Transcript WRFworkshop2010b.ppt

WRF Model Physics: Problems
and Progress
Cliff Mass and Dave Ovens
University of Washington
For Several Basic Parameters We Have
Made Substantial Progress During the
Past Ten Years With the Transition
from MM5 to WRF and Physics
Improvements
Case in Point: Precipitation
MM5
Precip
Bias for
24-h
90% and 160%
lines are
contoured
with dashed
and solid lines
For entire
Winter
season
24-h MM5 Precip. Bias Scores over W. WA
But some parameters are still a
problem: wind speed and wind
direction
• WRF generally has a substantial
overprediction bias, with winds being too
geostrophic.
• Not enough contrast between winds over land
and water.
• This problem is evident virtually everywhere
Dealing With Surface Wind Biases
• A consistent error in WRF, both here in the
Northwest and elsewhere, is the tendency for:
– a positive wind speed bias
– winds that are excessively geostrophic.
• Also noticed that there was insufficient contrast
between winds over the water and land (land
winds too large).
• A number of examples were discussed at the
NW Weather Workshop and the last consortium
meeting.
Dealing with surface wind biases
• Last year we experimented with all available
planetary boundary layer schemes (including a
number of new ones) and also tried varying
the number of vertical levels.
• None solved this problem.
• Earlier this year we started running at 1.3 km
grid spacing over western WA and the
problem seems to get much better.
Dealing with Wind Biases
• This led to a hypothesis that the problem is that
the model is not resolving subgrid scale
roughness elements at the surface at 12 and even
4-km resolution.
• Early experiments in increasing u*, which is
related to surface drag were very suggestive—it
decreased the wind and directional biases
significantly.
• This was good enough that we added it to the
real-time system on April 14th.
Old System Wind Speed Bias: Rerun
Jan 1-Feb 8, 2010
New—Lots of Improvements
Optimizing the Approach
• An alternative, and perhaps more
straightforward, way of doing the same thing is to
increase the surface roughness length (z0), and
others have played with this approach (like NCEP,
who has never published anything on it).
• Following the hypothesis, it made sense to make
the increase in roughness dependent on the
variance of the subgrid scale terrain.
• More variance of terrain—more roughness.
New Surface Drag Approach
• During the past few months we have
completed an extensive series of experiments
(view them at:
http://www.atmos.washington.edu/~ovens/windbias/) with
various surface drag approaches.
• Narrowing this down substantially, but here is
one of the best, with z0 dependent on surface
terrain variance over land using 1-km terrain
data base.
Old Wind Bias-00z
Latest Exp Wind Bias
Old-12z
Case Study
LSM Change
• The Noah LSM in the WRF 3.1.1 and 3.2 codes
has a strong cold bias in max temp over the
elevated terrain of the Intermountain West.
• Turning off the Noah LSM and switching to the
simpler 5-layer thermal diffusion scheme (as
was used in our MM5 runs) improves the
surface and 2-m temperatures greatly.
• This change will, however, introduce about a
1°F higher dewpoint temperature bias.
Feb 2-m temp MAE, 00Z
Corresponding Bias
LSM
No LSM
Progress is Slow in Some Parameters
• .
Wind Direction Mean Absolute Error (MAE)
60
36-km
12-km
4-km
40
30
20
10
Forecast Hour
48
45
42
39
36
33
30
27
24
21
18
15
12
9
6
3
0
0
MAE (degrees)
50