[RegCNET] Re: Sigma Levels
Pal Jeremy
jpal at ictp.it
Fri Oct 28 17:28:40 CEST 2005
Hi Neil,
It should be fine to leave the BATS timestep at 600. You could also try
dt=180 and abatm=540.
Jeremy
Neil Davis wrote:
> I tried this suggestion by changing the timesteps to those listed in
> the manual as being good for a dx of 50 km, So I changed dt from 200
> to 150 and abatm from 600 to 450 and now the job is running on the
> larger domain. Thanks everyone for the suggestions and help. I am
> wondering if there is an intermediate change that I could attempt
> however just to save some computational time. I wasn't sure if the
> timestep values had to be set to certain values or if I could change
> them some as long as abatm is 3 tims as large as dt.
> Neil Davis
>
> Solmon Fabien wrote:
>
>>
>> Hi Moet !
>>
>> Did you try to play also with the time step on the big domain ? the
>> vertical advection is sensitive to sigma level change, I would guess
>> that if you have a high vertical velocity in a part of the domain
>> you need a smaller time step if you reduce the grid step .... cheers
>>
>> fabien
>> moetasim wrote:
>>
>>> Hi Jeremy , Bi n All,
>>>
>>> Neil and I are trying to figure out the reason why he has floating
>>> point exception (FPE) while running RegCM with his modified sigma
>>> levels. What we got finally is that regcm works fine with his
>>> modified levels at a relatively smaller domain than the one which
>>> ends up with FPE. I had actually tried to run regcm with his
>>> modified levels at a smaller domain and found that it works fine.
>>> Same kind of result he got when he ran the model with his modified
>>> levels but, smaller domain. I can't really say anything with
>>> confidence that why is it happening? He has domain size = 74 x 136 x
>>> 23 for the simulation which ends up with FPE. The same domain works
>>> fine if he runs the model without modifying sigma levels.
>>> Is it happening because his modified levels are more concentrated in
>>> low level (say in PBL) which is making it more computationally tough
>>> for PBL model or what else? There is no reason other than
>>> computationally incompatibility to end up with a FPE for larger
>>> domain + modified sigma levels, if model works fine for both small
>>> domain + modified sigma levels, and larger domain + default sigma
>>> levels (its my point of view).
>>>
>>> (I have attached the gradual progress to this conclusion here under)
>>> MOet
>>>
>>>
>>> Quoting Neil Davis <nndavis at unity.ncsu.edu>:
>>>
>>>
>>>
>>>> I tried running the 50 X50X23 and that seems to be working, it has
>>>> run through the first month already so I am assuming that that
>>>> works. I am just wondering why the default 23 sigma levels would
>>>> run on my large domain, when the proposed ones do not. Is there
>>>> something with the stacking in the lower levels that adds computer
>>>> memory or something that I am missing.
>>>> Thank you again for all your help on this problem,
>>>> Neil Davis
>>>>
>>>
>>>
>>> Quoting moetasim <mashfaq at purdue.edu>:
>>>
>>>
>>>
>>>> To be on safe side just make a check with small domain like 50 x 50
>>>> x 23!
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> Quoting Neil Davis <nndavis at unity.ncsu.edu>:
>>>>
>>>>
>>>>
>>>>> How much of a decrease would you suggest? I did run the identical
>>>>> domain with the default 23 vertical levels. I am running the
>>>>> model on an IBM blade cluster running redhat linux. Glad to hear
>>>>> that the model is running somewhere with my proposed sigma
>>>>> levels. I will try your suggestion of a smaller domain and see if
>>>>> that fixes it. Neil
>>>>>
>>>>
>>>
>>> Quoting moetasim <mashfaq at purdue.edu>:
>>>
>>>
>>>
>>>> Hi,
>>>> I have an early conclusion for your problem, now. It seems that
>>>> increase
>>>
>>> in
>>>
>>>> vertical resolution and considerably large domain (that you have
>>>> for your experiment) is creating the crash. You may prove me
>>>> right/wrong by just decreasing your domain size and running the
>>>> model. I had read previously
>>>
>>> that
>>>
>>>
>>>> you mentioned RegCM working fine with its default defined 23 sigma
>>>> levels.
>>>> Did you run that experiment with same domain size as its in this
>>>> case of your modified sigma levels experiment.
>>>>
>>>> Once you are done with with your problem, I will post the final
>>>> answer at RegCNET. By the way, model is running without any problem
>>>> at my end with
>>>
>>> your
>>>
>>>
>>>> proposed sigma levels. Which machine do you have for running RegCM?
>>>>
>>>> MOet
>>>>
>>>>
>>>>
>>>> Quoting Neil Davis <nndavis at unity.ncsu.edu>:
>>>>
>>>>
>>>
>>> Quoting moetasim <mashfaq at purdue.edu>:
>>>
>>>
>>>
>>>> I have started running RegcM with your defined 23 sigma levels..At
>>>> the
>>>
>>> moment
>>>
>>>
>>>> it running..lets see hw it goes..
>>>>
>>>>
>>>
>>> Quoting moetasim <mashfaq at purdue.edu>:
>>>
>>>
>>>
>>>> Yeah just make sure that every thing is fine for 3D variables
>>>> written in
>>>
>>> ICBC
>>>
>>>
>>>> files.
>>>>
>>>>
>>>
>>> Quoting Neil Davis <nndavis at unity.ncsu.edu>:
>>>
>>>
>>>
>>>> I have attached the output and the error log of the model run.
>>>> This time it didn't seem to even start a few days. I will try your
>>>> other suggestions now, although what did you mean by check icbc at
>>>> different sigma levels, just look at the ICBC files and see if the
>>>> values look reasonable?
>>>> Thanks again for your help
>>>> Neil
>>>>
>>>>
>>>>
>>>>
>>>
>>>
>>> Quoting moetasim <mashfaq at purdue.edu>:
>>>
>>>
>>>
>>>> Ok, can you send me the output that you get after running regcm.x
>>>> (start
>>>
>>> to
>>>
>>>> the point where FPE occurs).
>>>>
>>>> and for the time being:
>>>>
>>>> 1) check icbc at differnt sigma levels. If everything is fine there
>>>> and
>>>
>>> tell
>>>
>>>
>>>> me.
>>>>
>>>> 2)Also, try to run the model with reduced time step and let me know
>>>> what happens.
>>>> But before this send the ./regcm.x output.
>>>>
>>>> MOet
>>>> Quoting Neil Davis <nndavis at unity.ncsu.edu>:
>>>>
>>>>
>>>>
>>>>> Sorry, the exception takes place when running the model. It goes
>>>>> a few time steps then the error occurs. ICBC and Terrain seem to
>>>>> have no errors when I run them.
>>>>>
>>>>>
>>>>>
>>>>
>>>> --
>>>>
>>>>
>>>
>>>
>>> Quoting moetasim <mashfaq at purdue.edu>:
>>>
>>>
>>>
>>>> But you didnt tell me at what step yuo have floating point exception?
>>>> Terrain, ICBC or while executing model?
>>>>
>>>
>>>
>>>
>>> Quoting Neil Davis <nndavis at unity.ncsu.edu>:
>>>
>>>
>>>
>>>> I just changed in terrain.f the lines 1423 to 1448 to this
>>>> else if(kz.eq.23) then ! MM5V3
>>>> sigma(1) = 0.0
>>>> sigma(2) = 0.05
>>>> sigma(3) = 0.1
>>>> sigma(4) = 0.16
>>>> sigma(5) = 0.23
>>>> sigma(6) = 0.31
>>>> sigma(7) = 0.39
>>>> sigma(8) = 0.47
>>>> sigma(9) = 0.55
>>>> sigma(10)= 0.61
>>>> sigma(11)= 0.66
>>>> sigma(12)= 0.69
>>>> sigma(13)= 0.72
>>>> sigma(14)= 0.75
>>>> sigma(15)= 0.78
>>>> sigma(16)= 0.81
>>>> sigma(17)= 0.84
>>>> sigma(18)= 0.87
>>>> sigma(19)= 0.9
>>>> sigma(20)= 0.93
>>>> sigma(21)= 0.96
>>>> sigma(22)= 0.98
>>>> sigma(23)= 0.99
>>>> sigma(24)= 1.0
>>>> else
>>>>
>>>> moetasim wrote:
>>>>
>>>>
>>>>> let me have a look at your changes.
>>>>>
>>>>> MOet
>>>>>
>>>>> On Tue, 25 Oct 2005 09:36:08 -0400, Neil Davis wrote
>>>>>
>>>>>
>>>>>
>>>>>> I am working with RegCM over tropical Africa and want to adjust
>>>>>> the sigma levels in the 23 level setting so that there are more
>>>>>> levels below 3000 m than in its current set up. I have been able
>>>>>> to run the model with the default 23 sigma levels but when I try
>>>>>> to change it to increased levels in the lower atmosphere I get a
>>>>>> floating point exception. I was wondering if it is at all
>>>>>> possible to change the sigma levels from the default, and if it
>>>>>> is possible what could be causing the floating point exception.
>>>>>> I can post my adjusted sigma levels if that will help.
>>>>>>
>>>>>> Neil Davis
>>>>>> NCSU climate modeling lab
>>>>>>
>>>>>>
>>>>>
>>>
>>> _______________________________________________
>>> RegCNET mailing list
>>> RegCNET at lists.ictp.it
>>> https://lists.ictp.it/mailman/listinfo/regcnet
>>>
>>>
>>>
>>
>>
>
> _______________________________________________
> RegCNET mailing list
> RegCNET at lists.ictp.it
> https://lists.ictp.it/mailman/listinfo/regcnet
More information about the RegCNET
mailing list