[Go-essp-tech] Expected number of variables for QC2

[martin.juckes at stfc.ac.uk]

Hello Frank,

Attached is the last QC document I have from Michael -- it specifies 4 tests for QC level 2. This appear to be very clear and unambiguous. What we need now, I think, is a way of evaluating whether a set of files satisfies these tests or not. Do all your 90 error codes relate to these 4 tests?

Cheers,
Martin

-----Original Message-----
From: Frank Toussaint [mailto:Toussaint at dkrz.de] 
Sent: 07 January 2011 16:31
To: Juckes, Martin (STFC,RAL,SSTD); Bentley, Philip
Cc: go-essp-tech at ucar.edu
Subject: Expected number of variables for QC2

Martin, Philip, et al,

indeed, the qc2 tools are somewhat more difficult.


1. The exception  codes of QC2
-----------------------------
1.1: The code list:
We try to give an overview on:
   http://www.leuchtturm-atlas.de/SCR/qc2list.html
(sorry - redmine seems not to accept pages from hand-knit tools... So I put it on my private domain... ;-) )
It is referenced in redmine, too: see the "quality pages" at
  https://redmine.dkrz.de/collaboration/projects/cmip5-qc/wiki and
  https://redmine.dkrz.de/collaboration/projects/cmip5-qc/wiki/Qc_l2
This list is supposed to contain all of the possible error codes which are about 90.

For practical use in CMIP5, errors split into  two ad hoc groups:
* flags F<m> refer to general checks, mainly on the time axis,
* others <n> refer mainly to checks on the headers and the data.
Flags are preceeded by "F", the others are preceeded by "W" or "E".
As the "W"/"E" decision depends on the tool's setup, lets simply forget these two letters of the exception key.

1.2: Comments:
We will supply a third column soon to allow for explanations and comments. Here you will find information like the one from Martina about the min/max checks this morning.

1.3: Weightings:
We will supply a fourth column to allow for weighting. Presently, we indicated the weighting of exceptions as a first guess by colors. We will need to discuss how to weight errors.


2. Actions on Exceptions
------------------------
How to react on exceptions?
To simply reject data from a certain exception level on is too easy, I'm afraid. An error like F4 (two identical time steps) is not a big problem, in case it occurs with the first or last pair of steps in a 6-hourly 1000 years run. However, if it occurs between all pairs of two consecutive time steps, you lost your time axis and will have to dump the data.

As far as different parameters are concerned: a bad error in only one variable might perhaps result in just dropping that atomicDS from the experiment - and declaring the QC-L2 checks of this experiment (which is the appropriate granularity) as passed anyway. Who is going to decide this?

So what we will need is a discussion on a sensible handling of these exceptions. Whatever pops up in the QC2-Check-code will be sent to the QCDB, including the output (.log, .txt, etc.) files. Later it will (hopefully) make its way to the quality-part of CIM for documentation.


Cheers... frank


-- 
/** Dr. Frank Toussaint   * Deutsches Klimarechenzentrum DKRZ
  * Pfitznerstr. 69        * M&D / World Data Center - Climate
  * 22761 Hamburg              * Bundesstr.45a * 20146 Hamburg
  * priv.Tel.: 040-3861 9285 * office phone: +49-40-460094-123
  * www.Leuchtturm-Atlas.de        * e-mail: Toussaint at dkrz.de */

-- 
Scanned by iCritical.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: CMIP5-AR5-QualityControl-100820.pdf
Type: application/pdf
Size: 711792 bytes
Desc: CMIP5-AR5-QualityControl-100820.pdf
Url : http://mailman.ucar.edu/pipermail/go-essp-tech/attachments/20110107/598821ca/attachment-0001.pdf