Modelling firebreaks in a two-dimensional dynamic fire spread simulator

Abstract

Wildfires cause damage to infrastructure, houses and ecosystems every year in many countries across the world. Modelling of large-scale wildfire propagation is important for both long-term future planning and operational fire management such as decisions around evacuation and suppression tactics. Disruptions such as firebreaks, roads and rivers can completely halt, partially inhibit or have very little effect on the spread of wildfires. The behaviour of the fire at these disruptions depends on factors such as the width of the disruption, the fire intensity and type and condition of the fuel the wildfire is burning through. However, there is currently limited understanding as to the best method for representing such disruptions in wildfire models and quantifying their effect. Here we investigate a range of potential implementations for linear disruptions, and apply these to a simplified disruption scenario. In this study, probabilistic and deterministic methods of implementing a disruption within a fire spread simulator are investigated. The methods are implemented in Spark, a two-dimensional, raster-based level set solver. The scenario investigated in this study comprises a fire under constant wind and fuel conditions impacting a disruption, representing a firebreak, running perpendicular to the wind direction. The simulation was started from an ignition with a radius of 10 metres, used a southerly wind with a constant speed of 30 km/h, homogeneous fuel type and load and ran for a one-hour period. McArthur’s Mk5 model was used for the rate of spread of the fire. A one metre wide disruption was imposed 100 metres downwind from the ignition point. For the probabilistic method, the disruption was implemented as a set of un-burnable raster cells, the state of which could be set to burnable (allowing the fire to cross the disruption) dependent on user-defined criteria. This criterion was based on a failure function giving the probability of a cell allowing the fire through at each time-step of the simulation. Care was taken to adjust the failure function for both the spatial resolution of the raster cells, and the dynamically changing simulation time step. Using a constant failure function and probabilities of crossing an individual cell of 0.2 and 0.4 respectively, two sets of 1,000 simulations ensembles were run. These simulations were used to obtain cumulative density functions for the length of time it took the fires to cross the firebreaks. Several deterministic approaches were investigated for emulating the ensemble behaviour of the probabilistic methods, as ensemble simulations might not be feasible for some time constrained simulations (such as operational fire prediction) especially if one has limited computational resources available. These included treating the firebreak as slowly burnable or with a time delay before the fire was allowed to cross the disruption. These deterministic methods yielded similar results to the median of the probabilistic burn area histograms. However, time delays much longer than the median crossing time were required to achieve this. Overestimation of burn areas using the median crossing time is likely due to the existence of multiple or continuous crossing locations compared to the few discrete crossings in the probabilistic implementation. Despite this, the deterministic methods presented here will be useful to emulate probabilistic methods and could be used in fire simulators to improve predictions involving disruptions.