WRFDA and RTTOV

Setting up the 3DVAR package within WRF.

WRFDA and RTTOV

Postby hnlim » Tue Apr 27, 2010 5:25 pm

Hi

I had installed wrf da V3.2 with rttov 8.7 but am running into problem of NaN values for cost function and gradient using test data.

I had checked my rttov8.7 installations and they passed the test run given in the rttov8.7 package.

Has anyone encounter this problem?

Agnes
hnlim
 
Posts: 74
Joined: Wed Apr 16, 2008 11:51 am

Re: WRFDA and RTTOV

Postby helio » Tue Jun 08, 2010 5:17 pm

Did you place a proper coefficient package into your working directory?
coefficient 7 is the main one. You should not exclude that.

If rttov 8.7. had any interface problem with WRFDA 3.2, then you should not have gotten da_wrfvar.exe.
helio
 
Posts: 4
Joined: Tue Apr 27, 2010 11:48 pm

Re: WRFDA and RTTOV

Postby hnlim » Tue Jun 08, 2010 7:51 pm

Hi Helio,

Yes I did. The coefficient files were linked.

In fact I had resolved the issue 2 weeks ago. This is due to a variable index in the version of rttov8.7 that i had.

BTW have u carry out any experiments using AIRS data with this set of system?
I encounter other issues when assimilating AIRS.


Agnes
hnlim
 
Posts: 74
Joined: Wed Apr 16, 2008 11:51 am

Re: WRFDA and RTTOV

Postby thomas2 » Wed Jul 07, 2010 7:43 am

Hi,

can you explain the variable index issue a bit more in detail? I am also running in NaN when using AIRS data.

Thomas
thomas2
 
Posts: 75
Joined: Mon Sep 01, 2008 4:56 am

Re: WRFDA and RTTOV

Postby hnlim » Wed Jul 07, 2010 9:49 am

in line 154 in In rttov_setup.F90 file, it should be inst instead of ninst.

Agnes
hnlim
 
Posts: 74
Joined: Wed Apr 16, 2008 11:51 am

Re: WRFDA and RTTOV

Postby thomas2 » Thu Jul 08, 2010 3:02 am

Unfortunately this does not solve the NaN Problem. It seems that the residuals are NaN when using AIRS which I do not understand.....

Thomas
thomas2
 
Posts: 75
Joined: Mon Sep 01, 2008 4:56 am

Re: WRFDA and RTTOV

Postby hnlim » Thu Jul 08, 2010 11:45 am

If you are running the test case and getting this problem, then the fix will do the job. There is no AIRS data in the test case.
If you are running your own real case, then this fix will not solve it. At least in my case, when I run it with my own case study, I am getting negative cost function values. An issue which I do not have any solution either other than using CRTM instead of RTTOV.
hnlim
 
Posts: 74
Joined: Wed Apr 16, 2008 11:51 am

Re: WRFDA and RTTOV

Postby thomas2 » Thu Jul 08, 2010 2:41 pm

After using gdb ans sending a backtrace to wrfhelp, got an answer from NCAR: The problem is, that some of the libraries are not OMP thread safe, so the options are serial or dm, not sm.

Thomas
thomas2
 
Posts: 75
Joined: Mon Sep 01, 2008 4:56 am

Re: WRFDA and RTTOV

Postby hnlim » Thu Jul 08, 2010 2:48 pm

Hi Thomas,

Could you elaborate more? I am not quite following what you are saying.

Agnes
hnlim
 
Posts: 74
Joined: Wed Apr 16, 2008 11:51 am

Re: WRFDA and RTTOV

Postby thomas2 » Mon Jul 12, 2010 3:12 am

When using OpenMP lets say on a QuadCore, you have four different threads accessing the libraries multiple times at the same time . According to how the code is written, this may be not possible and leads to a segmentation fault.

But obviously this is not the only problem. Now I get a negative cost function after 2 iterations...

Thomas
thomas2
 
Posts: 75
Joined: Mon Sep 01, 2008 4:56 am

Next

Return to Installation

Who is online

Users browsing this forum: No registered users and 2 guests

cron