Wildfires
Chun-Lung Lim and Charles Erwin
THE WEATHERTHE WEATHERRESEARCH ANDRESEARCH AND
FORECAST MODELFORECAST MODEL(WRF)(WRF)
What makes WRF next-generationWhat makes WRF next-generationmesoscalemesoscale forecast model ? forecast model ?
Advance the understanding and the prediction ofmesoscale precipitation systems and to promotecloser ties between the research and operationalforecasting communities.Efficiency, portability, maintainability, andextensibility.
SOFTWARESOFTWAREARCHITECTUREARCHITECTURE
Runge-Kutta MethodRunge-Kutta Method
Driver LayerDriver Layer
Handles run-time allocation and parallel decompositionof model domain data structures.Organization, management, interaction, and control overnested domains, including the main time loop in themodel.High level interfaces to I/O operations on modeldomains.It is interface to other components when WRF is part ofa larger coupled system of applications.
Mediation LayerMediation Layer
Encompasses one time-step of a particulardynamical core on a single model domain.The current WRF implementation uses theMessage Passing Interface (MPI) communicationpackage.Shared-memory parallelism over tiles in the solveroutines is using OpenMP.
ICCICC’’s High Performance Computings High Performance Computing
Model LayerModel Layer
Comprises the actual computational routines that makeup the model: advection, diffusion, physicalparameterizations, and so forth.The subroutines are called through a standard ModelLayer Interface.The interface ensures that a Model Layer packageincorporated into WRF will work on any parallelcomputer.Model layer routines have data dependencies rely on themediation layer to perform the necessary interprocessorcommunication.
The RegistryThe Registry
It is a concise database of information about WRF datastructures and a mechanism for automatically generatinglarge sections of WRF code.The Registry data base is a collection of tables that listsand describes the WRF state variables and arrays withtheir attributes such as dimensionality and so fourth.Registry generates code for interfaces between layers ofthe infrastructure, packing and unpacking code forcommunication and nesting, and field-by-field calls toroutines for model I/O.
Moving NestsMoving Nests
PERFORMANCEPERFORMANCE
Implementation of Implementation of WildlandWildlandFire Model Component inFire Model Component in
the WRF modelthe WRF model
Functionalities of the Functionalities of the WildlandWildland Fire Model Fire ModelComponentComponent
Extension of the Clark-Hall atmospheric model.Atmosphere-wildland fire simulation model has been developed torepresent a complex interactions between fires and local winds.Helps to track atmospheric wind velocities and calculate firespread more precisely.Handles large releases of buoyancy and accurately represent fine-scale motions in complex terrain.Ingest large-scale gridded data to incorporate a changingmesoscale atmospheric environment.Telescope down to the meter-sized fine dynamic scales of vorticesin the fire line through horizontal and vertical grid refinement.
BenefitsBenefits
The availability of a coupled atmosphere-fire model in a well-supported community model for eventual community research.Can build a stable framework in other scientific componentsrelated to the Wildland Fire Research and DevelopmentCollaboratory.Provide eventual operational capability with operationalapplications like smoke management …A test of WRF with strong forcing at small scales.Reduction in the redundancy of effort developing neededcapabilities within the Clark-Hall model .
Numerical modelNumerical model
• Background- 3-D nonhydrostatic atmospheric prediction model coupled with
an empirical fire spread model with sensible and latent heatflux from the fire feed back to the atmosphere to produce firewinds.
- The atmospheric winds drive the fire propagation.
• Wildfire simulation model represents the complexinteractions between a fire and local wind.
BURNUP algorithmBURNUP algorithm
Characterizes how the fire consumes fuels ofdifferent sizes over time.
Equation : 1-F = exp(-t/W)
Implementation of Fire Module in WRFImplementation of Fire Module in WRF
Fire module will be implemented as an added physicsoption in WRF.Fire-atmosphere coupling will occur through passingwinds from the lowest WRF level to the fire module.The fire module will use those winds to predict the firespread and subsequent heat and water vapor emissions.Heat and water vapor emissions from the fire will bepassed back to WRF and distributed vertically throughan assumed extinction depth.
Initialization of Fire EnvironmentInitialization of Fire Environment
Three environmental factors that influence firebehavior are :
1. Topography2. Weather3. Fuel
TopographyTopography
Fires spread much faster in upslope than flatground.Fine-scale topography features as a factor in localairflows plays a role in fire behavior.
WeatherWeather
Weather impacting the fire.The weather impacts on the fire can be obtainedby the winds model in WRF.Winds model could be used to simulate changesin dead fuel moisture which responds with timelags corresponding to the size of the fuelparticles.
- A second feedback loop between the fire and theenvironment.
FuelFuel
Not all vegetation is burnable.Live vegetation may be dried and ignited by fire(live fuel).Forest floor needles, cured grasses and branches(dead fuel) play more important role in firebehavior.
Links with other modulesLinks with other modules
WRF-Chem is an atmospheric chemistrypackage linked to WRF for simulation ofatmospheric chemistry and aerosols.
The combined application of WRF-Chem andWRF-Fire will allow the user to create asimulation of Wildland fire.
Numerical Simulation of WildfiresNumerical Simulation of Wildfires
ReferencesReferences
URLs
http://mathworld.wolfram.com/Runge-KuttaMethod.html http://www.wrf-model.org/index.php http://www.llnl.gov/computing/hpc/training/#training_materials http://box.mmm.ucar.edu/research/wildfire/wrf/wrf_summary.html http://box.mmm.ucar.edu/research/wildfire/wrf/wrf_fire.html
Research Papers
The Weather Research and Forecast Model: Software Architecture and Performanceby J Michalakes, J.Dudhia, D.Gill, T.Henderson, J.Klemp, W.Skamarock, W.Wang
Implementation of Wildland Fire Model Component in the Weather Research and Forecasting(WRF) modelby Janice Coen (MMM/RAP) and Ned Patton (MMM)
WRF-Fire: A Coupled Atmosphere-Fire Module for WRFby Ediward G.Patton and Janice L.Coen.National Center for Atmospheric Research, Boulder, CO
Wildfire Simulation SoftwareWildfire Simulation Software
Charles ErwinCharles Erwin
CS 521: Computational Science 2
Simple Wildfire Simulator from NOVASimple Wildfire Simulator from NOVA
http://www.pbs.org/wgbh/nova/fire/simulation.html(requires Flash)
Wildfire Simulator is a simple computer simulation thatpredicts the behavior of fire in a wildland environment.
Not meant for research, only to demonstrate some basicideas about wildfire simulation.
Programming for this feature is derived from FARSITE.
CS 521: Computational Science 3
EMBYR: EMBYR: ““Ecological Model for BurningEcological Model for Burningthe Yellowstone Regionthe Yellowstone Region””
Created by William W. Hargrove and Robert H. Gardner Designed to simulate wildfires, the subsequent pattern of
vegetation, and then the next generation of burnpatterns.
While the EMBYR model parameters could be adjustedto reproduce a particular historical wildfire exactly, it ismore important to reproduce any wildfire relatively wellon average.
EMBYR can generate "Risk Maps", which areconstructed from many replications of a single simulatedfire. Cells which burned in many of the replications arecolored black, while cells which burned in only a fewsimulations are colored white, with gray levels inintermediate cases.
CS 521: Computational Science 4
EMBYR Fire ModelEMBYR Fire Model
The fire model, EMBYR, depicts the landscape as a grid inwhich the dimension of each cell is 50 m (2500 m2).
Diffusive Spread: Fire spreads from each ignited cell toany of eight unburned neighbors (the four adjacent cellsand four diagonal cells) as an independent stochastic eventwith probability I, where I may range from 0 to 1.
Each cell burns for a single time step of variable length, andthe fire goes out if new sites are not ignited at each timestep.
Theoretical studies have demonstrated that if I is less thena critical value, fires are unlikely to propagate across thelandscape c
i
CS 521: Computational Science 5
EMBYR Fire Model (cont)EMBYR Fire Model (cont)
They estimated by performing 50 simulations for eachvalue of I (0.245 I 0.252 in increments of 0.001) ona 300x300 grid.
The proportion of simulations with fires reaching the topedge of the map after the entire bottom edge was ignitedwas 38% for I = 0.250 and 60% for I = 0.251. Since isthe threshold at which 50% of the fires reach theopposite edge of the map, these results indicated that
lies between 0.250 and 0.251.
!!ci
ci
ci
CS 521: Computational Science 6
EMBYR Fire Model (cont)EMBYR Fire Model (cont)
Simulating multiple fuel classes: EMBYR explicitlysimulates multiple classes of fuel by varying theprobability of fire spread as a function of fuel type.The fuel classes considered are four successional
stages of lodgepole pine forest, nonforested regionssuch as meadows, and nonflammable areas such asrock, roads, and water.
Derived probabilities on fire spreading betweendifferent types of fuel.
CS 521: Computational Science 7
EMBYR Fire Model (cont)EMBYR Fire Model (cont)
Variation in fuel moisture: EMBYR uses a standard firedanger measure known as percent 1000-h time-laggedfuel moisture. In this measure, an assumption is made about how
long fuel of a particular diameter would take to soak tothe core, or to dry out once soaked.
Current internal moisture in fuels of that diameter ismodeled with appropriately time-lagged ambientatmospheric humidity.
Obviously, if fuels are sufficiently wet, fires do notoccur.
CS 521: Computational Science 8
EMBYR Fire Model (cont)EMBYR Fire Model (cont)
Simulating the effects of wind: Three classes of windspeeds (WS), measured at a standard height of 6.1 m(20 ft) above the surface, are considered: WS 0, with speeds ranging from 0 to 3.1 kph (5 mph) WS 1, moderate winds ranging from 3.1 to 21.7 kph (5–35 mph) WS 2, strong wind with speeds greater than 21.7 kph
For each of the three wind speed classes, a bias value bis used to modify the probability of spread to eachneighboring cell.
CS 521: Computational Science 9
EMBYR Fire Model (cont)EMBYR Fire Model (cont)
CS 521: Computational Science 10
EMBYR Fire Model (cont)EMBYR Fire Model (cont)
Simulating the effects of firebrands: EMBYRsimulates a second mechanism of fire spread — theproduction of firebrands which are carried aloft in therising convection column, and then drift and fall onremote sites.
The ‘spotting’ effect of firebrands is simulated bypermitting each burning site to generate a fixed numberof firebrands as a function of fuel type.
CS 521: Computational Science 11
Simulation: homogeneous landscapesSimulation: homogeneous landscapes
Area burned (in cells) in a 500x500 cell homogeneous fuel class landscape with asingle fixed ignition as a function of the probability of fire spread, I, to the eightsurrounding neighbors where (a) fire is allowed to propagate by adjacent spreadonly (no firebrands), and (b) fire is allowed to propagate by adjacent spread and byfirebrands. The simulation was ended before fire could reach the edge of the map.Means and standard deviations are shown for five replications.
CS 521: Computational Science 12
Simulation: Using actual LandscapesSimulation: Using actual Landscapes
CS 521: Computational Science 13
Simulation: Using actual LandscapesSimulation: Using actual Landscapes
The cumulativefrequency of risk of firesof increasing size forfour alternative weatherconditions of (from leftto right) (a)Scenario 1: ‘moist’ with‘strong’ winds; (b)Scenario 2: ‘dry’weather with ‘moderate’winds; (c) Scenario 3:‘very dry’ weather with‘moderate’ winds; and(d) Scenario 4: ‘very dry’weather with ‘strong’winds
CS 521: Computational Science 14
Examples of EMBYR In actionExamples of EMBYR In action
CS 521: Computational Science 15
FARSITE: FARSITE: ““Fire Area SimulatorFire Area Simulator”” Two Dimensional model of fire behaviour and growth
simulator. A simple ellipse fit observed fire growth data as well as
other shapes. Regardless of the correct shape (if asingle one exists), the eccentricity of the fire is known toincrease with increasing windspeed or slope steepnessor both.
Cellular Model: Simulate fire growth as a discreteprocess of ignitions across a regularly spaced landscapegrid.
In general, cellular models have had diminishing successin reproducing the expected twodimensional shapes andgrowth patterns as environmental conditions becomemore heterogeneous.
CS 521: Computational Science 16
FARSITE ModelFARSITE Model
Problems with cellular models are avoided by the vectoror wave approach to fire growth modeling (Huygens’principle).
The fire front is propagated as a continuously expandingfire polygon at specified timesteps.
Essentially the inverse of the cellular method, the firepolygon is defined by a series of two-dimensionalvertices (points with X,Y coordinates). The number ofvertices increases as the fire grows over time (polygonexpands). The expansion of the fire polygon isdetermined by computing the spread rate and directionfrom each vertex and multiplying by the duration of thetimestep.
CS 521: Computational Science 17
FARSITE Model (cont)FARSITE Model (cont)
The reliance on an assumed fire shape, in this case anellipse, is necessary because the spread rate of only theheading portion of a fire is predicted by the present firespread model. Fire spread in all other directions isinferred from the forward spread rate using themathematical properties of the ellipse.
There are still many problems in accurately simulatingfire with this approach, different methods, however, willprobably be of little consequence to the practicalapplication of a fire growth model until the greateruncertainties are resolved as to how wind, slope, andfuels affect fire shapes.
CS 521: Computational Science 18
FARSITE Model (cont)FARSITE Model (cont)
CS 521: Computational Science 19
FARSITE Model: RichardsFARSITE Model: Richards’’ Equations Equations
The shape of an elliptical firedetermined from the conditionslocal to that vertex in terms ofdimensions.
a, b, c
The direction of maximum firespread rate.
The orientation of the vertex onthe fire front in terms ofcomponent differentials.
Xs, Ys
!
CS 521: Computational Science 20
FARSITE Model (cont)FARSITE Model (cont)
CS 521: Computational Science 21
FARSITE: Transformations for SlopingFARSITE: Transformations for SlopingTerrainTerrain
Richards’ equations were originally developed for flatterrain.
On flat terrain, a horizontal coordinate system remainsunchanged when projected onto the ground surface.This is not the case for sloping terrain.
This means that the inputs to equations [1] and [2] mustbe transformed from the horizontal to the surface plane,and outputs must be transformed from the surface planeback to the horizontal plane.
CS 521: Computational Science 22
FARSITE Model (cont)FARSITE Model (cont)
CS 521: Computational Science 23
FARSITE Model (cont)FARSITE Model (cont)
CS 521: Computational Science 24
FARSITE Model (cont)FARSITE Model (cont)
CS 521: Computational Science 25
FARSITE Model (cont)FARSITE Model (cont)
Other models used include the Van Wagner crown firemodel, and Albini’s spotting model.
For input, FARSITE uses GIS raster data in lieu of vectordata. For fuel moisture, BEHAVE and NFDRS equationsare used.
CS 521: Computational Science 26
FARSITEFARSITE
Raster Landscape inputlayers required from theGIS for FARSITEsimulation.
CS 521: Computational Science 27
FARSITEFARSITE
Animation of a FARSITEv4.0x simulation in a 3Dwindow.
CS 521: Computational Science 28
FARSITEFARSITE
Screen shot of aFARSITE v4.00simulationutilizing the post-frontalcombustionmodel.
CS 521: Computational Science 29
ReferencesReferences
http://www.pbs.org/wgbh/nova/fire/simulation.html http://fire.org Finney, Mark A. 1998. FARSITE: Fire Area Simulator-model
development and evaluation. Res. Pap. RMRS-RP-4, Ogden, UT:U.S. Department of Agriculture, Forest Service, Rocky MountainResearch Station
http://research.esd.ornl.gov/~hnw/embyr/ Hargrove, W.W., R.H. Gardner, M.G. Turner, W.H. Romme, and
D.G. Despain. 2000. Simulating fire patterns in heterogeneouslandscapes. Ecological Modelling 135(2-3):243-263