Coupled fire-atmosphere-fuel moisture-smoke online modeling with WRF-SFIRE Jan Mandel, University of Colorado Denver Adam K. Kochanski, University of Utah Sher Schranz, NOAA/CIRA Martin Vejmelka, AVAST Supported by NASA grant NNX13AH59G and NSF grant DMS-1216481 September 30, 2017 The 3rd Annual Meeting of SIAM Central States Section Colorado State University, Fort Collins, CO
Range of scales affecting fires Range of scales in WRF • Atmospheric and fire scales 1 m 10 cm Structural Fires Flames Flamelets Wildland Fires Navier-Stokes Global weather Mesoscale weather Large Eddy Simulator model model (LES) (DNS) boundary boundary boundary conditions conditions conditions
WR WRF-SF SFIRE comp mponents HRRR forecast Atmosphere!model!WRF Chemical!transport! model!WRFBChem Surface!air! temperature,! rela?ve! Heat!and! Fire! humidity, vapor! emissions! rain fluxes (smoke) Wind !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!SFIRE Fuel!moisture!model Surface!fire!spread!model Data assimilation Data assimilation Data assimilation RAWS fuel moisture stations Satellite moisture sensing VIIRS/MODIS fire detection
WRF-SFIRE origins and sources • USDA Forest Service wildfire modeling system: BEHAVE - fire properties at one point, FARSITE - surface fire spread • NCAR’s Coupled Atmosphere-Wildland Fire Enviroment (CAWFE), based on the Clark-Hall research weather code and fire propagation by tracers • The Weather Research and Forecasting model (WRF), a standard supported community weather code, free download, widely used • Level set method • In the US, government data is free: fuel maps from LANDFIRE, weather data from NOAA, satellite fire detection from NASA, high resolution terrain from USGS,…
A Brief History of WRF-SFIRE • 2004: Connection of fire model from CAWFE and WRF proposed (Patton and Coen) • 2006: Fire propagation by tracers connected to WRF and support of refined surface fire mesh (Patton, Michalakes) • 2007: Level set method • 2008: Real data (Beezley) • 2009: Distributed memory parallelism from WRF • 2010, 2011: Versions of the model included in WRF release as WRF- Fire • 2012: Integrate fuel moisture model • 2013: Coupling with chemical transport by WRF-Chem for smoke • 2013: Operational Israel National wildfire system MATASH • 2017: NCAR selects the version from WRF release as the foundation of the operational Colorado Fire Prediction System (CO-FPS)
Representation of the fire area by a level set function • The level set function is given on center nodes of the fire mesh • Interpolated linearly, parallel to the mesh lines • Fireline connects the points where the interpolated values are zero
Evolving the fireline by the level set method Level set function L Fire area: L< 0 ∂ L ∂ t = − R ∇ L Level set equation Right-hand side < 0 → Level set function goes down → fire area grows
The fire model: fuel consumption fuel ignition time Time constant of fuel: 30 sec - Grass burns quickly 1000 sec – Dead & down branches(~40% decrease in mass over 10 min)
Integrating fuel consumption over mesh cells, with submesh fire region representation
Coupling with WRF-ARW d Φ ( ) dt = R Φ • WRF-ARW is explicit in time ( ) Φ * = Φ t + Δ t • Physics packages 3 R Φ t including fire are called only in the last ( ) Φ ** = Φ t + Δ t 2 R Φ * Runge-Kutta substep ( ) Φ t + Δ t = Φ t + Δ tR Φ ** • Fire module inputs wind, outputs heat and vapor flux Runge-Kutta order 3 integration in time
The fire model is running on a finer mesh than the atmosphere model
Wind interpolation • Spread rates for different fuels depend on wind at “midflame” height given by the fuel time • Linear interpolation of wind as a function of log(height/roughness height). Exact if the wind profile is exactly logarithmic (just like piecewise linear interpolation is exact for linear functions) independently of the vertical mesh spacing wind speed • If there are no WRF nodes under 6m, mathematically equivalent to the BEHAVE wind reduction factors. • It gets tricky • The heights of the nodes are computed from the geopotential, which is a part of the solution • The geopotential varies a lot near the fire • The atmospheric and fire mesh have different resolutions • The result depends on the roughness length. midflame height • Take the roughness length from LANDUSE or fuels? roughness height 12
Structure of the coupled WRF-SFIRE code WRF : add tendencies WRF : call sfire_driver wind heat and moisture tendencies Driver : get grid variables, get flags, interpolation calls, OpenMP Atm : one tile: temperature and loops, DM halos moisture tendencies from heat fluxes Model : one time step, one tile: winds in, heat fluxes out Phys : sensible and latent heat fluxes from fuel loss, fire rate of spread Core : time step for the level set equation, compute fuel loss. Dimensionless. Util : interpolation, WRF stubs, debug I/O,… WRF : error messages, log messages, constants,…
Standalone Sfire code MAIN Model : one time step, one tile: winds in, heat fluxes out Phys : sensible and latent heat fluxes from fuel loss, fire rate of spread Core : time step for the level set equation, compute fuel loss. Dimensionless. Util : interpolation, WRF stubs, debug I/O,… Wrf_fakes : error messages, log messages, constants,…
WRF parallel infrastructure - MPI and OpenMP • Distributed memory (DM): halo exchanges between MPI grid patches : each patch patch runs in one MPI process; programmer only lists the variables to exchange halo • Shared memory (SM): OpenMP loops over tiles within the patch • Computational routines are OpenMP tile tile callable . • Fire model executes on the threads, same horizontal tiles as the multicore atmosphere model, in the same threads Example: 2 MPI processes 4 threads each The parallel infrastructure constrains the algorithms used.
Parallelism in WRF-SFIRE: a PDE solver in WRF physics layer (meant for pointwise calculations)
Summary of the model • Atmosphere modeled by a standard numerical weather prediction software (NWP) • Fire is 2D, parameterized by Rate of Spread formula (Rothermel), • The fire Rate Of Spread (ROS) is a function of • Fuel composition and fuel moisture • Slope (fire spread uphill faster) • Wind • Heat and water vapor are released by the fire into the atmosphere, the quantity decreases exponentially with time from start of burning • Much simpler and cheaper than physics-based models, faster than real time (making prediction possible) • Can capture an important range of fire behavior
Idealized LES simulation of a small-scale prescribed burn (FireFlux experiment) • FireFlux prescribed burn of 155 acres (0.63 km 2 ) prairie • Model setup: • 1 domain, 1000m x 1600m, 10m horizontal resolution • 80 vertical levels from 2-1200m AGL • Fire grid resolution – 1m MT ST FireFlux picture from Clements et al. 2008 18
FireFlux Experiment Simulation (2010) (microscale) Field experiment Craig Clements et al., 2011 Visualization by Bedrich Sousedik
Timing of the fire front passage through the towers (5m and 4.5m air temperature) 20
Wildland Fire Behavior and Risk Forecasting Coupled atmosphere-fire model can capture As of: March PI: Sher Schranz, CSU/CIRA 1, 2016 an important range of fire behavior 2013 Patch Springs Fire, UT
Fuel Moisture Model • 1st order time-lag: • In time T , E-m ( t ) decreases by 1/ e • Equilibrium E depends on the current atmosphere state in the surface layer (temperature, RH, pressure) • Assimilation of Remote Automated Weather Station (RAWS) 10h data • Trend surface (regression) to extent RAWS data to the whole domain • Extended Kalman filter on a coarse mesh • Mix T = 1h, 10h, 100h moisture at every location with proportions from actual fuel data
Fuel Moisture Nowcasting
Simulated fire area and fuel moisture for Barker Canyon fire 2012 Simulated fire area and fuel moisture in-plume concentration ~3000μg /m 3 (3mg/m 3 ) 50000 22.0% Simulated fire area 45000 20.0% Observed fire area 40000 18.0% Integrated fuel moisture simulated by the fuel moisture model 35000 16.0% Fuel moisture Fire area (ha) 30000 14.0% 25000 12.0% 20000 10.0% 15000 8.0% 10000 6.0% 5000 4.0% 0 2.0% -12 0 12 24 36 48 60 72 84 96 Time since 09.09.2012 00:00 local (h) 24
Example #1 Simulation of Barker Canyon Fire (smoke as a passive tracer) in-plume concentration ~3000μg /m 3 (3mg/m 3 ) Simulated fire perimeter Observed fire perimeter Simplified approach – no chemistry 96h simulation done in 12h 52min on 640 CPUs, with the first 24h forecast ready in 3h 13min 25
Example #1 Simulation of Barker Canyon Fire (smoke as a passive tracer) in-plume concentration ~3000μg /m 3 (3mg/m 3 ) Fuel Moisture 26
The Online System WRFXCTRL: Submit jobs WRFXWEB: Delivery to users WRFXPY: Retrieve data,run jobs, process output Online Data: NOAA, USGS, … Model: WRF-SFIRE
Run online
Delivery online
• http://www.openwfm.org/wiki/WRF-SFIRE_user_guide • https://readthedocs.org/projects/wrfxpy • https://github.com/openwfm - sources • http://demo.openwfm.org/ - cloud visualization server • http://www.openwfm.org/wiki/Publications Data driven modeling - assimilation of fire detection data from satellites: MS06 tomorrow 10:50am
Recommend
More recommend
