Setup Integration

For the "paleo" CSM 1.4 version, we have obtained an initial condition for the year 1900. This state was provided by Caspar Ammann from his historical integrations. From this state, we integrate the coupled model 40 years, prescribing the time dependant atmospheric composition, solar irradiance and sulphate and volcanic aerosols, kindly provided by Caspar Ammann. The state in 1940 is randomly perturbed and used as an initial condition for 62 parallel integrations, each differing only in the atmospheric initial condition. The initial state of the other parts of the climate system is not perturbed. Historical estimates of atmospheric composition, solar irradiance and sulphate and volcanic aerosols are used upto the year 2000. From 2000 until the year 2080, only the atmospheric composition changes according to the BAU (Business As Usual) scenario of NCAR, which is similar to the SRES A1b scenario.

In the DCP meetings on february 7 and march 21 we had a discussion about the type of integration we should do:

1. 50 members of 1 long continuous integration of 140 years
2. 1 long continuous "control" integration of 140 year and 50 members of 2 integrations of 40 years starting in 1960 and 2070

After a lot of brainstorming everybody agreed to do 1 long continuous integration. And instead of 50 members, the Teras supercomputer enables us to do 62. (option A).

pros

• This setup is a true representation of the prediction problem in a perfect model world, i.e: given a perfect model, a perfect initial condition, a perfect knowledge of the future climate forcing, which changes in rare extreme conditions can we expect in the future.
• Conceptually the simplest and most elegant
• It is possible to get enough realisations of 140 year (>50), which means that it is possible to assess changes in the Probability Density Function (PDF) in time.
• Output can be used for impact studies which require continuous time series
• Makes an assessment of ocean variability on long timescales possible and also detection of changes in variability on decadal and shorter timescales (e.g. changes in the timescale and intensity of ENSO)

cons

• Long integrations are expensive; decreases the number of members and with it the extent to which rare events can be studied.
• In the beginning of the ensemble the expected range of ocean states is smaller than later on in the integration because of the presence of variability on long timescales (10-1000 year) that is not related to changes in the forcing on that timescale but that also appear if the GHG concentrations are held constant (= internal variability). This internal variability possibly also affects the PDFs that we are going to look at (wind- temperature and precipitation above land and sea). Therefore the possible changes in the PDF that we will find do not have to be due to changes in the GHG concentrations.
  

  Solution:
  Start the ensemble sooner in the 20th century. It is a safe assumption that the influence of the initial condition gets smaller during the course of the integration. If need be, we could decide NOT to go so far into the 21th century. E.g. 1940-2080 ? We think this is solution is better than another possibility, which is to perturb the ocean initial state as well to increase the spread early on in the ensemble. We decided not to do this, because it is not at all clear how to perturb the ocean state. These initial perturbations might realise ocean states that never occur in an unperturbed integration and might affect the ocean evolution over decades, whereas the atmosphere forgets its initial condition in a couple of weeks. Discarding the first year in the analysis of the ensemble discards states that are realised only due to the perturbations added to the initial atmospheric condition.

For the sake of completeness we also list the pros and cons of integration type B.

B. 1 long continuous "control" integration of 140 year and 50 members of 2 integrations of 40 years starting in 1960 and 2060

pros

• The setup of both integrations is identical so this makes comparing both climates straightforward.
• Makes it possible to increase the number of realisations in the computation time available which enables the study of events that are more rare and extreme.
• Makes it easier to reintegrate individual members to study extreme events in more detail afterwards.

cons

• There is no information about changes in the PDF in time (perhaps the most interesting changes occur in the period in between the two periods chosen)
• Not all future possible states are sampled (the second initial condition of the ocean can be a rather rare extreme state)
• There is no information about low-frequent changes in the climate system.
• Again it is not possible to distinguish the influence of the initial condition of the ocean from the influence of the time-dependant external forcing on the changes in the PDF.
• The dataset is less generally applicable, for instance for applications in which continuous timeseries are needed.

Return to the Dutch Challenge Project main page