Physics-Based Modeling of Fire Behavior and Smoke Plume Development, How Much is Enough?

William Mell | U.S. Forest Service



Over the last decade there has been a significant increase in the availability of affordable and capable computing power (e.g., Amazon Web Services cloud computing). As a result, there is a potential to use more advanced (i.e., greater physical fidelity) computational fluid dynamics (CFD) based fire behavior and smoke plume development models as tools for prescribed burn planning. There are two main questions addressed in this project from a model usability point view:

  1. What is the lowest level of CFD-based model physical fidelity that results in predictions that are useful for the application in question (e.g., plume rise)?
  2. How computationally expensive are the models considered here? For example, can they be run on a desktop computer or are cloud computing resources needed? Provide measures of computing time and cost in dollars.

Another objective is the development of a prototype tool for running simulations on Amazon Web Services (AWS).

Back to Top

Technical Approach

Predictions from four fire behavior and three smoke plume models were compared for a number of test scenarios. These models, in order of decreasing physical fidelity are: Wildland-urban interface Fire Dynamics Simulator model suite (WFDS) using the full physics-based model for fire behavior, WFDS-LS1, WFDS-LS52, WFDS-LS13 (for fire behavior) and Fire Dynamics Simulator from National Institute of Standards and Technology, A Large Oil Fire plume Trajectory – Flat Terrain, Mercer and Weber (1994) (for smoke plume development). Limited model validation, through comparisons to observations, were conducted. Given that the range of relevant observational studies needed for a more complete model validation does not currently exist, the highest physical fidelity models were used as a standard against which the lower fidelity models were compared in the test scenarios.

1WFDS-LS: WFDS using the level set with full fire-atmosphere coupling and an empirical formula for the head fire spread rate dependence on the local wind (i.e., there is no explicit modeling of the thermal degradation of vegetation).

2WFDS-LS5: WFDS using the level set model with partial fire-atmosphere coupling and a constant head fire spread rate.

3WFDS-LS1: WFDS using the level set model with no fire-atmosphere interaction. Approach is consistent with, and can be made essentially identical to, the FARSITE model.

Back to Top

Interim Results

Fire behavior focused simulations, for which qualitative results are sufficient, ran 2 - 3.4 times (depending on model fidelity) slower than real time in a 300 meter x 300 meter domain. Higher resolution simulations that predicted locations of understory consumption ran 23 times slower than real time. These simulations were run on a high-end desktop (25 cpus). For single line fires, all smoke plume models gave similar centerline heights. The simulations suggest that multiple, spreading, fires can be approximated by static fires and simulated on coarse grids, allowing a relatively fast computation. Plume interaction from multiple fires can affect plume height and requires CFD-based models, the simplest of which ran significantly faster than real time. A prototype user interface for running CFD-based simulations on AWS was demonstrated.

Back to Top


Overall, the results of this project suggest that, depending on domain size and user needs, semi-routine CFD-based fire behavior and smoke plume simulations are within reach for use as prescribed-burn planning tools. This is potentially the case with a high-end desktop computer and definitely the case if cloud computing is available. Model runs with relatively quick turnaround times and qualitative accuracy could be used in prescribed burn training or to illustrate fire behavior relevant to fire safety. More complete model evaluation is needed (through both model-to-model comparison and model comparison to observations). It is essential that further development of the modeling tools includes the collaboration of prescribed burn planners who are active beta testers. In addition, the model source code and executables need to be open source with readily available and regularly updated user guides. This ensures transparency and supports a more collaborative advance of the modeling tools.

Back to Top

Points of Contact

Principal Investigator

Mr. William Mell

U.S. Forest Service

Phone: 206-430-2072

Program Manager

Resource Conservation and Resiliency