Numerical instability using adaptive time stepping?

[lkugler]

Hi,
Is it possible to get numerical instabilities when target_cfl < 1? Why should one use target_cfl=1.2 as in the template namelists? Is it because cfl is calculated from horizontal wind, whereas also vertical winds can violate the CFL criterium?
Is there a special reason why the value of target_hcfl in the template is exactly 0.84?
(Template: https://esrl.noaa.gov/gsd/wrfportal/namelist_input_options.html)

I noticed that numerical instabilities first show themselves in Q_vapor being < 0 in the first model level. Are there explanations /scientific literature describing how it comes to this phenomena?

Thank you!
Regards,
Lukas

 

[Ming Chen]

It is still possible to get numerical instability with target_cfl < 1. The numerical analysis done for dynamics only shows the model should be stable when cfl < 1.4. According to Hutchinson (2007, see below), cfl = 1.2 is taken from the paper by Skamarock and Wicker (2002). But for model that has moisture and full physics, the stability criterion will be more strict. In the paper by Hutchinson below, he shows that they do evaluation both horizontal and vertical cfls.

If you are interested, please take a look at this presentation on adaptive stepping:
http://www2.mmm.ucar.edu/wrf/users/workshops/WS2007/presentation/9-4_Hutchinson.pdf

I cannot say how general it is the QV that does negative first. I often look for perturbation MU.