David Ovens
Department of Atmospheric Sciences
University of Washington
Stability Issues in WRF High Resolution Model Runs
The Problem
WRF develops numerical instability, CFL errors, that cause
high-resolution runs (in our case 4/3-km) to fail on a frequent
basis. This occurs roughly 1 out of 20 runs. The location of the
instability is over steep mountains. In our domains, Mt Rainier
is the preferred location for this instability.
The Solution
As suggested by Jimy Dudhia to the wrf-users group in an email March
4, 2010, he addressed this problem in a 1995 MWR paper in regards to
the MM5. As he put it:
I just want to see if this is the same problem I addressed in my 1995
paper on MM5 and slope instability. epssm is the equivalent of beta
in MM5 and it biases the time average in w for sound wave
computations. According to that paper epssm should exceed the slope
for stability. MM5's default was 0.4, and you could try even higher
values up to 1.0 to see if anything works.
For us, we have tested 3 cases with cfl failures using epssm=0.2.
Each of these runs has succeeded.
Case by Case Examination
| Failure | Attempted Solution |
|---|
| Date | Error Hour | epssm | Result |
|---|
| 2011020500 | 11 | 0.2 | Success |
| 2011012000 | 9 | 0.3 | Success |
| 2011012000 | 9 | 0.2 | Success |
| 2009092800 | 37 | 0.2 | Success |
| 2010070700 | 1 | 0.2 | Success |
| 2011011912 | 22 | 0.2 | Success |
| | | |
Plot Loops
