Spatial estimates of fire risk in Victoria, Australia considering ignition likelihood and containment probability through Bayesian Network Analysis

Abstract

Accounting for multiple changing systems in environmental decision making is challenging and requires balancing several competing priorities. In fire risk, one approach which is increasingly used to capture uncertainty within multiple systems and to prioritise management efforts is Bayesian Network analysis. Here, we have used a Bayesian Network to understand the interactions between ignition likelihood, containment probability, fire behaviour, and fire weather, alongside the subsequent risks to people and property. We developed, populated and tested a Bayesian Network (BN) which classifies the likelihood of outcomes for each of these systems. We then apply this BN to grid of 72,000 potential ignition locations accross Victoria to predict house and life loss values under conditions capturing the top ten worst ranked weather days in the history of each location. We use Phoenix fire behaviour simulations and landscape scale raster data to populate the parent nodes for each ignition and extract the expected values for predicted nodes under different weather scenarios and varying levels of suppression. We found values predicted by the BN broadly matched the spatial patterns of risk produced in Phoenix i.e., areas where risk was highest and lowest in terms of fire area and house loss aligned. However, the values are rescaled by the BN as it takes into account the influence of ignition likelihood and containment probability on risk estimates. The BN is also able to capture uncertainty around the values presented from across the top ten Phoenix simulations, so the recorded values represent the likely outcome for each node given the range of potential weather conditions in those scenarios. We show that BNs can be a useful management tool for estimating fire risks across a range of weather scenarios and locations while still considering ignition likelihood and suppression effectiveness.