Issues Relating to Transient Ensembles

At the third GCEP meeting (19 Feb 2007) it was decided that we should run a set of ensemble experiments using the same experimental methodology. This involves starting ensembles from two model states which differ on some key property but preferably are otherwise fairly comparable. As an example we could choose to compare ensembles started from two model states which differ in their North Atlantic surface heat content, but neither of which have an El Nino signal. The two initial model states will be taken from one or more ensembles. To get a reasonable choice in initial model states two criteria must be met. First the ensemble(s) must have run for long enough to have diverged sufficiently and second the ensembles must have a large enough number of members to adequately sample uncertainty.

Two methods of achieving this have been suggested.

Option 1:

• Choose 3 or 4 points in the control run separated by at least 40-50 years
• For each point make a single transient run from 1860 to about 1950
• For each single run start an ensemble of about 4 members 1950-2000
• Choose the initial states from these ensembles

Option 2:

• Choose one point in the control run
• Run a single transient run from 1860 up to about 1940
• From this run start an ensemble of about 8-12 members 1940-2000
• Choose the initial states from this ensemble

-- LeonHermanson - 21 Feb 2007

UPDATE: The chosen method was option 1 with the exception that rather than running our own transient runs from 1860-1950, we took initial states from the Hadley Centre all forcings run (see ModelFormulation). At this point we have one ensemble with 5 members that has run to 1990. We had another but it was lost in the great pegasus data loss disaster of '07.