LES of an ensemble of wildfire spread

Description

The propagation of wildfires is a complex, dynamic process that is influenced by various factors, such as fuel, wind, terrain, and other environmental conditions. Accurately and reliably predicting the rate-of-spread of wildfires is of critical importance for fire management, rapid fire response, and fire mitigation. The Google FireBench dataset [1] aims to provide high-fidelity data to tackle these issues by providing an ensemble of large-eddy simulations that capture the three-dimensional wildfire-spread behavior and coupling with the atmospheric environment.

The spatial and temporal evolution of the combustion of solid fuel coupled with the atmospheric flow is described by a two-phase model [2]. The gas-phase is described by the Favre-filtered conservation equations for mass, momentum, oxygen-fraction, and potential temperature [3]: $$ \partial_t \overline{\rho} + \nabla \cdot (\overline{\rho} \widetilde{\boldsymbol{u}}) = S_\rho, $$ $$ \partial_t (\overline{\rho} \widetilde{\boldsymbol{u}} ) + \nabla \cdot (\overline{\rho} \widetilde{\boldsymbol{u}} \otimes \widetilde{\boldsymbol{u}}) = - \nabla \overline{p_d} + \nabla \cdot \overline{\tau} + [\overline{\rho} - \rho(z)] g \boldsymbol{\hat{k}_z} + \boldsymbol{f}_D + \boldsymbol{f}_C, $$ $$ \partial_t (\overline{\rho} \widetilde{Y_O}) + \nabla \cdot (\overline{\rho} \widetilde{\boldsymbol{u}} \widetilde{Y_O}) = \nabla \cdot \overline{\boldsymbol{j}_O} + \overline{\rho} \widetilde{\dot{\omega}_O}, $$ $$ \partial_t (\overline{\rho} \widetilde{\theta}) + \nabla \cdot (\overline{\rho} \widetilde{\boldsymbol{u}} \widetilde{\theta}) = \nabla \cdot \overline{\boldsymbol{q}} + \frac{\overline{\rho} \widetilde{\theta}}{c_p \widetilde{T}} [h a_v (T_s - \widetilde{T}) + \dot{q}_r + (1-\Theta) H_f \widetilde{\dot{\omega}}], $$ where $\widetilde{\cdot}$ denotes Favre-filtering and $\overline{\cdot}$ denotes Reynolds filtering. $\rho$ is the density, $\boldsymbol{u}$ is the velocity vector, $p_d$ is the hydrodynamic pressure, $\tau$ is the shear stress tensor, $g$ is the gravitational acceleration, $\boldsymbol{\hat{k}_z}$ is the unit vector along the gravitational direction, $f_D = - \overline{\rho} c_d a_v \boldsymbol{|\widetilde{u}| \widetilde{u}}$ is the drag force due to surface vegetation, $\boldsymbol{f}_C = f \boldsymbol{\hat{k}_z} \times \overline{\rho} (\widetilde{\boldsymbol{u}} - \boldsymbol{U}_\infty)$ is the Coriolis force, $Y_O$, $\boldsymbol{j}_O$, and $\dot{\omega}_O$ are the mass fraction, species diffusion, and source term of the oxidizer, $\theta$ is the potential temperature, $\boldsymbol{q}$ is the heat flux vector, $T$ is the gas-phase temperature, and $H_f$ is the heat of combustion. The heat exchange between the solid and gas phase is modeled with $h$ as the convective heat transfer coefficient, $a_v$ as the bulk fuel area-to-volume ratio, and $\dot{q}_r$ is the radiation source term. $\Theta = 1 - \rho_f/\rho_{f,0}$ is the fraction of the heat release that contributes to the increase of the solid phase temperature. $\dot{\omega}$ is the gas-phase combustion source term.

The dataset consists of 117 cases with 9 velocities and 13 slopes with data extracted 1.5 m and 10 m above ground level. In addition, data was extracted at a streamwise location of 100 m < x < 1000 m. Specifically, the cases span a range of mean inlet velocity at 10 m above ground level of 2 to 10 m/s with a step of 1 m/s, and a range of slopes from 0 to 30 degrees with steps of 2.5 degrees.

Quick Info

  • Contributors: Qing Wang, Matthias Ihme, Cenk Gazen, Yi-Fan Chen, John Anderson, Jen Zen Ho, Bassem Akoush
  • Nx = 900, Ny = 252
  • DOI
  • .bib
ID Conditions Size (GB) Links
0 u10 = 2 m/s 68 Kaggle
1 u10 = 3 m/s 42 Kaggle
2 u10 = 4 m/s 42 Kaggle
3 u10 = 5 m/s 42 Kaggle
4 u10 = 6 m/s 42 Kaggle
5 u10 = 7 m/s 42 Kaggle
6 u10 = 8 m/s 60 Kaggle
7 u10 = 9 m/s 42 Kaggle
8 u10 = 10 m/s 51 Kaggle

References

[1]. Q. Wang, M. Ihme, C. Gazen, Y. F. Chen, J. Anderson. A high-fidelity ensemble simulation framework for interrogating wildland-fire behaviour and benchmarking machine learning models. International journal of wildland fire (2024).

[2]. R. R. Linn. A transport model for prediction of wildfire behavior (No. LA-13334-T). PhD thesis. Los Alamos National Lab., NM, United States (1997).

[3]. Q. Wang, M. Ihme, R. R. Linn, Y. F. Chen, V. Yang, F. Sha, C. Clements, J. S. McDanold, J. Anderson. A high-resolution large-eddy simulation framework for wildland fire predictions using TensorFlow. International journal of wildland fire (2023).

Updated: