Developing a Probabilistic Fire Risk Model and Its Application to Fire Danger Systems
Presentation at Research Forum of the 2012 Bushfire CRC and AFAC Annual Conference.
Wildfires can result in significant economic losses. Management agencies have large budgets devoted to both prevention and suppression of fires, however the extent to which they alter the probability of asset loss is not quantified. Prediction of the risk of asset loss requires an understanding of complex processes from ignition, fire growth and impact on assets. These processes need to account for the effects of management, weather and the natural environment. Traditional analytical methods can only examine only a subset of these. Bayesian Belief Networks (BBNs) provide a methodology to examine complex environmental problems. Outcomes of a BBN are represented as likelihoods, which can then form the basis for risk analysis and management. Here we combine a range of data sources, including simulation models, empirical statistical analyses and expert opinion to form a fire management BBN. Various management actions have been incorporated into the model including landscape and interface prescribed burning, initial attack and fire suppression. Performance of the model has been tested against fire history datasets with strong correlations being found. Adapting the BBN presented here we are capable of developing a spatial and temporal fire danger rating system. Currently Australian fire danger rating systems are based on the weather. Our model accounts for existing fires, as well as the risk of new ignitions combined with probabilistic weather forecasts to identify those areas which are most at risk of asset loss. Fire growth is modelled with consideration given to management prevention efforts, as well as suppression resources that are available in each geographic locality. At a 10km resolution the model will provide a probability of asset loss which represents a significant step forward in the level of information that can be provided to the general public.