Citation

Plucinski MP, McCaw WL, Gould JS, Wotton BM. Int. J. Wildland Fire 2014; 23(4): 520-531.

Copyright

(Copyright © 2014, International Association of Wildland Fire, Fire Research Institute, Publisher CSIRO Publishing)

DOI

10.1071/WF13090

PMID

unavailable

Abstract

Data from bushfire incidents in south-west Western Australia from the Departments of Parks and Wildlife and Fire and Emergency Services were used to develop models that predict the number of human-caused bushfires within 10 management areas. Fire incident data were compiled with weather variables, binary classifications of day types (e.g. school days) and counts of the number of fires that occurred over recent days. Models were developed using negative binomial regression with a dataset covering 3 years and evaluated using data from an independent year. A common model form that included variables relating to fuel moisture content, the number of recent human-caused bushfires, work day (binary classification separating weekends and public holidays from other days) and rainfall was applied to all areas. The model had reasonable fit statistics across all management areas, but showed enough day-to-day prediction variability to be of practical use only in the more densely populated management areas, which were dominated by deliberate ignitions. The findings of this study should be of interest to fire managers in Mediterranean climatic regions where a variety of practices are used to manage wildfires.