Opened 10 months ago

Closed 4 months ago

#3119 closed help (fixed)

perturbed ensemble produces same results for each perturbation

Reported by: xd904476 Owned by: um_support
Component: UM Model Keywords:
Cc: Platform:
UM Version:

Description

Hi,
I am running a perturbed ensemble (suite u-bq215), but when I look at the results they are all the same.
I use a perturbation of the atmosphere and when I look at the perturbed files produced they are not the same (/share/data on archer).
This suggests to me that the problem is in the initial conditions that the coupled task uses. Or is this setup in the recon task?
any suggestions?
thanks,
dani

Change History (10)

comment:1 Changed 10 months ago by dcase

Dani,

if I could check this- you want to perturb the field number 236 (tile modified infiltration rate), and you claim that after the perturbation you can see a change here? I can't see a change in this field, and I think that it is because you have turned on the -r option for reproducibility. This takes the seed from the hash of the input file, but all input files to the perturbation script are the same (so the hash of their name will be).

Could you point me to the place where you see differences in the input files?
Let me know if I've gone down the wrong path here.

Dave

comment:2 Changed 10 months ago by xd904476

Hi Dave,
I have run the model with the ensemble more than one times and I have also changed the code so that the if control is practically switched off (both conditions are without the hash). (u-bq215,u-bq336,u-bq339).
I thought that the -r was getting me in a position to be able to reproduce a certain perturbation.
When I changed the field to be changed the model was crashing when producing the perturbation.
I have run manually 50 suites with the same perturb_ini.py file changing the stash 2 (potential air temp), and it works.
when I do a cmp in the start files in share data I get this kind of answer:

dflocco@eslogin008:~/cylc-run/u-bq336/share/data> cmp bq336.astart_0 bq336.astart_1
bq336.astart_0 bq336.astart_1 differ: byte 221916496, line 411089

therefore I thought the perturbation was working.
perhaps I should try with a different field?

thanks,
dani

comment:3 Changed 10 months ago by dcase

When I ran this suite, I didn't use reproducible random numbers. If you want to do this then you can use the hash of the file name, but this is just a way to make the same random number if you re-ran the suite. Also, if you are using one input file to make many different perturbed files, then you can't use hash([input file name]) or you'll make the same perturbation.

If you are happy with turning the -r option off then that is easiest (and what I did)- it's just that if you repeat the study you'll have slightly different answers.
As for the field to perturb, that is a scientific choice that you choose for your particular study, and I know too little to say anything.
Finally, if you are comparing numbers in um files, I would use a program which is designed to read these, rather than a general unix command. xconv should allow you to read the files.

comment:4 Changed 10 months ago by xd904476

Hi Dave,
I'll try again. I have already run the suite in this way.
I have used xconv to look at the data and compared the ice concentration results from the daily outputs, and these were the ones which were the same.
I have now changed the variable to convert to 17 (from xconv this should be the surface air temperture field, do u agree?).
The suite is running the coupled task at the moment. I'll try looking at the astart files to check on variable 17.
thanks

comment:5 Changed 10 months ago by dcase

The stash numbers are the things that appear in the stash master and so on. 17 is SILHOUETTE OROGRAPHIC ROUGHNESS. This webpage has a first start (http://puma.nerc.ac.uk/STASH_to_CF/STASH_to_CF.html) but look at where your stash is defined for the definitive definition at your version of the UM.

comment:6 Changed 10 months ago by xd904476

Hi Dave, I've run the suite again perturbing the same stash that I perturb when running 2 suites manually, but I keep getting the same results for the ice concentration during the first month.
\i guess I'll show you some figures next week.

thanks,
dani

comment:7 Changed 10 months ago by xd904476

Hi Dave, I've run the suite again perturbing the same stash that I perturb when running 2 suites manually, but I keep getting the same results for the ice concentration during the first month.
\i guess I'll show you some figures next week.

thanks,
dani

comment:8 Changed 10 months ago by xd904476

  • Resolution set to fixed
  • Status changed from new to closed

The recon task was overwriting the ASTART
It needs to be

astart='${ASTART}'
in the um app rose-app.conf file

Solved.
Thanks

comment:9 Changed 8 months ago by xd904476

  • Resolution fixed deleted
  • Status changed from closed to reopened

Hi Dave, I am having again the same issue with the suite ensembles producing the same results. The astart in the um app rose-app.conf is set as
astart='${ASTART}'
the initial conditions are modified (checked with mule-cumf). I believe that this problem has come up more or less when the RDF went down. I am not sure whether it has anything to do with the change of paths for the archiving though.
I am working with suite u-bq339 and the results of the ensembles are producing the same results. I believe it may have something to do with the recon task that overwrites the results, but I don't know how to fix it.
Could you help please?
thanks,
dani

comment:10 Changed 4 months ago by ros

  • Resolution set to fixed
  • Status changed from reopened to closed

Resolved offline.

Note: See TracTickets for help on using tickets.