High-fidelity ensemble simulations of large-scale wildfires for fire-impact assessment
Abstract
With the increasing severity and frequency of large wildfires, there is a critical need for improved modeling capabilities to inform mitigation plans for fire management and risk mitigation. In particular, high-fidelity modeling tools are needed to provide reliable predictions of fire-spread behavior to support scientific inquiry and fire-risk assessment, as well as landscape
management at an early stage. However, because of the computational complexity, physics-
based models are largely limited to simulating a few conditions. We present a high-fidelity
simulation framework that takes advantage of emerging programming paradigms, novel
computing hardware architecture, and ensemble calculations for simulating large-scale wildfires scenarios at affordable cost, thereby enabling the parametric study and statistical analysis of wildfires scenarios under consideration of changing environmental conditions, ignition probabilities, and vegetation and fuel-moisture regimes. We discuss details of the simulation framework that is based on TensorFlow and the utility of ensemble simulations to examine fire- spread behavior in the presence of coupled wind-slope conditions that remain an outstanding scientific challenge for fire-spread predictions.
management at an early stage. However, because of the computational complexity, physics-
based models are largely limited to simulating a few conditions. We present a high-fidelity
simulation framework that takes advantage of emerging programming paradigms, novel
computing hardware architecture, and ensemble calculations for simulating large-scale wildfires scenarios at affordable cost, thereby enabling the parametric study and statistical analysis of wildfires scenarios under consideration of changing environmental conditions, ignition probabilities, and vegetation and fuel-moisture regimes. We discuss details of the simulation framework that is based on TensorFlow and the utility of ensemble simulations to examine fire- spread behavior in the presence of coupled wind-slope conditions that remain an outstanding scientific challenge for fire-spread predictions.
