Why WRF?• An attempt to create a national mesoscale prediction
system to be used by both operational and research communities.
• A new, state-of-the-art model that has good conservation characteristics (e.g., conservation of mass) and good numerics (so not too much numerical diffusion)
• A model that could parallelize well on many processors and easy to modify.
• Plug-compatible physics to foster improvements in model physics.
• Designed for grid spacings of 1-10 kmeta
Two WRF Cores• ARW (Advanced Research WRF) (aka Mass Core)developed at
NCAR• Non-hydrostatic Numerical Model (NMM) Core developed at
NCEP• Both work under the WRF IO Infrastructure
NMM ARW
The NCAR ARW Core Model:(See: www.wrf-model.org)
Terrain following hydrostatic mass (p) vertical coordinate, arbitrary vertical resolution
Arakawa C-grid, two-way nesting, any ratio 3rd order Runge-Kutta time-split differencing Conserves mass, entropy and scalars using up to
6th order spatial differencing equ for fluxes (5th order upwind diff. is default)
NCAR physics package (converted from MM5 and Eta), NOAH unified land-surface model, NCEP physics adapted too
The NCEP Nonhydrostatic Mesoscale Model: NMM (Janjic et al. 2001)
Hybrid sigmapressure vertical coord. Arakawa E-grid, 3:1 nesting ratio Adams-Bashforth time differencing, time splitting Conserves kinetic energy, enstrophy and
momentum using 2nd order differencing equation Separate set of equations for hydrostatic versus
non-hydrostatic terms Modified Eta physics, Noah unified land-surface
model, NCAR physics adapted too Parallelized within WRF infrastructure
WRF Modeling System
Obs Data,Analyses
Post Processors,Verification
WRF Software Infrastructure
Dynamic Cores
Mass Core
NMM Core…
Standard Physics Interface
Physics Packages
StaticInitialization
3DVAR DataAssimilation
WRF Hierarchical Software Architecture• Top-level “Driver” layer
– Isolates computer architecture concerns– Manages execution over multiple nested domains– Provides top level control over parallelism
• patch-decomposition• inter-processor communication• shared-memory parallelism
– Controls Input/Output
• “Mediation” Layer– Specific calls to parallel mechanisms
• Low-Level “Model” layer – Performs actual model computations– Tile-callable– Scientists insulated from parallelism– General, fully reusable
Mediation Layer
wrf
initial_config alloc_and_configure init_domain integrate
solve_interface
solve
Model Layer
Driver Layer
prep
filt
er
big_
step
deco
uple
adva
nce
uv
reco
uple
scal
ars
phys
ics
adva
nce
w
The National Weather Service dropped Eta ( old NAM-North American mesoscale run) in June and replace by WRF NMM (new NAM).
The Air Force is now switching from MM5 to WRF ARW.
Most universities using WRF ARW
On June 13, 2006 starting with the 12 UTC model run, NCEP will replace the forecast model used in its North American Mesoscale (NAM) time slot. Currently the the Eta forecast model is used for the NAM, but on this date it will be replaced with the Non-hydrostatic Mesoscale Model (NMM) in the WRF framework
The WRF/NMM with continue to run over the same domain and same horizontal resolution (12 km) as the Eta and its output will be available at the same time. Specifics on the differences between the Eta and WRF/NMM systems are as follows:
1. Model Changes
- Replace Eta prediction model with WRF version of the Non-hydrostatic Meso Model (WRF-NMM) - Extended model top pressure from 25 mb to 2 mb - Replace Eta step-mountain vertical coordinate with NMM hybrid sigma-pressure vertical coordinate - Refined/retuned numerous aspects of the Eta model physics for use in the NMM - Replace Eta 3DVAR analysis system with the new unified GSI analysis system that has been adapted for application to the WRF-NMM
WRF-NMM
•Same domain as Eta
•Sixty levels like Eta
•Essentially same physics as ETA
•Much better in terrain…doesn’t share the eta’s problems.
Round OneSubjective
Impressions
• Surface and near surface wind and temperature fields are similar
• WRF has more intense, detailed, and more extensive precipitation structures.
Round TwoObjective Verifications
• Both WRF and MM5 were verified against large array of surface observations over the Pacific Northwest.
• Model output was linearly interpolated to observation sites within the 12-km domain encompassing the Pacific Northwest.
• Will show statistics from 12 UTC March 29 to 12 UTC June 6, 2005
2- m Temperature Mean Absolute Error
0
0.5
1
1.5
2
2.5
6 12 18 24
MM5WRF
Forecast Hour
oC
12-km domain, 12 UTC initialization, roughly 60,000 observations in each