A Technique for Developing a Fire Weather Index Using Surface Meteorological Observations and a Climatology of Daily Wildfire Occurrence

There is no universal definition for what represents a fire weather index because an index can be developed using different techniques depending on its intended purpose. Depending on the spatial and temporal scales involved, it is not even clear that the physical processes most relevant to predictions of fire danger and fire behavior are well understood. For this reason, fire weather indices often employ representations of physical processes in conjunction with statistical methods that link meteorological conditions and fire activity. However, it is unclear how to quantitatively assess the role of different meteorological parameters in the performance of a statistical index due to the often non-normal distribution of both the input variables and the outputs of the index. This presentation offers a statistical technique by which the impact of different meteorological input parameters on the performance of an index can be assessed quantitatively, to determine whether including the variable in a given index formulation improves the performance of the index.

For the purpose of this study, daily wildfire occurrence data and ASOS weather observations over the Northeast U.S. from 1999 to 2008 are employed. By comparing average values of meteorological variables on observed wildfire days to their climatological average using a bootstrap resampling approach, the potentially statistically significant weather variables are determined. Different formulations of a binomial logistic regression model using the potentially statistically significant variables are then tested to assess their ability to diagnose the probability of wildfire occurrence. Finally, a fire weather index is defined using probabilistic output from the optimal binomial logistic regression configuration. Independent verification with resampling is used to establish that the index is probabilistically more reliable and skillful at predicting wildfire occurrence than a climatology of observed fire occurrence probabilities.