Where's the fire? Quantifying uncertainty in a wildfire threat model

Models of wildfire threat are often used in the management of fire-prone areas for purposes such as planning fire education campaigns and the deployment of fire prevention and suppression resources. While the use of spatial or geographic data is common to all wildfire threat models, the key question arises: Is the accuracy of the spatial data used in wildfire threat models sufficient for the intended decision-making purpose? To help answer this question, a quantitative uncertainty assessment technique was applied to a wildfire threat model used by the Country Fire Authority in Victoria, Australia. The technique simulates known or estimated spatial data error by modifying data values to represent the range of all probable errors present in the input dataset. The wildfire threat model is then run multiple times using these modified ‘error’ layers in order to simulate and observe the effect these errors have on the model outputs. For the model concerned, the results suggest that errors in digital elevation surfaces have only minimal impact upon the outputs, resulting in relatively stable wildfire management decisions. On the other hand inaccuracies in land cover maps (with implied differences in fuel load estimations) result in larger changes in the model outputs, whereas changes in fire weather data can result in highly unstable outputs. Knowledge of these effects can facilitate better wildfire management since any improvements that are to be made to the model’s accuracy can be focussed directly upon the problem datasets.