Single Column Simulation with CCLM – in #8: General Questions

in #8: General Questions

<p> Dear all, <br/> I am planning to do single column simulations with <span class="caps"> CCLM </span> 5.0. <br/> Now I am wondering whether there is a maximum number of grid cells needed. </p> <p> When I choose my domain for example to be 9×9 grid cells there comes the following error in INT2LM: <br/> ERROR in tautsp(), src_vert_interpol.f90: while processing j-index: </p> <p> When I extend the domain a little bit, I get the following error in INT2LM: <br/> Caught signal 8 (Floating point exception) <br/> mxm_handle_error() </p> <p> And when I extend my domain up to more then 20×20 grid cells without changing anything else it works fine. <br/> Is there any namelist variable I have to change when using a smaller domain or is this not possible with <span class="caps"> CCLM </span> ? </p> <p> Thank you very much for your help and best regards, <br/> Eva </p>

  @evanowatzki in #cdf57aa

<p> Dear all, <br/> I am planning to do single column simulations with <span class="caps"> CCLM </span> 5.0. <br/> Now I am wondering whether there is a maximum number of grid cells needed. </p> <p> When I choose my domain for example to be 9×9 grid cells there comes the following error in INT2LM: <br/> ERROR in tautsp(), src_vert_interpol.f90: while processing j-index: </p> <p> When I extend the domain a little bit, I get the following error in INT2LM: <br/> Caught signal 8 (Floating point exception) <br/> mxm_handle_error() </p> <p> And when I extend my domain up to more then 20×20 grid cells without changing anything else it works fine. <br/> Is there any namelist variable I have to change when using a smaller domain or is this not possible with <span class="caps"> CCLM </span> ? </p> <p> Thank you very much for your help and best regards, <br/> Eva </p>

Single Column Simulation with CCLM

EH wrote

Dear all,
I am planning to do single column simulations with CCLM 5.0.
Now I am wondering whether there is a maximum number of grid cells needed.

When I choose my domain for example to be 9×9 grid cells there comes the following error in INT2LM:
ERROR in tautsp(), src_vert_interpol.f90: while processing j-index:

When I extend the domain a little bit, I get the following error in INT2LM:
Caught signal 8 (Floating point exception)
mxm_handle_error()

And when I extend my domain up to more then 20×20 grid cells without changing anything else it works fine.
Is there any namelist variable I have to change when using a smaller domain or is this not possible with CCLM ?

Thank you very much for your help and best regards,
Eva

View in channel
<p> I am not aware of a single column version in the <span class="caps"> CLM </span> -Community. Maybe there is one in the <span class="caps"> COSMO </span> consortium? </p>

  @burkhardtrockel in #e64448d

<p> I am not aware of a single column version in the <span class="caps"> CLM </span> -Community. Maybe there is one in the <span class="caps"> COSMO </span> consortium? </p>
BR replied

I am not aware of a single column version in the CLM -Community. Maybe there is one in the COSMO consortium?

View in channel
<p> There is a single column model from Matthias Raschendorfer, but this is not public and I don’t think that you used that. Maybe you configured the <span class="caps"> COSMO </span> -Model in a way that it runs on a very small domain (9×9 grid points). Or what did you do exactly? </p>

  @ulrichschättler in #a379436

<p> There is a single column model from Matthias Raschendorfer, but this is not public and I don’t think that you used that. Maybe you configured the <span class="caps"> COSMO </span> -Model in a way that it runs on a very small domain (9×9 grid points). Or what did you do exactly? </p>
US replied

There is a single column model from Matthias Raschendorfer, but this is not public and I don’t think that you used that. Maybe you configured the COSMO -Model in a way that it runs on a very small domain (9×9 grid points). Or what did you do exactly?

View in channel
<p> Yes, that is what I did. I run the model on a very small domain. But it only runs without errors when the domain is bigger than 20×20 grid points. </p>

  @evanowatzki in #084e938

<p> Yes, that is what I did. I run the model on a very small domain. But it only runs without errors when the domain is bigger than 20×20 grid points. </p>
EH replied

Yes, that is what I did. I run the model on a very small domain. But it only runs without errors when the domain is bigger than 20×20 grid points.

View in channel
<p> How do you control it? Is it an artificial data run? Or do you have initial and boundary data for it? Could you show us the namelist configuration ( <span class="caps"> YUSPECIF </span> )? </p>

  @ulrichschättler in #5a2fc18

<p> How do you control it? Is it an artificial data run? Or do you have initial and boundary data for it? Could you show us the namelist configuration ( <span class="caps"> YUSPECIF </span> )? </p>
US replied

How do you control it? Is it an artificial data run? Or do you have initial and boundary data for it? Could you show us the namelist configuration ( YUSPECIF )?

View in channel
<p> I use <span class="caps"> ERA </span> -Interim as initial data. Here the <span class="caps"> YUSPECIF </span> of my simulation. Thank you very much </p>

  @evanowatzki in #d91321f

<p> I use <span class="caps"> ERA </span> -Interim as initial data. Here the <span class="caps"> YUSPECIF </span> of my simulation. Thank you very much </p>
EH replied

I use ERA -Interim as initial data. Here the YUSPECIF of my simulation. Thank you very much

View in channel
<p> Ok, I see one problem: You have ie_tot=je_tot=21 and nprocx=nprocy=24! I wonder that it works at all. Because this means that you distribute the 21×21=441 grid points to 24×24=576 processors. <br/> You have to use nprocx=nprocy=1 (!), so only 1 <span class="caps"> CPU </span> . Then it should also work with less grid points. You have nboundlines=3 (what you need for the dynamics), so can try to use ie_tot=je_tot=7. This is one grid point in the interior domain and 3 grid points of boundaries around. </p> <p> I am sorry that I cannot help further, because this is my last day before I leave for 3 weeks of holidays. Good Luck! </p>

  @ulrichschättler in #cb8361a

<p> Ok, I see one problem: You have ie_tot=je_tot=21 and nprocx=nprocy=24! I wonder that it works at all. Because this means that you distribute the 21×21=441 grid points to 24×24=576 processors. <br/> You have to use nprocx=nprocy=1 (!), so only 1 <span class="caps"> CPU </span> . Then it should also work with less grid points. You have nboundlines=3 (what you need for the dynamics), so can try to use ie_tot=je_tot=7. This is one grid point in the interior domain and 3 grid points of boundaries around. </p> <p> I am sorry that I cannot help further, because this is my last day before I leave for 3 weeks of holidays. Good Luck! </p>
US replied

Ok, I see one problem: You have ie_tot=je_tot=21 and nprocx=nprocy=24! I wonder that it works at all. Because this means that you distribute the 21×21=441 grid points to 24×24=576 processors.
You have to use nprocx=nprocy=1 (!), so only 1 CPU . Then it should also work with less grid points. You have nboundlines=3 (what you need for the dynamics), so can try to use ie_tot=je_tot=7. This is one grid point in the interior domain and 3 grid points of boundaries around.

I am sorry that I cannot help further, because this is my last day before I leave for 3 weeks of holidays. Good Luck!

View in channel
<p> Thank you so much for your help. I will try it that way. I wish you happy holidays! </p>

  @evanowatzki in #d31ac61

<p> Thank you so much for your help. I will try it that way. I wish you happy holidays! </p>
EH replied

Thank you so much for your help. I will try it that way. I wish you happy holidays!

View in channel