review of urban modeling program at llnl
DESCRIPTION
Review of Urban Modeling Program at LLNL. CRTI-02-0093RD Project Review Meeting Canadian Meteorological Centre August 22-23, 2006. FEM3MP – An Urban Dispersion Model. Massively parallelized CFD model based on solving 3D time- dependent Navier-Stokes equations for large-scale problems - PowerPoint PPT PresentationTRANSCRIPT
Review of Urban Modeling Program at LLNL
CRTI-02-0093RD Project Review Meeting
Canadian Meteorological Centre
August 22-23, 2006
FEM3MP – An Urban Dispersion Model
• Massively parallelized CFD model based on solving 3D time- dependent Navier-Stokes equations for large-scale problems
• Finite element method for effective treatment of terrain, complex geometries and flows
• Simple and advanced turbulence closures
• Sub-models for canopies, aerosols, UV radiation decay, surface heating, etc.
• Validated against data from wind tunnel and urban field experiments
Governing Equations
• Plus Smagorinsky SGS turbulence model & Plus Smagorinsky SGS turbulence model & wall damping function by Piomelli, et al. (1987)wall damping function by Piomelli, et al. (1987)
)(
)(
0
cuu
u
uuuup
uu
jjxjxc
jtc
jjx
idjijxixjxiu
jit C
FEM3MP Applications
• Model flow and dispersion in urban areas• Perform simulations to optimize utilization of resources
in the design of field experiments• Generate realistic scenarios to support emergency
planners in planning of special events• Use model results to provide improved parameterization
in larger scale models• Source inversion for contaminant plume dispersion in
urban areas
AUDIM – LLNL’s Next-generation Urban Dispersion Modeling Capability
AUDIM
Adaptive Urban Dispersion Integrated Model
Adaptive mesh refinement for enhanced fidelity: release points, building entrances, etc.
Complex release scenarios: moving sources, etc.
CFD code for urban dispersion
FEM3MPParallel adaptive
mesh support
SAMRAIRapid geometry to
mesh capability
Overture
Automatic mesh construction from building datasets.
Geometrically complex buildings and cityscapes
Diverse urban environments: stadiums, arenas, subways, etc.
Immersed Boundary Formulation
• Ghost-cell method of Tseng and Ferziger (2003)– set values at “ghost points” inside boundary using interpolation from
outside neighbors
– interpolation to enforce conditions at boundary– Conditions applied: u = 0, dp/dn = 0
Immersed boundary
Ghost point
Nearest neighbors
Enforce zero velocities on the immersed boundary
normalBCs applied at boundary point
closest to ghost point
T. Chow
,)()1( uuzacft
udroof
,)()2( vvzacft
vdroof
,)()3( wwzacft
wdroof
Momentum Equations:
(froof: roof fraction, cd: urban drag coef., a(z): roof surface area density profile)
Urban Canopy Parameterization (UCP)
TKE Equation:
),()()(
)4(333wvuzacf
t
TKEdroof
Potential Temperature Equation:
zRff
Bz
qf
zRf
c ptNc
roofurburb
urbN
urb )[()1
1()1{(1
)5(1
]},)(C
qczbf
roof
roofproof
Roof Surface Energy Equation:
).()()1()6(_
4 TTVccTRRqroof roofroofdproofLWSW
Street Canyon
Roof-Top
Anthropogenic
furb= froof+ fcnyn
(Chin et al., 2005, MWR)
Key urban surface and building infrastructure parameters of UCP are derived from USGS land-use data using a table conversion method.
urban thermal properties
anthropogenic heating
canopy heating & cooling
drag
radiation attenuation
turbulence production
radiation trapping
Seamless Coupling Between Regional and Urban Scale Models
Urban scale models resolve small scale flows which must be parameterized in large scale models – considerable current scientific interest
Downtown
REGIONAL SCALE4km grid size
URBAN SCALE
1m grid size
Coupling will provide accurate boundary conditions for urban scale simulations
MESOSCALE
Possible release locations are identified to within a ~25m x 150m area including the actual source
Inflow wind Sensors ( )
Markov chain sampling
Possible source locations
Actual sourcelocation
Histogram shows simultaneous determination of release rate to within 10% of actual value
Actual release rate
Computational approach uses Green’s function methodology• 2560 pre-computed unit source simulations • Total CPU = 13,056hrs (12+ hrs on 1024 2.4 GHz Xeon processors)• Event reconstruction requires ~2 minutes (20000 Markov iterations)
Event Reconstruction - Computational framework will support multiple stochastic algorithms, models, and platforms
Output Handler
Input Handler
MCMC SMC
HYBRIDMULTI-RES.
Informed prior and proposal sampling
with nonlinearoptimization
Job Distributor
MODEL DRIVER
Model Handler
Input Handler
Output Handler
Urban Puff Model
Output Handler
Input Handler
3D Particle Model
Output Handler
Input Handler
2D Puff Model Urban CFD Model
STOCHASTIC TOOLS
...
SYSTEM HARDWARE
PC workstation
Massively parallel system