linear state-space models. By the choice of variables included in the state space, we can handle many situations, for example, the history of the process (locations and times of fires) up to the present in addition to characteristics of the environment that might serve as explanatories. We found the model to be useful for assessing the importance of commonly used fire danger indices and for predicting expected numbers of fires in a region.