A stochastic fire spread model for north Patagonia based on fire occurrence maps

Understanding fire spread in different ecosystems is of fundamental importance for conservation, management and anticipating the effects of environmental changes. Tailoring existing fire spread models to particular landscapes is challenging because it demands a substantial data collection effort. Here we develop an objective way to fit simple stochastic fire spread models based on readily available data from documented fire events (i.e. approximate ignition point, preexisting vegetation, final perimeter, topography, and average wind direction). We use a simulation-based approach founded on Approximate Bayesian Computation, which allows for a thorough exploration of parameter space as well as the quantification of uncertainty around best estimates. As illustration, we use data from nine fire events that occurred during dry years in northern Patagonia, Argentina. We found that fire spreads readily in shrublands, while forests tend to act as firebreaks. Topography has a strong effect not only because fire moves easily upslope but also because it modulates wind direction. Finally, aspect affects fire spread mainly in forests, probably due to its effects on fuel moisture. Simulating fire spread sampling parameters from the approximated joint posterior distribution resulted in individual fires roughly similar to the ones used for model fitting. Furthermore, the fitted model was able to produce simulated fire-size distributions in good agreement with the historical record for dry years in Nahuel Huapi National Park, Patagonia. The approach presented here can be used in places where standard fuel models have not yet been developed.