Published works

The objective of the Probabilistic Framework Project is to develop a new consequence-based fire danger rating system able to integrate a wide range of variables and link their complex interactions to the probability of property loss. The project aims at delivering a spatially-explicit framework capable of generating daily maps representing the distribution of the probability of property loss at 5km spatial resolution.

In year two, a “consequence-based” system (developed in Year 1) was refined and applied in two case study regions: the Sydney Basin, and the Victorian East Central Risk Landscape. The BN framework was successfully integrated with GIS facilities to generate spatially explicit predictions of the probability of a fire spreading to and reaching the urban interface and then burning there at high intensity (>4000kWm-1, hereafter unsuppressible fires). Overall, the model indicated that the highest risk areas may potentially be identified by accounting for not only fire weather, but also fuels, the distribution of property, plus features inherent in the landscape that influenced fire spread.

The objective for the third stage of the project was to develop and test an operational application of the daily fire danger rating Bayesian Network model for two case study areas - Sydney Basin and the East Central Risk Landscape in Victoria. In this report, we present the results of the operational application of the model, quantify sensitivities of the model and finally we make recommendations for the future of the modelling approach.
We developed a prototype Fire Danger Rating (FDR) website for updating fire risk in real-time (30-60 seconds per case study). Maps are generated through use of landscape data and additional daily 3pm weather data downloaded from the Bureau of Meteorology. Across the 2014/15 season spatial variation was seen between the predictions from the FDR and that of FFDI. It is important to note that these variations exist despite the fact FFDI is an input into the model influencing ignition probability, fire spread and fire intensity. One of the main reasons for these differences is that FFDI only accounts for weather and does not consider topography, fuels, spatial arrangement of assets or the directionality of the wind.

The sensitivity analysis suggested that the model is performing well relative to expectations. Logical relationships and coarse scale patterns are holding true. The results indicate strong reliance on the empirical analysis of ignition probabilities in the landscape. FFDI was found to be the input node that required the greatest accuracy.

A number of recommendations were made by state agencies during an end of study review of the project. These included expanding the FDR website to new landscapes, with particular interest in assessing the model for grassland environments. In addition, the group thought the model had considerable capacity for longer term planning of fuel treatments, accounting for changing human patterns and future climates.

Overall, the project has succeeded in delivering a spatially-explicit framework capable of generating daily maps representing the distribution of the probability of property loss at 5km spatial resolution.