The effect of fire channelling on fire severity in the 2009 Victorian fires, Australia
© International Congress on Modelling and Simulation, MODSIM 2013.All right reserved. Empirical studies of the 2003 fires in Canberra have highlighted a phenomenon called fire channelling, whereby a fire spreads rapidly laterally across lee-facing slopes, thus increasing the rate of spread and the power of the fire. Analyses of wind, terrain and fire data have identified thresholds in slope and aspect relative to the wind, which define necessary conditions for the phenomenon to occur. It is expected that areas affected by fire channelling will burn with high severity. Separate empirical analysis of fire severity patterns in the Victorian fires of 2009 identified higher prevalence of crown fire on leeward slopes as compared to that on windward slopes. In this study, we reconcile these observations by using the thresholds to identify where fire channelling was likely to have occurred in the 2009 fires, and reanalysing the severity patterns to quantify the extent to which high severity corresponded to the identified fire channelling prone areas. The sample is 4500 regularly spaced points (500 m separation) from four time periods within four fires. Fire channelling prone parts of the landscape were identified using a binary function χ(σ,δ), which depends on a threshold slope σ and a threshold aspect discrepancy δ. Fire channelling prone parts of the landscape were defined as those satisfying χ(σ,δ) = 1, with other parts of the landscape characterized by χ(σ,δ) = 0. Values of χ(σ,δ) were calculated using three different DEM resolutions: 30 m, 90 m and 240 m. In each case a binary grid was obtained across the landscape, which was then used as a predictor of fire severity. Initial analyses considered χ(σ,δ) as the only predictor of fire severity, but subsequent analyses followed a full model selection process as was done by Price and Bradstock (2012), though in this case with χ(σ,δ) replacing aspect. This approach identified the best model and supported alternatives, including testing all two-way interactions between variables. The best model obtained in this way was compared with the model developed by Price and Bradstock (2012) (i.e. with aspect). The analyses revealed that locations identified as prone to the fire channelling phenomenon were more likely to experience crown fire, and that this effect was most obvious at larger spatial scales. At the largest spatial scale considered, crown fire occurred in 27.2% of fire channelling prone points and 20.5% of non-fire channelling prone points (33% greater). The binomial regression analyses also indicated that when fitted on its own fire channelling proneness was a significant predictor or fire severity. Overall, the effect of fire channelling on fire severity was substantial and amounted to an 11% increase in crown fire likelihood. Given that the analyses entailed a comparison with all parts of the landscape, including windward slopes where the highest likelihood of crown fire would normally be expected, the effect is very important. Combined with information arising from other extreme bushfires, including the 2003 Canberra fires, the analyses considered here suggest that fire channelling may be a common phenomenon in large and intense fires. The fire channelling phenomenon therefore poses some serious challenges for firefighting, fire spread prediction, the management of environmental assets and risk planning. Given the effects of fire channelling on fire severity demonstrated by this study, this will be particularly pertinent in high-value water catchments. The results of the analyses also have significant implications for the management of soil erosion and biodiversity conservation in rugged terrain after large fires.