Uncertainty Quantification of Wildfire Forecast Error

Abstract:

Wildfire evolution depends upon the fire interaction with local wind conditions, in particular with unsteady atmospheric turbulent structures. In order to predict the fire front propagation, computational models to capture the coupled fire-atmosphere dynamics have been developed in the past decades. These models typically consist of an atmospheric flow solver (for example, based on large-eddy simulations), which provides the driving wind field, and a fire-front spreading model, which parameterizes the combustion process and tracks the evolution of the fire front and the heat released. However, the physical processes affecting local wind conditions, hence wildfire spreading, occur across multiple spatio-temporal scales, ranging from mesoscale circulation (on the order of kilometers) to small-scale turbulence (on the order of meters). The atmospheric solver is generally the most computationally intensive task of the numerical model, and for this reason the grid resolution is usually limited to spacings in the order of 100 m or larger. Thus, the effect of smaller-scale structures on the fire evolution is filtered out increasing the uncertainty in the prediction of wildfire propagation. In this work, we seek to quantify such uncertainty as a function of the numerical resolution and the time elapsed from ignition. We propose a methodology based on polynomial chaos expansion (PCE). PCE provides a response function for a model output (in this case, the forecast error) in terms of orthogonal polynomials in the input parameters (resolution and elapsed time). The response function is herein obtained from a set of numerical simulations of wildfires with varying degree of resolution. The simulations are performed with an in-house coupled fire-atmosphere code, based on large-eddy simulations and the Rothermel fire spread model. This methodology provides an operational tool to determine the wildfire forecast accuracy as a function of resolution and time elapsed from ignition.