representing diverse scales in olam advantages and challenges of locally-refined unstructured grids...
DESCRIPTION
OLAM background OLAM is based partly on RAMS, a limited area model specializing in mesoscale and cloud scale simulations. The original motivation for OLAM was to provide a unified global-regional modeling framework in order to avoid the disadvantages of limited area models. External GCM domain Traditional RCM domain Information flow Numerical noise at lateral boundary OLAM global lower resolution domain OLAM local high resolution region Well behaved transition region Information flowTRANSCRIPT
Representing Diverse Scales in OLAM Advantages and Challenges of Locally-Refined Unstructured Grids
Robert L. Walko
Rosenstiel School of Meteorology and Physical OceanographyUniversity of Miami, Miami, FL
Presented at IPAM, UCLA, April 16, 2010
Acknowledgments: 1) Roni Avissar – Dean of Rosenstiel School2) Pratt School of Engineering, Duke University3) NSF, NASA, DOE4) William R. Cotton, Roger A. Pielke, Sr.
Outline:
1. Background and motivation for OLAM (the Ocean-Land-Atmosphere
Model)
2. Summary of OLAM dynamic core – local mesh refinement – some physics
3. Numerical experiments to examine how convection responds to variable
resolution
4. What the experiments tell us – some suggestions for future research
5. Summary
OLAM background
OLAM is based partly on RAMS, a limited area model specializing in mesoscale and cloud scale simulations.
The original motivation for OLAM was to provide a unified global-regional modeling framework in order to avoid the disadvantages of limited area models.
External GCM domain
Traditional RCM domain
Informationflow
Numerical noise at lateral boundary
OLAM global lower resolution domain
OLAM local highresolution region
Well behaved transition region
Informationflow
OLAM went through several versions, including overlapping polar-stereographic projections, but eventually we settled on the geodesic mesh because of its facility in local mesh refinement.
OLAM works with either triangles or hexagons as the primary mesh. With hexagons, a few pentagons and heptagons are required for local mesh refinement.
Physical parameterizations were adapted from RAMS, with others added later.
OLAM’s dynamic core was a new formulation, not from RAMS.
iiiiii FgvpVvtV
2
HVt
)(
QVsts
)(
V
d
V
P CR
CC
vvdd pRRp
0
1
Mass & Momentum conserving FV dynamic core
vV
Momentum conservation(component i)
Total mass conservation
conservation
Scalar conservation(e.g. )
Equation of State
Momentum definition
)253,max(1
TCq
p
lat = potential temperature = ice-liquid potential temperature
cvd Total density
/vvs
MVt
dd
dFdVdt m
dFdVdt
dFdVsdst s
.
Discretized equations:
Integrate over finite volumes andapply Gauss Divergence Theorem:
dFdgdvdxpdVvdV
t iiii
ii 2
d
Terrain-following coordinates
OLAM uses cut (“shaved”) grid cells
Anomalous vertical dispersion
Wind
Terrain-following coordinate levels
Terrain
Investigate behavior of parameterized and/or resolved convection across grid scales.
We look only at accumulated surface precipitation at end of 6 hours.
Choose very simple horizontally-homogeneous forcing of environment (no topography, no land/water, no large-scale flow or disturbances).
Begin with horizontally-homogeneous, conditionally unstable atmosphere at rest.
Impose constant surface sensible heat flux (~300 W m-2).
Numerical experiments
Grid number Grid cell size 1 200 km 2 100 3 50 4 25 5 12 6 6 7 3
Cumulus parameterization only – no microphysics
Average parameterized convective precipitation over each refinement zone
We need to ask ourselves:
Do we want convective parameterization to give uniform precipitation at all scales where it is applied?Or not?
Is convective parameterization performing as intended, i.e., according to its design? In particular, is it responding correctly to the host model’s ability (and tendency) to generate stronger W on finer grids?
If we choose to take the “route 1” approach described by A. Arakawa, can we adjust the fractional updraft area in a way to achieve the desired result?
We find that area-averaged parameterized precipitation is insensitive to grid spacing for cells larger than about 30 km, but increases (by about 50%) as cells reduce to 3 km.
Microphysics only – no cumulus parameterization
Average resolved convective precipitation over each refinement zone
Should the model even be allowed to produce convective-type vertical motion on 6 km, 12 km, or larger cells? Such convection is unrealistically wide, and is not well represented on the grid.
Perhaps a convective parameterization should be retained at these resolutions to remove convective instability (the “route 1” approach)
Cumulus parameterization and microphysics together
Horizontally averaged combined precipitation
Horizontally-averaged parameterized convective precipitation
Obviously, in locations where convection is resolvable, we want resolved convection to prevail and parameterized convection to become inactive.
This will not not happen unless parameterized convection is switched off or is somehow supressed.
Can this transition be made smoothly?
Many more complicating factors need to be considered:
Ambient wind
Large-scale disturbances
Orographic lifting
High-CAPE convection (not generally permitted by
parameterized convection)
Can blending methods be made to work well for all situations?
Possible configuration for embedded “superparameterization” grid:
Inner cells fine enough to resolve primary updraft and rain-cooled downdraft
Overall cluster of cells wide enough to encompass mesoscale subsidence
Achieves both goals with fewer cells than uniform embedded grid
Motion of convection grid
Summary:
Local mesh refinement within a single model framework enables numerical experiments that examine how a model represents parameterized and resolved convection:
1) at different scales2) where grid scale changes
OLAM has the appropriate tools for this investigation, including extensions for topographic, land/sea, and realistic large-scale forcing
Both “route 1” and “route 2” approaches are being investigated.