Sensitivity Analysis of a Model of the 2003 Simi Wildfire Event

A sensitivity analysis of the FARSITE fire spread model was performed using the fall 2003 Simi wildfire event as a test case. Model inputs that were varied included two different fuels maps and two different sources of wind speed and direction. The model parameter that was varied was perimeter resolution, essentially the spatial resolution at which the model was run. Model accuracy was assessed through a comparison with the MODIS active fires product, using the Lee-Sallee spatial correspondence metric. The FRAP fuels map predicted initial fire growth most accurately, but produced too large of a burn at the later time steps. A fuels map created from a classification of Landsat imagery, smoothed to 270m cells, produced the most accurate fire size at the later time steps. Spatially varying wind inputs increased variability in model output. This research demonstrates that FARSITE is sensitive to model inputs and parameters and suggests that fire behavior analysts should perform multiple runs in order to bracket the range of possible fire behavior