grid resolution
Adriana Huerta-Casas
Dear GETM users,
Is there any grid-size limitation in GETM? I would like to use it to include a module concerning some biogeochemical cycles in a coral reef ecosystem (around 14 equations). I am thinking in two domains:
1) x, y, z: 10Km x 10Km x 10m, with an horizontal grid of around 20m.
2) x, y, z: 70m x 40m x 10m, with an horizontal grid spacing of 1m.
Are these domains possible and stable in GETM?
Thanks a lot!
adriana
Ulf Gräwe
Hi Adriana,
there are no limitations on the grid size, beside the numerical
limitations. Since GETM is based on the hydrostatic assumption, you have
to check whether nonhydrostatic effects can effect your solution. This
shouldn’t be a problem for the first domain. However, your aspect ratio
of horizontal grid spacing to water depth is rather distorted in your
second domain. Mostly, this ratio is much larger than one. However, in
your case it is even much smaller than one.
I never used a model setup with such a small ratio. I can only advise
you that you have to be very careful with the interpretation of the
model output, whatever it is.
Hope that helps,
ulf
> Thanks a lot!
> adriana
>
>
>
--
Ulf Gräwe
Leibniz Institute for Baltic Sea Research Warnemuende
Dept. for Physical Oceanography and Instrumentation
Seestrasse 15
D-18119 Rostock-Warnemuende
Germany
Tel. : +49 (381) 5197-358
Fax : +49 (381) 5197-440
Lars Umlauf
GETM is a hydrostatic model, implying that the horizontal scales you
resolve must be "much larger" than the vertical scales. Your first
example is at the limit in that sense - but in the second example your
are certainly beyond anything admissable with a hydrostatic model. In
this case, you need non-hydrostatic physics, which is not yet available
with GETM.
Best wishes,
Lars
Karsten Bolding
But if the phenomenon you want to solve is hydrostatic does it then
'harm' to solve in higher resolution? If you want to solve the tides in
the North Sea in a 6nm resolution I guess you are still allowed to run
wiith a 1s timestep - aren't you also allow to run with 1m resolution?
Karsten
Hans Burchard
> Hello
Hi Karsten,
>
> But if the phenomenon you want to solve is hydrostatic does it then
> 'harm' to solve in higher resolution? If you want to solve the tides in
> the North Sea in a 6nm resolution I guess you are still allowed to run
> wiith a 1s timestep - aren't you also allow to run with 1m resolution?
>
No, this would not be permitted because you resolve internal waves at
lenghts where non-hydrostatic pressure gradients will be essential. Thus,
you generate short internal waves that will propagate at long internal
wave speed.
> Karsten
>
Hans.
--
Hans Burchard
Leibniz Institute for Baltic Sea Research Warnemuende
Dept. for Physical Oceanography and Instrumentation
Seestrasse 15
D-18119 Rostock-Warnemuende
Germany
Tel. : +49 - 381 - 5197 -140
Fax : +49 - 381 - 5197 -114
web : http://www.io-warnemuende.de/hans-burchard-en.html
E-Mail: hans.b...@io-warnemuende.de
skype : hans.burchard
Adriana Huerta-Casas
Thanks for your feedback. Now I am concern about the computational cost of my setup, I will include biogeochemistry in a coral reef atoll (around 14 equations). I have access to 24 nodes.
my setup: horizontal grid of 10Km x 10Km with 20m grid spacing. Vertically 15m, with at least 10 layers, following terrain coordinates.
How long does a year run will take with a 5min timestep?
cheers,
adriana
Karsten Bolding
>
> Thanks for your feedback. Now I am concern about the computational cost of my setup, I will include biogeochemistry in a coral reef atoll (around 14 equations). I have access to 24 nodes.
The most expensive in running a bio-model is actually the advection -
the process calculation is much smaller than the time it takes to advect
the 13 variables around.
24 nodes - how many cores does each have? And do you know the CPU type?
>
> my setup: horizontal grid of 10Km x 10Km with 20m grid spacing. Vertically 15m, with at least 10 layers, following terrain coordinates.
The domain will be 500x500x10 which is by far the biggest GETM setup.
The problem here will be the timestep as you have a dx of only 20m.
With a sub-domain size of 50x50 you'll need a maximum of 100 cores and
you'll have that on the 24 nodes.
I good way forward would be to use a test setup with like 40x40 (or even
coarser) for a lot of the testing.
>
> How long does a year run will take with a 5min timestep?
This will be a gross violation of the CFL-criterion. You are talking
about an external time-step of around 1 sec and maybe a split-factor of
10-20 so the bio is to be called every 10-20 secs.
To run the setup is doable - do you need to make annual simulations?
>
> cheers,
> adriana
>
Karsten
Karsten Bolding
<snip>
> > my setup: horizontal grid of 10Km x 10Km with 20m grid spacing. Vertically 15m, with at least 10 layers, following terrain coordinates.
>
> The domain will be 500x500x10 which is by far the biggest GETM setup.
should have read:
is far from making it the biggest GETM setup.
so the opposite :-)
>
Adriana Huerta-Casas
Thanks. I am reconsidering my setup. I will use the refine resolution in a 2D setup x-z, and a coarser one in 3D, around 100m grid spacing, then 100 X 100, with a timestep as small as permitted for both.
The cluster software says: 28processors available (84.94GHz), 4 unavailable (8GHz)
OS is MacOS Server 10.6
I believe a year run for the 2D is reasonable and will consider it for the 3D, depending in the cost.
Then, which will be my computational cost?
cheers,
adriana
----- Original Message -----
Bjarne Büchmann
> No, this would not be permitted because you resolve internal
> waves at lenghts where non-hydrostatic pressure gradients will
> be essential. Thus, you generate short internal waves that
> will propagate at long internal wave speed.
I think you imply that GETM will not converge spatially(?)
Personally, I find that highly disturbing. Normally, the solution
should not depend on the resolution as long as the resolution is
"fine enough". Obviously, you may resolve phenomena, which you
are not really interested in, but what you write implies that
the (coarse) grid spacing is used to "filter" unwanted solutions.
From a numerical point of view, that raises some issues...
> Hans.
MED VENLIG HILSEN
Bjarne Büchmann
Civilingeniør, PhD
Overgaden oven vandet 62 B Postboks 1919
DK - 1023 København K
TLF. +45 3268 697
B...@FRV.DK
Hans Burchard
> Hans,
Hej Bjarne,
>
>> No, this would not be permitted because you resolve internal
>> waves at lenghts where non-hydrostatic pressure gradients will
>> be essential. Thus, you generate short internal waves that
>> will propagate at long internal wave speed.
>
> I think you imply that GETM will not converge spatially(?)
> Personally, I find that highly disturbing. Normally, the solution
> should not depend on the resolution as long as the resolution is
> "fine enough". Obviously, you may resolve phenomena, which you
> are not really interested in, but what you write implies that
> the (coarse) grid spacing is used to "filter" unwanted solutions.
>>From a numerical point of view, that raises some issues...
The convergence you talk about will take place froma certain grid
resolution downwards, depending on the values of the horizontal
diffusivities and viscosities. We are often running GETM with zero
horizontal values for those quantities, and that would certainly converge
to a physically unsound solution. Just think about resolving surface waves
which are much shorter than the water depth. For a hydrostatic model those
will still propagate at a phase speed of sqrt{gH}, which certainly is
wrong. Only a proper amount of horizontal viscosity woul be able to kill
all those "wrong" waves. And I have no clue which values should be used
for that.
It is therefore esential to apply proper horizontal mixing and disspation
operators.
All the best,
Hans.
>
>> Hans.
>
> �MED VENLIG HILSEN
> �
> Bjarne B�chmann
> �Civilingeni�r, PhD
>
> Overgaden oven vandet 62 B Postboks 1919
> DK - 1023 K�benhavn K
> TLF. +45 3268 697
> B...@FRV.DK
>