A Space-Time Conditional Intensity Model for Evaluating a Wildfire Hazard Index.

Numerical indices are commonly used as tools to aid wildfire management and hazard assessment. Although the use of such indices is widespread, assessment of these indices in their respective regions of application is rare. We evaluate the effectiveness of the burning index (BI) for predicting wildfire occurrences in Los Angeles County, California using space-time point-process models. These models are based on an additive decomposition of the conditional intensity, with separate terms used to describe spatial and seasonal variability as well as contributions from the BI. We fit the models to wildfire and BI data from the years 1976-2000 using a combination of nonparametric kernel-smoothing methods and parametric maximum likelihood. In addition to using the Akaike information criterion (AIC) to compare competing models, we use new multidimensional residual methods based on approximate random thinning and rescaling to detect departures from the models and to ascertain the precise contribution of the BI to predicting wildfire occurrence. We find that although the BI appears to have a positive impact on wildfire prediction, the contribution is relatively small after taking into account natural seasonal and spatial variation. In particular, the BI does not appear to take into account increased activity during the years 1979-1981 and can overpredict during the early months of the year. [ABSTRACT FROM AUTHOR]