Advancing Coupled Fire-Atmosphere Simulation: Fuel Representation and Fire Spotting
Loading...
Authors
Shamsaei, Kasra
Issue Date
2025
Type
Dissertation
Language
en_US
Keywords
Fire Spotting , Fuel , Machine Learning , Statistical Modeling , Wildfire
Alternative Title
Abstract
Wildland fires can lead to substantial ecological, social, and economical losses as well as health burdens due to smoke produced from biomass burning. Accurate wildland fire simulation is an essential capability in predicting and managing these losses, benefiting both pre-fire risk mitigation and preparedness as well as active-fire emergency response management. The goal of this research is to assess the performance and limitations of a state-of-the-art coupled wildland fire-atmosphere simulation platform, known as WRF-Fire, and advance the fire simulation capabilities by developing models and parametrization to introduce key missing physics in the current platforms. First, the performance of WRF-Fire is evaluated in simulating historic fires including the 2018 Camp Fire. Simulation results are compared with high temporal-resolution fire perimeters derived from weather radar observations, depicting non-negligible discrepancies between the simulated and observed rate of spread (ROS) and spread direction. Then, the sensitivity of the simulation to different modeling parameters and assumptions are investigated. While the sensitivity analyses show that refining the atmospheric grid in complex terrains improves fire prediction capabilities, the model still comprises systematic sources of modeling errors including errors in the fuel bed representation and lack of fire spotting capability. These two elements are the focal point of the advancements in the next step.
Wildland fuels not only drive the fire propagation, but also affect the generated fire heat, which leads to localized fire-induced weather conditions in the atmosphere. The preliminary simulation results of historic fires indicate the likely inaccurate characterization of the biomass fuel load, which results in the underestimation of heat fluxes in the simulation. WRF-Fire coupled fire-atmosphere models currently only consider the effects of surface fuels for heat flux calculation. Thus, the fuel bed representation in the platform is enhanced by adding canopy fuel load and a new heat release scheme. The results show that while the improved fuel bed and heat release scheme have limited impacts on the simulated fire perimeters, the resulting plume structure better agrees with the observations. This study provides guidance on the next level of scientific research and computational developments needed to further address the fuel characterization and heat flux calculation.
Another source of wildland fire modeling error is the lack of fire spotting simulation. Fire spotting is a process during which new fires are ignited in front of the active fire line by firebrands. For this process, a fire spotting framework capable of simulating firebrand generation, transport, and ignition is developed and implemented in the WRF-Fire recently. The objective of this work is to advance the current implementation and develop simple yet robust models for firebrand generation and spot fire ignition. For firebrand generation, a data-driven model is developed based on the existing experimental data. The model predicts distribution of firebrand mass and projected area given fuel characteristics and wind speed. Validation studies against independent experimental data show reasonable accuracy of the firebrand generation model.
Unlike the firebrand generation, limited experimental data are available for spot fire ignition due to its complex nature. Therefore, to develop a spot fire ignition model, a high-fidelity combustion-based fire simulation platform, known as Fire Dynamics Simulator (FDS), is utilized to simulate heat flux from smoldering firebrands with different properties. Energy balance and heat conduction theories are utilized to simulate the heat transfer between the firebrand and recipient fuel bed and its temperature increase due to firebrand heat flux. After validating the framework with experimental data, a dataset of recipient fuel ignition by firebrands given varying firebrand properties, as well as recipient fuel bed characteristics are generated. This dataset is then utilized to train machine learning models to enable spot fire ignition prediction given firebrand and fuel bed characteristics. The firebrand generation and spot fire ignition models are essential contributions to advance the coupled wildland fire-atmosphere simulation capabilities.
This dissertation contributes to developing solutions to address two key sources of modeling error within the existing wildland fire simulation platforms, namely heat simulation from non-surface fuels and fire spotting simulation, with the goal of improving the existing platforms accuracy in landscape-scale fire simulations. Specifically, this dissertation develops innovative and novel models for crown fire heat release, firebrand generation, and spot fire ignition of vegetative fuel beds. Moreover, the work contributes to shedding light on the critical scientific and technical challenges that can pave the way for future research and technical development for wildland fire simulation platforms.