Detail on design of short term experiments

1
Detail on design of short term experiments

• James Murphy
• WGCM Paris, Sept 2008

Following Aspen Global Change Institute meeting,
June 2008 on Climate Prediction to 2030 Is
it possible, what are the scientific issues, and
how would these predictions be used ? Co-chairs
Jerry Meehl, Ron Stouffer, Lisa Goddard, James
Murphy
2
Basic experiments (1.1)

• 10 year integrations from 1960, 1965, 1970,,
  2000, 2005
• 3 ensemble members, optionally O(10)
• Initialise ocean based on some representation of
  either observed anomalies, or full observed
  fields
• Observed forcings for past dates, as CMIP5
• Recommend RCP4.5 for future dates
• A single scenario for anthropogenic aerosol
• Volcanoes Observed values for past dates (e.g.
  hindcast from 1990 assumes knowledge of Pinatubo
  in 1991)
• Assume no further eruptions after 2005

3
Details on 1.1

• Initial dates towards the end of the year (e.g.
  1st Nov), at convenience of modelling group
• Runs long enough to give 10 complete
  hindcast/forecast years, 1961-70, 1966-75,.. etc
• Choice of ocean initial conditions up to each
  group, e.g. use their own dedicated
  analysis/assimilation technique, or use one of a
  number of available ocean analyses transferred
  into their ocean model
• Initialisation (or not..) of land, sea-ice and
  atmosphere also up to each group.
• 180 model years (420 years if 10 member ensembles)

4
Basic experiments (1.2)

• Extend initialised integrations from 1960, 1980,
  2005 to 30 years
• 3 ensemble members, optionally O(10)
• 180 (optionally 420) model years

• Note CMIP5 proposal (Table 8) AMIP runs with
  high resolution and/or more complex chemistry.
• These propose to simulate 2030-2040 using
  prescribed changes in SST/sea-ice, and recommend
  RCP4.5.
• cf plan for near-term initialised projection from
  2005, which also recommends RCP4.5, finish in
  2035.
• Coordinate to enable grand ensemble of
  projections for 2030s with different value-added
  characteristics ?

5
Requested experiments (1.3)

• 10 year integrations from 2001, 2002, 2003, 2004,
  2006.
• Further runs from 2007, encouraged where
  possible
• 3 ensemble members, optionally O(10)
• Take advantage of improved ocean data coverage
• Step towards real-time prediction
• 150 model years (optionally 350 years)

6
Requested experiments (1.4)

• Parallel no observed initial data runs from
  1960-2035, but with same external forcing as
  1.1/1.2, to allow assessment of impact of
  initialisation.
• Obvious method is to do additional CMIP5 20th
  century/RCP4.5 runs, for groups who can.
• CMIP5 plan asks for one mandatory 20th
  century/RCP4.5 run, with additional ensemble
  members for the historical period recommended
• Additional CMIP5 runs should ideally have
  statistically independent realisations of
  internal variability during 1960-2035, e.g. by
  starting well before 1960.

7
Experiments (1.4) contd..

• For groups lacking basic CMIP5 runs (notably a
  20th century run with historical forcings)
• Suggest a 100 year control simulation, and a 70
  year run with 1 per year increase in CO2.
• Allows evaluation of model drift, TCR, ocean heat
  uptake, natural variability characteristics
• But how to design this to also support
  initialisation of near-term projections with
  observed anomalies ?
• Could (e.g.) run control for 100 years from
  Levitus, with constant 1950 anthropogenic forcing
• Then run from 1950-2000 with obs forcings (in
  addition to 1 CO2 run), starting from the end of
  the control run.
• Gives a 1950-2000 model climate, cf obs 1950-2000
  for definition of initial condition anomalies for
  assimilation

8
Optional no volcanoes addition to experiments
(1.1, 1,2 and 1.4)

• Repeat hindcasts from 1960, 1980 and 1990 without
  Agung, El Chichon, Pinatubo.
• Would support no cheating estimates of past
  skill consistent with the no volcano assumption
  made in future forecasts.
• EU ENSEMBLES project has produced decadal
  hindcasts from 1960, 1965,.., 2005 (using A1B for
  future forcing), done with no volcanoes after
  initialisation.
• Also suggest additional no volcano, no obs
  initial data 20th century runs, started from
  existing CMIP5 historical runs at 1960, 1980,
  1990.

9
Further optional studies

• Comparison of initialisation strategies e.g.
  repeat 1.1/1.2 using different ocean analyses, or
  a different initialisation method
• Repeat forecast from 2005 with high and/or low
  anthropogenic aerosol scenario
• Repeat forecast from 2005 with an imposed
  Pinatubo eruption in 2010
• Impacts of interactive ozone chemistry
• Air quality (ozone and air quality studies
  coordinated with chemistry intercomparison
  activities)

10
Outputs/Data Handling

• needs to be defined.
• Initial assumption is that the output will be the
  same as for the main CMIP5 runs.
• From what subset of the near-term runs will data
  be archived as part of the general AR5 archive ?
  Section 1.1 ? More ?