Source-linked AI summary
Coupled atmosphere-wildland fire modeling with WRF-Fire
Jan Mandel, Jonathan D. Beezley, Adam K. Kochanski
TL;DR
Wildland fires affect millions of people and cause major damage, motivating better tools for modeling fire behavior. This paper describes WRF-Fire, a level-set fire-spread model coupled with WRF, and reports successful simulations while noting that validation remains in progress.
Problem
Wildland fires affect millions of people and cause major damage, creating a need for better tools to model fire behavior for suppression, controlled burns, and fire-danger assessment.
Method
The paper describes WRF-Fire’s physical model, numerical algorithms, software structure, level-set fire-spread module, coupling with WRF, ignition schemes, real-data inputs, and parallel computing.
Results
Successful WRF-Fire simulations reproduced fireline fingering, agreed well with FireFlux data, and obtained realistic fire and atmosphere behavior in a real-data Meadow Creek simulation.
Takeaways & Limitations
WRF-Fire is publicly available, supports ideal and real runs, and is currently suitable for research and education purposes.
Takeaways & Limitations
Validation is still in progress, and simulations can spread too quickly because ignition parameterization and unresolved fuel irregularities limit accuracy.
Abstract
from arXiv · showhide
We describe the physical model, numerical algorithms, and software structure of WRF-Fire. WRF-Fire consists of a fire-spread model, implemented by the level-set method, coupled with the Weather Research and Forecasting model. In every time step, the fire model inputs the surface wind, which drives the fire, and outputs the heat flux from the fire into the atmosphere, which in turn influences the atmosphere. The level-set method allows submesh representation of the burning region and flexible implementation of various ignition modes. WRF-Fire is distributed as a part of WRF and it uses the WRF parallel infrastructure for parallel computing.
1 Introduction
WRF-Fire addresses the need for practical wildland-fire modeling by coupling a mesoscale weather model with a semi-empirical fire-spread model. The paper documents its physical model, coupling, numerical infrastructure, new ignition and data features, and validation progress.
- Wildland fires motivate improved modeling tools for suppression management, controlled-burn planning, and fire-danger assessment.
- Two-way fire–atmosphere interaction is needed because wind drives fire spread while fire heat and vapor fluxes influence the atmosphere.
- WRF-Fire combines WRF and its ARW dynamical core with a semi-empirical fire-spread model for faster-than-real-time prediction.
- Compared with legacy CAWFE components, WRF-Fire uses a supported parallel community weather code designed for real runs and easier modification.
- The paper adds practically important ignition schemes, boundary-layer wind interpolation, parallel computing, real-data input, and reproducible coupling descriptions.
2 Physical fire model and fuels
The physical fire model uses a semi-empirical, fuel-dependent spread formulation together with exponential fuel consumption and heat-flux calculations. Fuel properties are configurable through preset or custom categories.
- The physical model specifies fire spread rate and heat fluxes, using a formulation based on CAWFE and BEHAVE.
- Fuels: Fuel is specified through 13 preset Anderson categories, while users may modify them or define completely new custom categories.
- Fire spread rate: The spread rate uses modified Rothermel components for no-wind spread, wind effects, and slope effects, with inputs from fuel properties, wind, and terrain.
- Fire spread rate: WRF-Fire includes a special chaparral formulation, limits slope and wind speed, and reduces wind from 6.1 m height to midflame height.
- Fuel burned and heat released: Each location begins with fuel fraction F = 1, then burns exponentially after ignition; Tf is the time to reach 1/e of the original fuel.
- Fuel burned and heat released: For compatibility with existing fuel models and literature, WRF-Fire inputs the coefficient w instead of the derived burn time Tf.
- Fuel burned and heat released: The model computes average sensible and latent heat flux densities, with latent heat conversion using a water-to-dry-fuel mass ratio of 0.56 and L = 2.5×10^6 J kg^-1.
3 Domain, grids, and nodes
WRF-Fire couples the fire model to the finest atmospheric domain within a nested, three-dimensional atmospheric grid. The atmospheric and fire components use distinct grid placements and resolutions.
- The atmospheric model uses a logically quadrilateral 3-D grid with horizontally nested domains, and only the finest domain couples to the fire model.
- Atmospheric scalar variables occupy 3-D cell centers, while wind components are stored at staggered midpoints of cell faces.
4 Mathematical core of the fire model
WRF-Fire represents the burning region with a level-set function and advances it using wind- and terrain-dependent spread rates. Fuel consumption and ignition are computed over subcells to represent burning and heat release below the fire-mesh scale.
- Level-set representation: The burning region is the set where ψ ≤ 0, the fireline is ψ = 0, and ψ > 0 denotes no fire.
- Level-set representation: The level-set equation advances the fireline using the normal spread rate S and gradient magnitude ∥∇ψ∥.S is evaluated throughout the domain, and S ≥ 0 prevents the fire area from decreasing.
- Numerical propagation: Heun’s second-order Runge–Kutta method advances the discretized level-set equation with upwinding and artificial viscosity.
- Numerical propagation: Boundary extrapolation is constrained to prevent ψ from decreasing and triggering a spurious fire near the boundary.
- Fuel consumption and heat release: Fuel fraction is integrated over each fire-mesh cell, while subcell interpolation of ψ and ignition time represents gradual burning as the fireline crosses a cell.Cells are split into four subcells, and heat fluxes are later aggregated from fire cells to atmospheric cells.
- Fuel consumption and heat release: The fuel-fraction approximation is asymptotically accurate for slow burning and exact burning-area approximation, while the heat-release scheme preserves total heat released over time.
5 Atmospheric model
WRF-Fire couples its fire module to WRF-ARW’s prognostic atmospheric system, formulated in a hydrostatic-pressure vertical coordinate. Atmospheric tendencies are integrated with an explicit third-order Runge–Kutta method.
- Atmospheric formulation: WRF-ARW uses hydrostatic pressure η as its vertical coordinate, scaled from η = 1 at the surface to η = 0 at the domain top.
- Atmospheric formulation: The atmospheric state evolves prognostic variables Φ, while pressure, temperature, and height are computed as diagnostic variables.Variables containing μ are identified as coupled variables.
- Time integration: The governing equations are discretized with explicit third-order Runge–Kutta time integration and finite-difference spatial operators.Physics tendencies, including the fire module, are updated only at the third Runge–Kutta step.
6 Coupling of the fire and the atmospheric models
At each atmospheric time step, WRF-Fire interpolates near-surface wind to the fire mesh, advances fire state and fuel, and returns heat fluxes to the atmosphere. Fire fluxes enter atmospheric equations as vertically decaying forcing terms.
- Spatial interpolation: Terrain gradients are computed at the best available terrain resolution and interpolated to the fire mesh during preprocessing.Computing gradients after interpolation could create jumps that affect fire propagation.
- Fire-to-atmosphere coupling: Wind is interpolated to 6.1 m above terrain using a logarithmic profile, then bilinearly interpolated to fire-subgrid cell centers.
- Fire-to-atmosphere coupling: The fire step updates active ignitions, advances the level set, records new ignition times, updates fuel fraction, and computes sensible and latent heat flux densities.
- Atmospheric forcing: Heat flux densities are averaged from fire cells into atmospheric cells before being inserted into the atmospheric model.
- Atmospheric forcing: Sensible and latent fire heat fluxes are inserted as atmospheric forcing terms in a layer above the surface with exponential decay with altitude.The scheme is used because WRF does not support flux boundary conditions.
7 Software structure
WRF-Fire uses WRF’s patch-and-tile parallel infrastructure and separates WRF-dependent coupling code from the fire-model implementation. This organization supports distributed execution while isolating most fire code from WRF internals.
- Parallel execution: WRF divides the horizontal domain into patches executed by MPI processes and tiles executed by OpenMP threads.Fire-grid tiles are colocated with atmospheric-grid tiles, with inter-patch communication handled through HALO calls.
- Software layers: The driver layer handles tile-level data exchange, while interpolation, grid coupling, and fire code remain tile-callable.
- Software layers: Only the driver and atmospheric-physics layers depend on WRF; the remaining fire-module layers are isolated from WRF.
8 Recommended WRF settings
WRF-Fire supports ideal and real runs, but its fine fire-simulation grids require settings that differ from ordinary weather applications. Recommended configuration addresses surface fluxes, mesh refinement, timestep stability, and high-resolution terrain.
- Run modes: WRF-Fire can run in ideal or real mode, with real cases additionally requiring wrfbdy boundary-condition files.Both modes use wrfinput initialization files; real.exe creates the required initialization and boundary files.
- LES configuration: Domains finer than 100 m should use LES rather than a boundary-layer parameterization.The supplied guidance recommends disabling boundary-layer parameterization and using surface-model fluxes for LES.
- Mesh resolution: The fire mesh should be about 10 times finer than the atmospheric mesh to release heat gradually near the moving fireline.Refinement factors must be selected before real.exe because they determine fire-variable dimensions and input-file structure.
- Mesh resolution: An atmospheric mesh step of about 60 m or less is recommended for proper wind feedback to the fireline.Larger steps can produce incorrect fire spread.
- Numerical stability: Fine fire domains generally require substantially smaller timesteps, especially with high-resolution topography or steep terrain.Strong vertical velocities can violate the CFL condition, limiting numerical stability.
9 Data input
WRF-Fire accepts idealized experiments and real-event simulations through WRF preprocessing workflows, with fire-specific fuel and topography data represented on refined subgrids. Real cases use atmospheric initialization and boundary data, while high-resolution inputs often require user preparation.
- Run types: Ideal runs use artificial data and an ideal.exe executable, while real runs support prediction and analysis of natural events.Real runs require user-supplied atmospheric initial and boundary conditions.
- Required data: WRF-Fire input files contain meteorological, topographical, and fire-related information such as fuel maps and high-resolution topography.Real cases use wrfinput and wrfbdy files generated through the WRF preprocessing workflow.
- Preprocessing workflow: WPS prepares atmospheric and static data through geogrid, ungrib, and metgrid before real.exe creates WRF initial and boundary files.Geogrid handles static fields, ungrib extracts GRIB data, and metgrid interpolates atmospheric data onto the computational mesh.
- High-resolution data: WRF-Fire extends WPS for refined surface meshes, but users must download and convert high-resolution fields for each simulation.The distributed convert_geotiff.x utility converts GeoTIFF data into geogrid binary format.
- High-resolution data: Fuel behavior categories are the only strictly additional dataset required for a WRF-Fire simulation.Fuel and topography variables can be defined on the refined fire subgrid, unlike standard atmospheric fields.
- Atmospheric initialization: Atmospheric datasets are often limited to about 10 km resolution, motivating nested grids with 3-to-1 refinement and long spin-up before ignition.The stated goal is to reconstruct local atmospheric conditions near the fire.
10 Computational simulations
Computational tests include verification of level-set fire-shape advection, idealized fingering behavior, and simulations of observed or real fires. These cases report agreement with reference behavior, observations, or realistic coupled fire–atmosphere behavior.
- Verification: The level-set fire module advects fire shape correctly on verification examples also used for CAWFE’s tracer code.This verification was reported by Kim (2011).
- Ideal simulations: An idealized sufficiently long fireline reproduced fingering behavior with results similar to earlier Clark et al. studies.The behavior was demonstrated by Jenkins et al. (2010).
- Real-fire simulations: WRF-Fire produced good agreement with data in a simulation of the FireFlux grass-fire experiment.The FireFlux case was used to demonstrate model validity against observations.
- Real-fire simulations: A Meadow Creek simulation used six nested domains to scale from 32 km atmospheric initialization to a 10 m fire grid.The finest domain used 10-times fire-subgrid refinement, and realistic fire and atmosphere behavior was obtained.
11 Discussion
The discussion identifies limitations in WRF-Fire’s atmosphere–fire coupling, ignition and spread parameterizations, heat-flux treatment, numerical discretization, data assimilation, and supported fire types. Several issues motivate future model development and validation.
- Additional features: WRF-Fire does not yet support canopy fire, and adding smoke coupling remains under consideration.The stated development strategy is to add features after they can be verified and validated.
- Atmosphere: Rothermel’s spread model assumes wind unaffected by the fire, whereas coupled simulations feed back fire-modified winds into spread.The discussion also questions logarithmic-profile and Monin–Obukhov assumptions near strongly convective firelines.
- Atmosphere: WRF-Fire evaluates wind at a common 6.1 m height and converts it to mid-flame height using fuel-specific correction factors.This follows the classical Rothermel and Baughman–Albini approach rather than using different heights for different fuels.
- Atmosphere: Fire-generated vertical winds require short timesteps to avoid violating the vertical CFL condition.The discussion suggests an implicitly vertical NMM core as a possible alternative for strong convection.
- Fire: FireFlux spread was reasonably close after ignition, but the simulated fire reached observation towers too soon because ignition started too quickly.The Meadow Creek fire was also too fast, attributed to unresolved irregularities and laboratory-derived spread rates.
- Fire: Heat-flux calculations omit evaporation of moisture already present in fuel, although the resulting error is typically only a few percent.The discussion proposes dynamic fuel moisture and differentiated radiative and convective heat-flux treatments.
- Numerics: The level-set discretization introduces slight nonlocal fireline behavior, while fuel-fraction estimates can have significant near-fireline subgrid error.An alternative second-order method is being tested but is more computationally expensive.
- Data assimilation: Data assimilation methods remain insufficiently robust because statistical perturbations can create spurious fires that do not dissipate.A simpler prescribed-fire-history approach is being considered for initializing from a known fire perimeter.
12 Conclusions
WRF-Fire is a coupled atmosphere-fire model available for ideal and real runs, with visualization and diagnostic utilities. The authors position it for research and education while validation continues.
- WRF-Fire couples an atmosphere model with a fire model and is publicly available for ideal and real runs.
- Visualization and diagnostic utilities accompany the software.
- The model is currently intended for research and education purposes.
- Validation is in progress.