[RegCNET] Re: Question about time step
moetasim
moetasim at gcisc.org.pk
Mon Oct 24 21:28:34 CEST 2005
>tell me onething that why model crahses at the larger time step and
>why it works at the shorter time step, as the time step at which the>
>model crahses(at bigger one) comes in the shorter time step also....
>Hoping to get ur kind answer
>Adnan
I am CCing answer at RegCNET so that others may comment and correct me or may get benefit.
The time step restriction is applied for maintaining numerical calulations stability. The stability
restrcition is usually determined by the horizontal resoltution of your grid. Time step should be
smaller than the grid size divided by the maximum velocity. So normally 3 x resolution is general
criteria for time step in most models.
In RegCM, when you have your map factor ~ 1, this means the grid boxes are almost equal in size
thorughout the domain so based on the 3xresolution criteria, you can choose your time step but if
map factor largely fluctuates above n below '1' then this means that your grid boxes are not of equal
size thorughout the domain and ration of x/y is more or less than '1' . This is becasue your selected
domain extends largely northward or southward. In such situation you have to decrease your time
step in following way:
If you have map factor max value say 1.25 and your horizontal gird resoltuion 50 km then in this
case, you should use a modified formula for stable time step i.e
3 x resoltuion/max. map factor
=> 3 x 50/ 1.25 = 120
So this is the maximum time step that you can have in this case.
To avoid the problem of unstable domain, we use different map projections for different regions of
the world. Like in your case, you use rotated mercator map projection but if you try to extend your
domain in north and south direction while keeping your centre somehwere near 30.0 degree N, you
will see that domain becomes unstable and model crashes at normal time step.
Also, Don't confuse model output frequecny with numerical calculation time step. Model output
frequency determines maximum resolved temporal scales. I believe that nyquist criterian should be
followed over here as well which tells that maximum spatial resolvable scale is at least = 2 deltax
(strictly speaking it is 4 x deltax). This implies that maximum temopral resolved scale is = 2 x output
frequency (strictly speaking it should be 4 x output frequency). This means with RegCM surface 03
hourly output you can look into sub-daily times scales provided that you have enough high
resoltuion.
MOet
More information about the RegCNET
mailing list