--- NESL, the NCAR
        Earth System Laboratory ---

The Coupled Atmosphere-Wildland Fire Environment

Model (CAWFE) for Wildland Fire Modeling

          and visualization of the 2012 High Park wildfireimage
Figure 1.  VAPOR (www.vapor.ucar.edu) visualization of CAWFE simulations of the (a) High Park fire and (b) the Troy Fire.

The Coupled Atmosphere – Wildland Fire - Environment (CAWFE) Model contains two parts: a numerical weather prediction model and a fire behavior model that simulates the growth of a wildfire in response to weather, fuel conditions, and terrain. These are two-way coupled to constantly exchange information so that heat and water vapor fluxes from the fire alter the atmospheric state, notably producing fire winds, as the evolving atmospheric state and changes in humidity (including effects from the fire) simultaneously affect fire behavior, notably how fast and in what direction the fire propagates. The model is described in Clark et al. (2004) and Coen 2005a.  Coen (2013) documents the model equations.



(1) Studies using CAWFE have shown that complex interactions between a fire and the atmosphere are behind even the most basic fire fundamentals (Coen 2011), including the elliptical shape of the fire itself and the formation of the rapidly moving, most intense burning head, flanking regions where the wind blows parallel to the interface, and backing region that creeps against the wind. In contrast to FARSITE, which forces the evolving shape of the fire to follow an ellipse, the interplay of the fire with the atmosphere in CAWFE allows this often-observed behavior, to evolve from the physics (Figure 2).


Thus, although simpler tools such as BEHAVE+ and FARSITE can in some circumstances, with daily calibration of fuel loads, produce acceptable estimates of anticipated rates of spread, much more detailed understanding/insights are now possible with coupled weather-fire models.  Because the predicted rate of spread for a given input wind is fixed for tools such as BEHAVE+ or FARSITE, it is not possible to capture these feedbacks with incremental improvements; models such as CAWFE are fundamentally different because the forces on the air created by the fire change the winds that in turn direct the direction and rate of spread of the fire.  This becomes even more important when simulating the unfolding of a large fire event, where more complex results of this coupling become apparent.


Visualization of
            fire in idealized conditions

Figure 2.  Heat produced by the fire (more intense colors are hotter), smoke (misty purple field), and surface winds (longer arrows indicate stronger winds, the arrow indicates direction).  In this simulation, a fire began as a line in winds that were all coming at 3 m/s from the left, but which created a fire with a head, flanks, and backing region, and shaped the winds in the fire vicinity to be moving rapidly forward at the fire head, parallel to the flanks, and weak in the backing region. (Animation at http://www.mmm.ucar.edu/people/coen/files/newpage_f.html ).  The winds along the flank carry small perturbations, which grow into fire whirls, along to the head of the fire where they may hook together and shoot forward in bursts.


(2)  Numerical weather prediction models represent the three-dimensional wind structure in mountainous terrain.  These flows depend on many simultaneous factors, including the steepness, height, and width of terrain, the wind speed and direction and how this changes with height, the temperature profile in the air (the atmospheric stability), heat released by the formation or evaporation of precipitation and clouds, the solar heating of slopes, upwind terrain features, and the roughness of the surface.  The past 30 yrs of atmospheric science research has established that no simpler model than a full weather model can consider all these factors together and predict which of many types of flow might occur at one particular time and location. In mountainous terrain, weather information from a weather station is extremely unlikely to be representative of the winds driving the fire.


A striking example of the limitations of current approaches is the 2007 Esperanza Fire in Riverside County, California, which ignited on the upwind edge of the San Jacinto mountains during dry, windy Santa Ana conditions. CAWFE simulations show the simulated weather conditions, fire growth, and smoke production and transport. This work is the first to simulate simultaneously the evolving meteorological flow, fire behavior, and fire-induced flow for a landscape-scale naturally evolving fire. The simulation captures how strong upper-altitude winds from the east-northeast were brought down to the surface in terrain-generated atmospheric waves, driving the fire and smoke to the west-southwest.  (Figure 3). Local surface weather stations (RAWS) were located in Banning Pass and caused FARSITE to predict a slow spread due west.


Other features characteristic of this fire are brought out in the CAWFE simulation - the splitting of the fire, the fire drawing itself up canyons along the flanks, the most intense burning often being along the flanks rather than at the leading edge (the "head") of the fire.  These can only be reproduced in models that allow the fire to modify the winds.  For infrared imagery of this fire from U.S.D.A. Forest Service research aircraft, see  http://fireimaging.com/ .




smoke heat flux
            from simulated Esperanza fire

Figure 3.  View: towards the south.   Cabazon, CA, is in the foreground.  Animations are at http://www.mmm.ucar.edu/people/coen/files/newpage_m.html . The misty field represents smoke, colored by concentration - higher concentrations are more opaque (linearly with concentration) and darker.  The colors identifying the burning parts of the fire are inspired by the radiant temperature color bar at fireimaging.com - brighter colors like yellow reflect higher surface fire sensible heat fluxes.  Darker browns are lower fluxes.  The other field on the surface is the 'fuel load remaining' - where the fire has passed, the surface appears dark brown. (The spots of dark brown ahead of the fire reflect places where the fuel was light grass and the load was small, not burned out.) The boxiness to the fireline shows the atmospheric grid sizes, onto which the fire fluxes on the fire fuel cell scale (5x5 within each atmospheric cell) have been summed. 



vectors from
            simulated Esperanza fire

Figure 4. Similar to Figure 3, but the arrows show the winds at 1500 m above sea level, near the surface.  The length and direction of the vectors show the strength and direction of the horizontal winds.  The colors superimposed on the arrows represent the vertical velocity: white is 0, warm colors (yellow to orange to red) reveal air traveling up, cold colors (green increasing to blue increasing to violet) represent air traveling downward.   Over the fire area, the arrows are greenish, which is interpreted as a slight downdraft at this elevation.  Some areas are mostly white, which means there is no up or down component there.  A few vectors near lower right and sometimes over Cabazon Peak (center left) show upward motion.  The location and strength of the upward and downward air motions over the fire vary with time. These waves are set off by being forced over the San Bernardino mountain range upwind, and then, encountering the San Jacintos, are complicated by smaller-scale terrain features like Cabazon Peak, shear-generated motions, and convection generated by the fire.


(3)   Just how strong is the effect of the fire on its surrounding environment and why is it important?


The modification of the winds by the fire is the cause of virtually all phenomena that create the individual character of large event -  splitting of fire fronts, draws of flanks up canyon perpendicular to overall fire spread, how fires drawn themselves together (ex.: deliberately set fires may either be drawn into large wildfires or turn into wildfires themselves), and in the extreme, the generation of fire whirls and blowups, where the firestorm-like connection/grip/bond between the increasing fire intensity and the atmosphere tightens, such that the ‘fire creates its own weather’.  


This simulation shows several hours in the early period of the Big Elk Fire (Figure 5), a 2200 ha Colorado wildfire. Fire behavior was extreme reflecting the extremely dry conditions throughout Colorado (including the lowest fuel moistures ever recorded in the area). Initial spread was rapid, moving up a south slope of ponderosa pine mixed with Douglas fir with crowning and torching into high density thin-stemmed lodgepole pine at upper elevations.  This case represents a relatively simple scenario, with no large-scale weather features - winds were driven primarily by solar heating of mountain slopes, producing weak afternoon upslope conditions during the active fire periods, and by the fire-induced winds themselves. 


Big Elk fire

Figure 5. The orange lines outline topography contours, increasing toward Kenney Mountain, in the center.  The red field shows where the air was warmed at least 10 degrees by the heat released from the fire.  The misty white field represent smoke, with denser areas representing higher concentrations.  The wind speed is shown by the length of the arrows (longer arrows indicate stronger winds) and direction near the surface.  


The magnitude of the effect of the fire on the winds is shown in Figure 6.

difference between
            simulation with and without fire feedbacks to atmosphere

Figure 6. The effect of the fire on the winds, shown by the difference between a simulation and one in which the fire could not modify the winds. The perspective is the same as Figure 5, but looking straight down. The strength and direction of the fire’s change of the wind is shown both in the arrows and in the contours.  The effect of the fire-atmosphere interactions is that the fire draws air into the base of the plume at its leading edge, creating strong winds over the leading edge of the fire that, as in Figure 2, increase the spread of the fire. 


Two important points: (1) The magnitude of the fire’s effects are to change the winds by 10-12 m s-1 near the fire, but they may also changes the winds in the fire’s environment by several m s-1 even 5 km from the fire.  This is how one fire may alter the winds affecting a nearby fire.  (2) The fire’s effects on the winds make them unsteady, which may cause wind shifts that are a further hazard.  Again, these effects are only seen in models that capture the fire’s ability to alter the winds in its environment.



·      CAWFE has been used on case studies of large wildfires (e.g. the Big Elk Fire in Colorado (Coen 2005a); the Esperanza Fire (Coen and Riggan 2010, 2013), the 2012 High Park Fire, the Simi Fire (CA), Troy Fire (CA), Spade Fire (MT), Hayman Fire (CO)) to aid understanding of these complicated events. See the examples.

·      CAWFE has also been tested as a real-time forecasting tool (Coen 2005b) and can currently be configured to run in real time on a single processor of a desktop computer.  Many possible usage scenarios exist: strategic vs. tactical, wildland fire use planning, anticipating prescribed fire burning windows, testing “What-If ?” fuel mitigation scenarios) should be tested and examined has not yet occurred; the current capabilities are appropriate for testing and feedback in conjunction with a practical setting.  

·      CAWFE can be run with the initialized in progress with perimeters from infrared airborne fire mapper or satellite fire detection data (Coen and Schroeder 2013).



This material is based upon work supported by the National Science Foundation under Grants No. 0324910, 0421498, and 0835598, the National Aeronautics and Space Administration under Awards NNX12AQ87G and NNH11AS03, and the Federal Emergency Management Agency under Award EMW-2011-FP-01124. The National Center for Atmospheric Research is sponsored by the National Science Foundation. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. No endorsement by the U.S. Department of Agriculture is implied.  The model was developed with support from the USDA Forest Service Riverside Fire Laboratory and Missoula Fire Laboratory and contributions from individuals at NCAR, the USDA Forest Service Riverside Fire Laboratory (now Pacific Southwest Research Station), and The Australian Bureau of Meteorology.

Microsoft Word - CS_090313_upl Microsoft Word - CS_090313_upload_WS.d


Clark, T. L., J. Coen, D. Latham: Description of a coupled atmosphere-fire model, 2004: Intl. J. Wildland Fire, 13, 49–64.

Coen, J. L., 2013: Modeling Wildland Fires: A Description of the Coupled Atmosphere-Wildland Fire Environment Model (CAWFE). NCAR Technical Note NCAR/TN-500+STR. 38 pp. http://nldr.library.ucar.edu/repository/collections/TECH-NOTE-000-000-000-866.

Coen, J. L., 2011. Some new basics of fire behavior.  Fire Management Today.  71(1), 37-42.  

Coen, J. L. 2005a Simulation of the Big Elk Fire using coupled atmosphere-fire modeling. Intl. J. Wildland Fire, 14, 49–59.

Coen, J. L., 2005b, Applications of coupled atmosphere-fire modeling: Prototype demonstration of real-time modeling of fire behavior. Amer. Meteor. Soc. Joint 6th Symp. on Fire & Forest Meteor./Interior West Fire Council Conf. 25-27 October. Canmore, AB, Canada.  CD-ROM, Paper 8.1

Coen, J. L. and P. J. Riggan, 2010: A landscape-scale wildland fire study using a coupled weather-wildland fire model and airborne remote sensing. Proceedings of 3rd Fire Behavior and Fuels Conference, October 25-29, 2010, Spokane, Washington, USA. Published by the International Association of Wildland Fire, Birmingham, Alabama, USA.  CD-ROM.  12 pp.    

Coen, J. L. and P. J. Riggan: Simulation and thermal imaging of the 2006 Esperanza wildfire in southern California: Application of a coupled weather-wildland fire model. Intl. J. Wildland Fire. In review.

Coen, J. L. and W. Schroeder, 2013: Use of spatially refined remote sensing fire detection data to initialize and evaluate coupled weather-wildfire growth model simulations. Geophys. Res. Lett. 40:1-6. (doi:10.1002/2013GL057868)

Keywords:  wildfire model, fire behavior, forest fire, fire model, wildland fire model