Re: [ferret_users] Ensemble realization analysis

[Brockmann Patrick <Patrick.Brockmann@xxxxxx>]

Ansley Manke a écrit :
> Hi -
> I would agree; though the variables in the datsets are of type FLOAT, 
> so Ferret's use of single precision for variables isn't causing a loss 
> of accuracy in this case.
Hi,

If you plot


yes? use 
"http://dods.extra.cea.fr/cgi-bin/nph-dods/data/p86denv/IPSLCM5A/PROD/rcp45/v3.rcp45.1/ATM/Analyse/TS_MO_YE/v3.rcp45.1_20060101_21651231_1Y_t2m.nc";
yes? use 
"http://dods.extra.cea.fr/cgi-bin/nph-dods/data/p86denv/IPSLCM5A/PROD/rcp45/v3.rcp45.2/ATM/Analyse/TS_MO_YE/v3.rcp45.2_20060101_21001231_1Y_t2m.nc";
yes? plot/l=1:40 T2M[i=@ave,j=@ave,d=1]
yes? plot/l=1:40/over T2M[i=@ave,j=@ave,d=2]

to be compared to
let a=zcat(T2M[d=1],T2M[d=2])
plot/l=1:40 a[i=@ave,j=@ave,k=2], a[i=@ave,j=@ave,k=1]

You will see that this not an accuracy problem.
I would rather suspect the ZCAT function to be problematic.

Patrick
> Ansley
>
> On 4/26/2011 9:41 AM, Martin Schmidt wrote:
> > Hi Patrick,
> >
> > I can reproduce your finding. I GUESS, the i/j-average of the 
> > difference between both
> > the experiments is pretty small, even if there are significant 
> > differences in some areas. The average runs over 10^4
> > points. The quantities are of the order 200, the difference is less 
> > than +/-10.
> >
> > Hence it may be that you see simply the limited accuracy of ferret.
> >
> > I do not know nothing about the experiments, but may it be that some 
> > procedure has been applied to
> > keep the average values fixed?
> >
> > Greetings,
> > Martin
> >
> >
> >
> >
> > Brockmann Patrick wrote:
> > > Hi all,
> > >
> > > Sorry for the uncompleted previous message.
> > >
> > > When you analyze ensemble realizations
> > > (read 
> > > http://cmip-pcmdi.llnl.gov/cmip5/docs/cmip5_data_reference_syntax.pdf
> > > Ensemble menber paragraph), you would like to work with
> > > a variable over its realizations seen as a nth dimension and then 
> > > apply on it power of ferret transformations.
> > >
> > > Unfortunatelly, there are only 4 dimensions in ferret.
> > > But sometimes, you also may have the chance that the variable you 
> > > analyze
> > > has 3 dimensions and has a XZT shape.
> > >
> > > I have imagined to use the ZCAT function to concat the XYT variable 
> > > over the Z dimension
> > > and then get a XYZT variable with the Z dimension representing a 
> > > realization dimension.
> > >
> > > It seems to work since for one time step I can shade a difference 
> > > but when I plot
> > > a time evolution I get the same plot from the 2 realizations. That 
> > > is not correct.
> > >
> > > Try the attached script if interested.
> > >
> > > Any share of experience on this topic is welcome. The challenge of 
> > > course is not
> > > to physically save files.
> > >
> > > Patrick


--
LSCE/IPSL, Laboratoire CEA-CNRS-UVSQ
Data Analysis and Visualization Engineer
ICMC - IPSL Climate Modelling Centre
--