Bushfire prediction in Victoria, Australia using Generalized extreme value distribution

Abstract

To facilitate appropriate design, management, and improvement of infrastructures for both rural and urban communities, the prediction for the occurrence of future extreme climatic events e.g. bushfire is intrinsic. Widely used extreme climatic event prediction method, Generalized Extreme Value (GEV), was used in this study in predicting annual peak Forest Fire Danger Index (FFDI), which is a measure of fire behaviour used in different parts of Australia. The GEV distribution was fitted for 17 selected stations spreading all around Victoria, Australia. The estimation of the parameters for the GEV distribution was performed using Maximum Likelihood Estimation (MLE) method. Three goodness of fit tests such as: Kolmogorov-Smirnov (KS) test, Anderson-Darling (AD) test, and Chi-square test result were used to verify the efficacy of GEV distribution. Fire frequency curves (FFCs) were derived for Victoria to identify fire prone regions. The study revealed that Fréchet (type II) extreme value distribution is suitable for modelling the annual peak of FFDI for most of the selected stations. Annual peak FFDI for the 100 years return level (time average of every 100 years) can vary from 34 (Dartmouth) i.e. high fire danger event to 146 (Walpeup) i.e. catstrophic bushfire event. The study noted that three stations in Victoria namely: Nhill, Walpeup and Ouyen are vulnerable to catastrophic fire danger situation (FFDI ≥100) with at least 1% probability of occurrence (1 in 100 year). The developed FFCs for Victoria will guide key public and private stakeholders namely: catchment management authorities, insurance companies, structural designers, traffic modellers, and forest hydrologists when designing and managing infrastructures in fire prone areas.