Sensitivity of COSMO-CLM Regional Climate Model to the domain selection – in #8: General Questions

in #8: General Questions

Cookies disclaimer

Our site saves small pieces of text information (cookies) on your
device in order to verify your login. These cookies are essential
to provide access to resources on this website and it will not
work properly without.
Learn more

<p>
Dear
<span class="caps">
CCLM
</span>
Community,
</p>
<p>
I conducted an experiment using slightly different domains from the default one over Europe region.
<br/>
The experiment indicates that by shifting just the domain, model will produce different results in monthly values.
<br/>
A mini-report on this issue can be found here: http://users.met.fu-berlin.de/~BijanFallah/CCLM.pdf
<br/>
I am looking forward to receiving your Ideas or suggestions about this issue.
<br/>
The model set up for the default domain (without shifting and nboundlines=3) can be found here: http://users.met.fu-berlin.de/~BijanFallah/YUSPECIF .
</p>
<p>
With very best wishes,
<br/>
Bijan Fallah
</p>

<p>
Dear
<span class="caps">
CCLM
</span>
Community,
</p>
<p>
I conducted an experiment using slightly different domains from the default one over Europe region.
<br/>
The experiment indicates that by shifting just the domain, model will produce different results in monthly values.
<br/>
A mini-report on this issue can be found here: http://users.met.fu-berlin.de/~BijanFallah/CCLM.pdf
<br/>
I am looking forward to receiving your Ideas or suggestions about this issue.
<br/>
The model set up for the default domain (without shifting and nboundlines=3) can be found here: http://users.met.fu-berlin.de/~BijanFallah/YUSPECIF .
</p>
<p>
With very best wishes,
<br/>
Bijan Fallah
</p>

Sensitivity of COSMO-CLM Regional Climate Model to the domain selection

I conducted an experiment using slightly different domains from the default one over Europe region.
The experiment indicates that by shifting just the domain, model will produce different results in monthly values.
A mini-report on this issue can be found here: http://users.met.fu-berlin.de/~BijanFallah/CCLM.pdf
I am looking forward to receiving your Ideas or suggestions about this issue.
The model set up for the default domain (without shifting and nboundlines=3) can be found here: http://users.met.fu-berlin.de/~BijanFallah/YUSPECIF .

<p>
Hallo Bijan,
<br/>
I hope, I understand it correctly. You show
<span class="caps">
RMSE
</span>
of monthly mean values. If this is the case, the result is as expected and shows simply the internal variability due to sensible dependence on initial condition.
</p>
<p>
Regards, Andreas
</p>

<p>
Hallo Bijan,
<br/>
I hope, I understand it correctly. You show
<span class="caps">
RMSE
</span>
of monthly mean values. If this is the case, the result is as expected and shows simply the internal variability due to sensible dependence on initial condition.
</p>
<p>
Regards, Andreas
</p>

Hallo Bijan,
I hope, I understand it correctly. You show
RMSE
of monthly mean values. If this is the case, the result is as expected and shows simply the internal variability due to sensible dependence on initial condition.

<p>
Hello Andreas,
</p>
<p>
thank you for your response. If that is due to internal variability, then my concern is already solved. My motivation for this experiment was also to create different climate simulations (with different initial conditions). Given that the error pattern (shifted minus Default domain) seems to have large values near the boundaries, I was afraid if I am doing something wrong.
</p>
<p>
With very best wishes,
<br/>
Bijan
</p>

<p>
Hello Andreas,
</p>
<p>
thank you for your response. If that is due to internal variability, then my concern is already solved. My motivation for this experiment was also to create different climate simulations (with different initial conditions). Given that the error pattern (shifted minus Default domain) seems to have large values near the boundaries, I was afraid if I am doing something wrong.
</p>
<p>
With very best wishes,
<br/>
Bijan
</p>

thank you for your response. If that is due to internal variability, then my concern is already solved. My motivation for this experiment was also to create different climate simulations (with different initial conditions). Given that the error pattern (shifted minus Default domain) seems to have large values near the boundaries, I was afraid if I am doing something wrong.

<p>
Hi Bijan,
</p>
<p>
I have two questions:
</p>
<p>
1. Could you explain in some more detail how you determined the forcing data for the simulations in the different domains? The question came up because your “INT2LM domain” is larger than the various
<span class="caps">
CCLM
</span>
domains.
</p>
<p>
2. You used a lateral relaxation zone (sponge zone) of about 590 km (see rlwidth Namelistparameter in your
<span class="caps">
YUSPEFIC
</span>
) and the explicit formulation of the boundary conditions, what means, that rlwidth plays a role.
Did you ignore this sponge zone in your analyses? If not, then this might be the reason for large differences near the boundaries.
</p>
<p>
Hans-Juergen
</p>

<p>
Hi Bijan,
</p>
<p>
I have two questions:
</p>
<p>
1. Could you explain in some more detail how you determined the forcing data for the simulations in the different domains? The question came up because your “INT2LM domain” is larger than the various
<span class="caps">
CCLM
</span>
domains.
</p>
<p>
2. You used a lateral relaxation zone (sponge zone) of about 590 km (see rlwidth Namelistparameter in your
<span class="caps">
YUSPEFIC
</span>
) and the explicit formulation of the boundary conditions, what means, that rlwidth plays a role.
Did you ignore this sponge zone in your analyses? If not, then this might be the reason for large differences near the boundaries.
</p>
<p>
Hans-Juergen
</p>

1. Could you explain in some more detail how you determined the forcing data for the simulations in the different domains? The question came up because your “INT2LM domain” is larger than the various
CCLM
domains.

2. You used a lateral relaxation zone (sponge zone) of about 590 km (see rlwidth Namelistparameter in your
YUSPEFIC
) and the explicit formulation of the boundary conditions, what means, that rlwidth plays a role.
Did you ignore this sponge zone in your analyses? If not, then this might be the reason for large differences near the boundaries.

<p>
Hi Hans-Juergen,
</p>
<p>
thanks a lot for the response. I have used
<span class="caps">
ERAI
</span>
nterim data and yes the INT2LM domain is 12 grid points larger (6 grid points from each side). I have replotted Figure 1 with all the grids over it:
</p>
<p>
http://users.met.fu-berlin.de/~BijanFallah/Figure_test.pdf
</p>
<p>
And for analysis I use the overlapping area of the domains so automatically it will be 4 grid points smaller than the not-shifted domain (inner solid black box in the figures). And then in the next step I even make the
<span class="caps">
CCLM
</span>
domain smaller in Figure 3. Given this I imagine that the sponge zone is already ignored?
</p>
<p>
Best Regards,
<br/>
Bijan
</p>

<p>
Hi Hans-Juergen,
</p>
<p>
thanks a lot for the response. I have used
<span class="caps">
ERAI
</span>
nterim data and yes the INT2LM domain is 12 grid points larger (6 grid points from each side). I have replotted Figure 1 with all the grids over it:
</p>
<p>
http://users.met.fu-berlin.de/~BijanFallah/Figure_test.pdf
</p>
<p>
And for analysis I use the overlapping area of the domains so automatically it will be 4 grid points smaller than the not-shifted domain (inner solid black box in the figures). And then in the next step I even make the
<span class="caps">
CCLM
</span>
domain smaller in Figure 3. Given this I imagine that the sponge zone is already ignored?
</p>
<p>
Best Regards,
<br/>
Bijan
</p>

thanks a lot for the response. I have used
ERAI
nterim data and yes the INT2LM domain is 12 grid points larger (6 grid points from each side). I have replotted Figure 1 with all the grids over it:

And for analysis I use the overlapping area of the domains so automatically it will be 4 grid points smaller than the not-shifted domain (inner solid black box in the figures). And then in the next step I even make the
CCLM
domain smaller in Figure 3. Given this I imagine that the sponge zone is already ignored?

<p>
Now I not sure whether I really understood what you did.
<br/>
I assumed that you performed an individual simulation for each of your domains, the reference domain, and those with the shifts.
<br/>
That was the reason why I asked my first question.
</p>
<p>
But now I interpret your answer in such a way that you performed only one simulation for the domain that corresponds to the “INT2LM-domain”,
<br/>
and for your post-processing, that means your T_2M analyses, you have chosen the different “sub-domains”.
<br/>
If this is the case, then I am sure that you did not ignore the sponge zone.
</p>
<p>
This zone has a size of about 590 km, in terms of grid-points that are about 12 points, plus the 3 “nboundlines” points.
<br/>
Thus, in units of grid-points your sponge zone has a size of 15 grid-points at each side of your “INT2LM-domain”.
<br/>
But with your various shifts, large parts of your sub-domains are within the sponge zone of the “INT2LM-domain”.
</p>
<p>
Hans-Juergen
</p>

<p>
Now I not sure whether I really understood what you did.
<br/>
I assumed that you performed an individual simulation for each of your domains, the reference domain, and those with the shifts.
<br/>
That was the reason why I asked my first question.
</p>
<p>
But now I interpret your answer in such a way that you performed only one simulation for the domain that corresponds to the “INT2LM-domain”,
<br/>
and for your post-processing, that means your T_2M analyses, you have chosen the different “sub-domains”.
<br/>
If this is the case, then I am sure that you did not ignore the sponge zone.
</p>
<p>
This zone has a size of about 590 km, in terms of grid-points that are about 12 points, plus the 3 “nboundlines” points.
<br/>
Thus, in units of grid-points your sponge zone has a size of 15 grid-points at each side of your “INT2LM-domain”.
<br/>
But with your various shifts, large parts of your sub-domains are within the sponge zone of the “INT2LM-domain”.
</p>
<p>
Hans-Juergen
</p>

Now I not sure whether I really understood what you did.
I assumed that you performed an individual simulation for each of your domains, the reference domain, and those with the shifts.
That was the reason why I asked my first question.

But now I interpret your answer in such a way that you performed only one simulation for the domain that corresponds to the “INT2LM-domain”,
and for your post-processing, that means your T_2M analyses, you have chosen the different “sub-domains”.
If this is the case, then I am sure that you did not ignore the sponge zone.

This zone has a size of about 590 km, in terms of grid-points that are about 12 points, plus the 3 “nboundlines” points.
Thus, in units of grid-points your sponge zone has a size of 15 grid-points at each side of your “INT2LM-domain”.
But with your various shifts, large parts of your sub-domains are within the sponge zone of the “INT2LM-domain”.

<p>
Hallo Bijan and Hans-Juergen,
<br/>
I agree with Hans-Juergen, that the strong values near the boundaries are in particular strong due to domain shifting and thus shifting of the boundary condition. However, the strong differences are outside the sponge zone overlap. So, what you see here is the boundary condition effect in your simulation, which is quite strong (in comparison with optimised configuration cases.
</p>
<p>
Greetings, Andreas
</p>

<p>
Hallo Bijan and Hans-Juergen,
<br/>
I agree with Hans-Juergen, that the strong values near the boundaries are in particular strong due to domain shifting and thus shifting of the boundary condition. However, the strong differences are outside the sponge zone overlap. So, what you see here is the boundary condition effect in your simulation, which is quite strong (in comparison with optimised configuration cases.
</p>
<p>
Greetings, Andreas
</p>

Hallo Bijan and Hans-Juergen,
I agree with Hans-Juergen, that the strong values near the boundaries are in particular strong due to domain shifting and thus shifting of the boundary condition. However, the strong differences are outside the sponge zone overlap. So, what you see here is the boundary condition effect in your simulation, which is quite strong (in comparison with optimised configuration cases.

<p>
Hallo Andreas and Hans-Juergen,
</p>
<p>
That is a good point! The sponge zone is now 12 times the grid mesh size.
<br/>
So I will choose a bigger domain and cut the sponge zone of the shifted + default domain for analysis.
</p>
<p>
Thanks a lot for comments.
</p>
<p>
With very best wishes,
<br/>
Bijan
</p>

<p>
Hallo Andreas and Hans-Juergen,
</p>
<p>
That is a good point! The sponge zone is now 12 times the grid mesh size.
<br/>
So I will choose a bigger domain and cut the sponge zone of the shifted + default domain for analysis.
</p>
<p>
Thanks a lot for comments.
</p>
<p>
With very best wishes,
<br/>
Bijan
</p>

That is a good point! The sponge zone is now 12 times the grid mesh size.
So I will choose a bigger domain and cut the sponge zone of the shifted + default domain for analysis.

<p>
Dear Colleagues,
</p>
<p>
I have repeated the experiment with a larger domain (155×150 grid points) and calculated the
<span class="caps">
RMSE
</span>
for T2M (http://users.met.fu-berlin.de/~BijanFallah/Figure07_RMSE.pdf).
</p>
<p>
As can be seen the previous error pattern near the boundaries for analysis are removed. I guess given that the domain is very large now, the model has more “freedom” and the error patterns in the centre of the domain are simply the internal variability that is larger than the previous settings.
</p>
<p>
Yours Sincerely,
<br/>
Bijan
</p>

<p>
Dear Colleagues,
</p>
<p>
I have repeated the experiment with a larger domain (155×150 grid points) and calculated the
<span class="caps">
RMSE
</span>
for T2M (http://users.met.fu-berlin.de/~BijanFallah/Figure07_RMSE.pdf).
</p>
<p>
As can be seen the previous error pattern near the boundaries for analysis are removed. I guess given that the domain is very large now, the model has more “freedom” and the error patterns in the centre of the domain are simply the internal variability that is larger than the previous settings.
</p>
<p>
Yours Sincerely,
<br/>
Bijan
</p>

I have repeated the experiment with a larger domain (155×150 grid points) and calculated the
RMSE
for T2M (http://users.met.fu-berlin.de/~BijanFallah/Figure07_RMSE.pdf).

As can be seen the previous error pattern near the boundaries for analysis are removed. I guess given that the domain is very large now, the model has more “freedom” and the error patterns in the centre of the domain are simply the internal variability that is larger than the previous settings.

## Sensitivity of COSMO-CLM Regional Climate Model to the domain selection

Dear CCLM Community,

I conducted an experiment using slightly different domains from the default one over Europe region.

The experiment indicates that by shifting just the domain, model will produce different results in monthly values.

A mini-report on this issue can be found here: http://users.met.fu-berlin.de/~BijanFallah/CCLM.pdf

I am looking forward to receiving your Ideas or suggestions about this issue.

The model set up for the default domain (without shifting and nboundlines=3) can be found here: http://users.met.fu-berlin.de/~BijanFallah/YUSPECIF .

With very best wishes,

Bijan Fallah

Hallo Bijan,

I hope, I understand it correctly. You show RMSE of monthly mean values. If this is the case, the result is as expected and shows simply the internal variability due to sensible dependence on initial condition.

Regards, Andreas

Hello Andreas,

thank you for your response. If that is due to internal variability, then my concern is already solved. My motivation for this experiment was also to create different climate simulations (with different initial conditions). Given that the error pattern (shifted minus Default domain) seems to have large values near the boundaries, I was afraid if I am doing something wrong.

With very best wishes,

Bijan

Hi Bijan,

I have two questions:

1. Could you explain in some more detail how you determined the forcing data for the simulations in the different domains? The question came up because your “INT2LM domain” is larger than the various CCLM domains.

2. You used a lateral relaxation zone (sponge zone) of about 590 km (see rlwidth Namelistparameter in your YUSPEFIC ) and the explicit formulation of the boundary conditions, what means, that rlwidth plays a role. Did you ignore this sponge zone in your analyses? If not, then this might be the reason for large differences near the boundaries.

Hans-Juergen

Hi Hans-Juergen,

thanks a lot for the response. I have used ERAI nterim data and yes the INT2LM domain is 12 grid points larger (6 grid points from each side). I have replotted Figure 1 with all the grids over it:

http://users.met.fu-berlin.de/~BijanFallah/Figure_test.pdf

And for analysis I use the overlapping area of the domains so automatically it will be 4 grid points smaller than the not-shifted domain (inner solid black box in the figures). And then in the next step I even make the CCLM domain smaller in Figure 3. Given this I imagine that the sponge zone is already ignored?

Best Regards,

Bijan

Now I not sure whether I really understood what you did.

I assumed that you performed an individual simulation for each of your domains, the reference domain, and those with the shifts.

That was the reason why I asked my first question.

But now I interpret your answer in such a way that you performed only one simulation for the domain that corresponds to the “INT2LM-domain”,

and for your post-processing, that means your T_2M analyses, you have chosen the different “sub-domains”.

If this is the case, then I am sure that you did not ignore the sponge zone.

This zone has a size of about 590 km, in terms of grid-points that are about 12 points, plus the 3 “nboundlines” points.

Thus, in units of grid-points your sponge zone has a size of 15 grid-points at each side of your “INT2LM-domain”.

But with your various shifts, large parts of your sub-domains are within the sponge zone of the “INT2LM-domain”.

Hans-Juergen

Hallo Bijan and Hans-Juergen,

I agree with Hans-Juergen, that the strong values near the boundaries are in particular strong due to domain shifting and thus shifting of the boundary condition. However, the strong differences are outside the sponge zone overlap. So, what you see here is the boundary condition effect in your simulation, which is quite strong (in comparison with optimised configuration cases.

Greetings, Andreas

Hallo Andreas and Hans-Juergen,

That is a good point! The sponge zone is now 12 times the grid mesh size.

So I will choose a bigger domain and cut the sponge zone of the shifted + default domain for analysis.

Thanks a lot for comments.

With very best wishes,

Bijan

Dear Colleagues,

I have repeated the experiment with a larger domain (155×150 grid points) and calculated the RMSE for T2M (http://users.met.fu-berlin.de/~BijanFallah/Figure07_RMSE.pdf).

As can be seen the previous error pattern near the boundaries for analysis are removed. I guess given that the domain is very large now, the model has more “freedom” and the error patterns in the centre of the domain are simply the internal variability that is larger than the previous settings.

Yours Sincerely,

Bijan