Unipost strange high, mid, low cloud cover output

All other topics about postprocessing model data (GrADS and other software), about other numerical weather prediction software (including WRF-NMM and WRF-ARW discussion unrelated to UEMS/WRF EMS), and general meteorology talk go in this forum.
surgeon
Posts: 46
Joined: Wed Dec 08, 2010 1:34 pm
Location: Poland
Contact:

Re: Unipost strange high, mid, low cloud cover output

Post by surgeon » Wed Oct 22, 2014 10:14 am

Try without mpiexec.
Is there any reason to compile WRF in dmpar if we use only one computer?

meteoadriatic
Posts: 1588
Joined: Wed Aug 19, 2009 10:05 am

Re: Unipost strange high, mid, low cloud cover output

Post by meteoadriatic » Wed Oct 22, 2014 11:49 am

surgeon wrote:Is there any reason to compile WRF in dmpar if we use only one computer?
It is faster. wrfelp wrote about that somewhere.

And no difference with one thread only? In fact I have similar (but not same) issues with cloud cover when run UPP on multiple threads for ARW output.

meteoadriatic
Posts: 1588
Joined: Wed Aug 19, 2009 10:05 am

Re: Unipost strange high, mid, low cloud cover output

Post by meteoadriatic » Wed Oct 22, 2014 12:12 pm

This is the issue I have with ARW postprocessing in UPP when I use more than one thread:

Image

I could not correct that and it is present in v2.1 and v2.2. However I just drop unipost to one thread and issue is gone. Still this looks much different than your issue, this is why I don't think it is directly related.

oib
Posts: 118
Joined: Mon Jan 31, 2011 3:29 pm

Re: Unipost strange high, mid, low cloud cover output

Post by oib » Wed Oct 22, 2014 1:10 pm

S..ooOOLVED! :D

Yesterday morning I must have looked distractedly the results obtained by performing unipost onto a single core. :roll:

In fact, warned by your post, meteoadriatic, a few minutres ago, I wanted to try again running unipost.exe (dmpar compiled) on one only core.

And I saw the problem disappeared :P

Now the grid values in GRIB output file from unipost.exe for the following fields looks correct:
+ Total Cloud Cover [%] (3D field on pressure levels)
+ Low Cloud Cover [%]
+ Middle Cloud Cover [%]
+ High Cloud Cover [%]

Now I ask, for ARW running purpose, should exists any reason to re-compile UPP in "Linux x86_64, Intel compiler (serial)" mode? Or I can stay with my dmpar and execute it on one core?

Note: as said previously, for ARW {Low, Mid, High} Cloud Cover calculated by unipost do not look so great (detailed) as the NMM ones.

-----------

Finally I would like to again thank warmly meteoadriatic for the competent and quick help and ideas that he provided to me about that problem along last two weeks.
It would be great for the world of weather modelling if the official WRF forum would work just as well as wrfems.info!

meteoadriatic
Posts: 1588
Joined: Wed Aug 19, 2009 10:05 am

Re: Unipost strange high, mid, low cloud cover output

Post by meteoadriatic » Wed Oct 22, 2014 2:13 pm

Oh, then we have same issue in fact even though looks is bit different, probably because you used much more threads than I did :mrgreen: I don't know now is this good or bad news? :twisted: Why? Because on one core unipost is much slower. So it would be better to try to solve original issue, instead of going around it. But I was not successful.

How about others who read this topic? Do you have same problem on multithreaded unipost?

In fact I tried in serial mode but then there is no performance advantage over running dpmar on one thread only. So basically I can't get at same time multithreaded unipost run and no cloud cover problems. Either one or another, both impossible.

About cloud output being not detailed as in NMM I guess that source code changes are necessary to improve that.

oib
Posts: 118
Joined: Mon Jan 31, 2011 3:29 pm

Re: Unipost strange high, mid, low cloud cover output

Post by oib » Thu Oct 23, 2014 7:00 am

meteoadriatic wrote:Oh, then we have same issue in fact even though looks is bit different, probably because you used much more threads than I did :mrgreen:
I have done some test executing running unipost.exe with different number of cores.
And I have noticed that the cloud cover plot is like it's calculated splitting the domain into horizontal tiles over the available cores.
The parallel calculation must have some problem and the resulting tiles do not perfectly match each other.
So the resulting image looks segmented.
As you can see the horizontal bands are as many as the number of cores assigned to parallel execution of unipost.exe.

For example this is from the command

Code: Select all

mpiexec -n 2 unipost.exe
Image
Note that the cloud cover over the domain looks like it is divided in two horizontal bands and I have used two cores.

This instead is from

Code: Select all

mpiexec -n 12 unipost.exe
Image
And as you can verify the image looks like segmented in 12 horizontal bands.

Maybe this might help us understand how to fix it...

meteoadriatic
Posts: 1588
Joined: Wed Aug 19, 2009 10:05 am

Re: Unipost strange high, mid, low cloud cover output

Post by meteoadriatic » Thu Oct 23, 2014 7:18 am

I also noticed that before. Was not much helpful to me. But I can't understand why this happen only on cloud cover, but not on other fields. This is why it is so strange.

oib
Posts: 118
Joined: Mon Jan 31, 2011 3:29 pm

Re: Unipost strange high, mid, low cloud cover output

Post by oib » Thu Oct 23, 2014 7:48 am

meteoadriatic wrote:I also noticed that before. Was not much helpful to me. But I can't understand why this happen only on cloud cover, but not on other fields. This is why it is so strange.
Very strange, sure...

In any case, I have just submitted a bug report to the WRF user support (wrfhelp@ucar.edu).

Antonix
Posts: 260
Joined: Fri Oct 16, 2009 8:53 am

Re: Unipost strange high, mid, low cloud cover output

Post by Antonix » Thu Oct 23, 2014 11:19 am

1)you have the same problem in precipitation?
2)you use numtil>1???
3)what is your cumulus parameterization?

I had a similar problem with
numtile > numcore for sigle cpu
cu = KF

meteoadriatic
Posts: 1588
Joined: Wed Aug 19, 2009 10:05 am

Re: Unipost strange high, mid, low cloud cover output

Post by meteoadriatic » Thu Oct 23, 2014 12:05 pm

Hello Antonix,

should not be a problem because, if we can get OK result with one UPP thread, then it is certanly not any issue with WRF but with UPP. But to answer questions:
1) No,
2) Yes,
3) depends on domain, in particular example here, none

Post Reply