Hi Simon,

Postproc may give this error if you are writing compressed netcdf format (byte*2), which calculates scale_factor and add_offset based on min and max of particular variable. Now, this error may not necessarily come from RegCM output. There are many variables, which are calculated in the post-processing step. Sometimes, I have seen this out of range error for snow water equivalent (swe) variable, which comes from RegCM3 directly.  Also, recently one of my colleague found out of range error for divergence and vorticity, which are calculated in postproc. 

I do not have any smart way of defining min and max for all variables which work for everyone all the time over every domain. The best practice is what you do i.e. adjust the range if out of range error is encountered. However, one has to be very careful in the choice of min and max, because the values of variables in the compressed netcdf format largely depend on them. Anyway, it is quite fortunate that I have never seen this error for the variables which can not be derived from other variables except for swe. 

Anyway, in short, I would recommend that if anyone is using compressed netcdf output option and gets out of range error, he should be careful in the choice of min and max for the problematic variable. A comparison between byte*2 and byte*4 netcdf output would be useful to check that the newly adjusted  min and max values are not creating difference between compressed and uncompressed netcdf format at the order of more than 10E-2 for that particular variable.

Meanwhile, I am also trying to get bit smarter to resolve this postproc issue in more useful way.

Moet

 


On Jan 29, 2010, at 12:29 AM, Simon Krichak wrote:

Dear All,
 
I have an additional issue for discussion. In some cases my simulations are completed successfully without any Nan error. And the  Postproc produces reasonable monthly averaged files. But the same code prints the following ,'Values Out of Range: FIELD=',vnambc(ni)when I am running the code with daily averaging. Of course I always can  change the min max limits in the code. Still the diagnostics apparently means that the model produced results are not physically reasonable.
What are the REGCNET recommendations for such situations?
 
Best 
 
Simon Krichak
 
----- Original Message -----
Sent: Friday, January 29, 2010 7:20 AM
Subject: [RegCNET] Nan error during model run-Common problem


Dear Sara and All,
Thanks for your mail and detail information about the Nan error in the RegCM3 model run.
As this Nan error is a common problem in RegCM, I have some point in continuation with this discussion.
Mainly, even though model is encountering Nan error during the integration still it proceeds up to completion of the integration. For long simulation people don't monitor continuously whether model running smoothly or there is a chance of nan error.
Sometimes, people see that they have completed their simulation successfully but at the time of analyse their model output they realize that there is Nan error and the model output is no use.
So they need to start or restart the simulation again.
In my opinion it will better if there will be facility such that, In case there is a Nan error during the run it should come out immediately and stop the integration automatically without proceed further.
In this way we can save lot of time.

Thanking you.

Best regards
Satyaban

On Thu, 28 Jan 2010 22:24:41 +0530 wrote
>Dear Dwijenra,

This is the most common question on the list (in fact, it was answered 
the other day). Everyone, please read the user's guide! It contains 
valuable information that will help you run your simulations.
http://users.ictp.it/RegCNET/regcm.pdf

The model is sensitive to the time step used. The first rule of thumb is 
to use 3 times the horizontal grid spacing. In your case, 60 km, so 180 
is the best to start with. However, in cases of steep topography you may 
need to decrease this time step. You might want to try 150, for example.

If you are doing a long simulation, and this just happens once or twice, 
you can work around this. Set the time step to a lower value, restart 
the model to a point before where it blew up, and then after this point, 
stop the model, put the time step back to its original value, and 
restart the model. So for example, if my model had died on January 5th, 
I would probably restart at the beginning of January, do that month at a 
lower time step, and then continue the simulation in February at the 
original time step if that's been working. Note that this is acceptable 
to do once or twice in a long simulation (like > 5 years), but if it 
happens more often, you should restart the simulation from the beginning 
at a shorter time step.

In the regcm user guide, there is a table that gives suggestions (these 
are just suggestions) for time steps for different grid spacing (Table 
12, page 40):
Table 12: Time steps with different resolutions:

dx(km) dt(sec) abatm(sec) abemh(hr) radfrq(min)
10 30 90 18 30
20 60 120 18 30
30 100 300 18 30
45 150 300 18 30
50 150 450 18 30
60 200 600 18 30
90 225 900 18 30

There is discussion of how to set the time step on page 35 in the user's 
guide.
cheers
sara


d d wrote:
> Dear RegCNET Users,
>
> I am having similar problem as Dr. Satyaban getting the NaN error 
> while running RegCM for 6 months simulation. The ICBC files are OK 
> when i checked them on grads. The Model can run successfully upto 4 
> months and after that it encounters NaN error for last two months.
> I had tried to restart the simulation but still the same
> domain.param configuration
> Idate1=1997010100 start day of simulation
> Idate2=1997063000 End day of simulation
> ds = 60km
>
> regcm.in configuration
> Idate0=1997010100
> Idate1=1997010100
> Idate2=1997063000
> dt=180
> radfrq=30
> abemh=18
> abatm=540
>
> please kindly suggests whar might be the problem
>
> Best regards
> Dwijendra
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> RegCNET mailing list
> RegCNET@lists.ictp.it
> https://lists.ictp.it/mailman/listinfo/regcnet


-- 
Sara A. Rauscher
T-3 Fluid Dynamics
MS B216
Los Alamos National Lab
Los Alamos, NM 87545 USA
(505) 606-0512

_______________________________________________
RegCNET mailing list
RegCNET@lists.ictp.it
https://lists.ictp.it/mailman/listinfo/regcnet


-------------------------------------------------------------------------------
Dr. Satyaban Bishoyi Ratna
Computational Atmospheric Sciences Team
Scientific & Engineering Computing Group
Centre for Development of Advanced Computing (C-DAC)
Pune University Campus, Ganeshkhind
Pune - 411 007, Maharashtra, India
E-mail: satyabanb@cdac.in
Tel (O): +91-020-2570 4226, (Cell): +91-094223 56326
--------------------------------------------------------------------------------


_______________________________________________
RegCNET mailing list
RegCNET@lists.ictp.it
https://lists.ictp.it/mailman/listinfo/regcnet



No virus found in this incoming message.
Checked by AVG - www.avg.com 
Version: 9.0.733 / Virus Database: 271.1.1/2653 - Release Date: 01/28/10 16:55:00
_______________________________________________
RegCNET mailing list
RegCNET@lists.ictp.it
https://lists.ictp.it/mailman/listinfo/regcnet