Quantifying the environmental limits to fire spread in grassy ecosystems

Anabelle W. Cardoso and Carla Staver, Ecology and Evolutionary Biology, Yale University, New Haven, CT, Sally Archibald, Natural Resources and Environment, CSIR, Pretoria, South Africa, William J. Bond, South African Environmental Observation Network, National Research Foundation, Cape Town, South Africa, Matthew Forrest, Senckenberg Biodiversity and Climate Reasearch Centre, Frankfurt, Germany, Navashni Govender and Tercia Strydom, Scientific Services Kruger National Park, Skukuza, South Africa, David Lehmann, Loic Makaga, Josue Edzang Ndong and Flore Koumba Pambo, Agence Nationale des Parcs Nationaux (ANPN), Libreville, Gabon, David Tilman, Bren School of Environmental Science & Management, University of California Santa Barbara, Santa Barbara, CA, Peter D. Wragg, Department of Forest Resources, University of Minnesota, St. Paul, MN

Background/Question/Methods
Fire spread can be modelled like an infectious disease: a burning patch of fuel infects, or ignites, a neighbouring patch of fuel, which infects another neighbor, and so on. Emergent from such an infection model are fire spread thresholds, whereby a minimum threshold value of fuel connectivity and infection probability is needed for fire to spread. This non-linear, process-based approach to modelling fire is favoured by theoreticians, but many large-scale fire models apply a linear, quasi-empirical approach instead. In this study, we seek to resolve the apparent conflict between theoretical and applied fire models by addressing whether we should account for fire’s threshold dynamics when modelling fire spread at larger scales. We built an infection model to simulate fire spread and compared our simulations to over one thousand real savanna fires.
Results/Conclusions
Our simulations recapitulated observed patterns in real fires better than competing linear or non-linear models. Furthermore, real fires occurring at the fire spread threshold had a near perfect (slope=0.99) correlation with model predictions. Having established that fire spread threshold do govern fire spread at the landscape scale, we parameterized these thresholds using measured fuel and weather data. The fuel connectivity threshold was predicted by grass biomass while the infection probability threshold was predicted by fuel moisture content and, to a lesser extent, vapour pressure deficit. Altogether, results emphasize that modelling fire spread as a linear process is not appropriate, although patterns may appear linear when averaged over spatial and temporal heterogeneity.