WRF-Fire Simulation of Lateral Fire Spread in the Bendora Fire on 18 January 2003.

On the afternoon of 18 January 2003, a number of separate fires located to the west of Canberra, Australia, began major runs under extreme fire weather conditions, and impacted upon the city. These events were well documented by a range of instruments, including a multispectral line-scanning instrument attached to an aircraft. Analysis of the data collected revealed that a number of the fires had exhibited atypical lateral fire spread, in a direction transverse to the background wind, on steep leeward slopes. These lateral fire spread events often contributed considerably to the size and impact of the fire. In one particular instance, a fire burning to the west of Bendora Dam (35◦ 28’ S, 148◦ 50 E), which had been burning for the previous ten days subject to control efforts, breached control lines and rapidly developed into a large conflagration. The Bendora fire then ran into the edge of urban areas, where it combined with other fires that had escalated significantly on that day. The fact that the Bendora fire developed so rapidly, from a relatively small breach of the control line, attests to the abrupt transitions that bushfires can exhibit under extreme weather conditions. In this study, the WRF numerical weather prediction model was coupled to the WRF-Fire wildland fire physics module at high resolution and used to simulate the early development of the Bendora fire on 18 January. The modelled fire spread was compared to the multispectral line-scan data with the fire to atmosphere coupling enabled and disabled. With the coupling enabled, the fire advanced around 1 km further laterally to the south, and this lateral fire spread occurred predominantly in the lee of a ridge. The lateral fire spread was partly driven by pyrogenic vorticity that formed in the lee of the ridge due to the interaction between the opposing pyrogenic and background winds. Additionally, differences between the modelled and actual fire spread on the windward side of ridges suggest that more careful consideration of the combined effects of wind and slope on the rate of spread is required in future versions of WRF-Fire. A large number of near-surface vortices, with a large component of vertical vorticity, were identified over the leeward slopes and downwind of the fire when the fire to atmosphere coupling was enabled. Additionally, a region of high turbulent kinetic energy extending to the southeast of the fire supports the notion that the fire was carried across the Bendora Reservoir by mid to long-range spotting. As there is no spotting model in WRF and WRF-Fire, the fire was unable to cross the Bendora Reservoir in the numerical simulations, as was observed in the multispectral line-scan data. The results demonstrate that WRF and WRF-Fire can model atypical lateral fire spread across steep leeward slopes in more realistic terrain than has previously been considered. It may therefore be possible to investigate other known atypical lateral spread events, such as the 2003 Broken Cart fire in Canberra, the 2009 Jesusita fire in Santa Barbara and the 2013 Wambelong fire near Coonabarabran, using this or a similar model configu- ration. However, we caution that these results were obtained using a single coupled atmosphere-fire model for a highly idealised configuration, and further work is required to replicate this fire behaviour in other coupled models.

Figure 1. Top (a): multispectral line-scan image of the region surrounding the Bendora fire, taken at 1446 on 18 January 2003. The yellow colour is indicative of active flame (Sharples et al., 2012). Bottom (b,c): time of ignition for the non-coupled (left) and coupled (right) simulations, with terrain height contours at 20 m intervals. The solid black region indicates the spatial extent of the Bendora Reservoir taken from the SRTM Water Body Dataset. The circular markers show the time and location of vortices identified by a fire whirl identification algorithm. The dashed region shows the ignition location, which approximates the location of the control line breach that occurred.