Development of spatial models for bushfire occurrence in South-Eastern Australia

Abstract

Australia is one of the most flammable continents in the world. The southeastern region of the continent, where the population is densely settled, is susceptible to low frequency and high intensity bushfires that can threaten human lives and cause extensive loss of properties. This paper describes the development and validation of spatial models for bushfire occurrence in South-Eastern Australia, especially in New South Wales, Victoria and the Australian Capital Territory. The active fire database from the Moderate Resolution Imaging Spectroradiometer (MODIS) is used as original data source of fire activity over the 11-year period 2003-2013. Those active fire detections are grouped into burning events using the Fire Spread Reconstruction approach (FSR) algorithm based on the spatial and temporal proximity between fire detections. Fire occurrence point is defined as the ignition point of each identified event. Univariate and multiple logistic regression models are investigated for the comprehensive understanding of determinants contributing to the spatial distribution of bushfires. The probability of bushfire occurrence in South-Eastern Australia is also studied for the prediction of future fire occurrence. Bushfires in the study area are significantly influenced by both environmental and anthropogenic variables. The mean annual precipitation positively influences the fire incidence, because the semi-arid regions lack the fuels necessary for a fire to start, while the coastal regions with abundant rain provide ample fuels for fire ignition. This finding is inconsistent with that at a small landscape scale. Fire probabilities are different regarding various land cover types. Forests are most likely to burn because they are covered by heavy fuel loads. Savannas are equivalently fire-prone because they are fundamentally easy to ignite. Permanent wetlands are also susceptible to fire possibly due to the influence of climate change and urban expansion. Shrublands are less fire-prone because of the low-level shrub canopy cover. Fires are also found to distribute in areas near the zero meso-scale elevation residual contour, which is consistent with the previous finding. Anthropogenic variables also show predictive power because of the influence of human activities on fire occurrence. The final model for the probability of bushfire occurrence include mean annual precipitation, MODIS land cover, distance to zero meso-scale elevation residual contour, distance to secondary road and distance to railway. The bushfire probability map was generated accordingly. From the information provided by the quantitative statistics and the bushfire probability map, bushfires in the study area mostly likely to occur in coastal and mountainous areas close to various types of infrastructure and zero meso-scale elevation residual contours, as well as on forests, savannas and permanent wetlands, while they rarely occurred inland. It is concluded that the proposed model provides practical guidance for fire management actions in South-Eastern Australia.