CT 4

Core Theme 4:  Predictability of the THC

The last decade has seen substantial developments in coupled ocean-atmosphere models, in global data assimilation systems for the ocean and the atmosphere, and in ocean observing systems. With these data assimilation systems, retrospective global analyses of the ocean and atmosphere state have been made over the past 40 years, and from these, for example, an assessment of seasonal climate predictability has been made using state-of-the-art coupled models. These developments suggest that the time is now ripe to assess decadal predictability using similar techniques, ie from ensembles of state-of-the-art coupled model integrations, initialised using realistic ocean-atmosphere analyses. Since the THC is one of the key physical processes for which there is evidence of decadal predictability, a focus on the THC would be relevant for a first phase of this type of assessment of decadal predictability. Science outputs of this core theme will feed into an overall assessment of priorities for the development of observing systems and models, required for a quasi-operational decadal THC prediction system (Deliverable D6.1.5). This will be one of the major scientific and technological outputs of THOR.


The work described in this Work Package is exploratory in nature. During the project, important techniques will be developed - most importantly the ability to create multi-model decadal-timescale ensemble forecasts from analysed ocean-atmosphere initial conditions. Spin-up and drift issues will be carefully studied. By the end of the project a fully developed system will be available for making a definitive multi-model study of decadal climate predictability (e.g. along the lines achieved by DEMETER for seasonal predictability). It is envisaged that the development of such a short-term climate prediction system will be of enormous importance for forthcoming IPCC assessment reports.

Lead: W. Hazeleger (KNMI)
Participants: ECMWF, UREAD, IFM-GEOMAR, MET-O, MPG-M, KNMI

WP 4.1 Coordinated Multi-model Ensemble Experiments to Assess the Predictability of the THC

The key scientific question to be addressed in WP4.1 is to what extent is the THC predictable on the decadal timescale using realistic ocean-atmosphere initial conditions? Supplementary questions include:

  1. What is the impact of initial conditions relative to external forcing on the skill of THC predictions?
  2. Are estimates of decadal predictability of the THC dependent, (i) i on the specific ocean-atmosphere analyses used, and (ii) i on the coupled ocean-atmosphere models used?
  3. Are ensemble-based estimates of decadal predictability of the THC dependent on the types of initial perturbations used (e.g. quasi-random vs. singular vector)?

To address these questions, two initial dates will be taken from distinct decades, here suggested 1965 and 1994 (addressing years which some analyses suggest had anomalously strong and weak North Atlantic sub-polar gyre circulation and, in addition, linking to the work under the FP6 EU ENSEMBLES project).  Oceanic initial conditions will be provided by Work Package 2. These would be interpolated onto the relevant model grids to provide suitable initial conditions. Atmospheric initial conditions would be provided either by ECMWF or by internal atmospheric-model spin up.
Coupled modelling groups would then run four 20-year 3-member ensembles from as many of these analyses as possible:

A   1965 initial conditions, observed GHGs (including aerosols) from 1965
B   1994 initial conditions, observed GHGs from 1994
C   1965 initial conditions, observed GHGs from 1994
D   1994 initial conditions, observed GHGs from 1965

By comparing A with B CT4 can gauge the overall level of THC predictability arising from both different initial conditions and different GHG forcings. By comparing A with D, and B with C, CT4 has two estimates of THC decadal predictability (arising from having different initial conditions and the same GHG forcing). By comparing A with C, and B with D, the theme has two estimates of the impact of GHG forcings (since initial conditions would be the same) on the THC.

Two types of initial perturbations are envisaged for generating ensembles:

I Analysis perturbations associated with uncertainty in atmospheric wind stress forcing and SST (provided by ECMWF)
II Oceanic singular vectors targeted onto the THC (provided by University of Reading)

Technical initialisation issues will be studied. For example, in the light of model drift, what is the best way to initialise the ocean from a set of analyses? For example, adding analysed anomalies (analysis - climatology) to some pre-existing state spun-up with climatological (e.g. Levitus) forcing might be best to try to reduce initial imbalances. Studies will be made using the ECMWF comparing anomaly coupling and standard coupling. In addition, a full set of decadal hindcasts from 1960-present, made using the HadCM3 model, will be assessed against the same set of observations used to assess the analyses in WP 2.1. These hindcasts include a set of ‘perturbed physics’ integrations, which allow the impacts on forecast skill of uncertainty in model physics to be assessed.

The relative importance of initial conditions and external forcing on the predictability of the THC will be assessed by comparing two parallel hindcast sets made with and without explicit assimilation of the observed state of the ocean and atmosphere. This will capitalise on work carried out for the EU Ensembles Project, using HadCM3 decadal hindcasts covering the period 1960 to the present day.

Lead: Wilco Hazeleger (KNMI)
Participants: ECMWF, UREAD, IFM-GEOMAR, MET-O, MPG-M, KNMI

WP 4.2 Coordinated Multi-model Ensemble Experiments to Assess the Predictability of the THC

Considerable effort has been put into observing the oceans in recent years. It is important to assess the impact these observations have on estimates of decadal predictability. Hence in this work package, decadal ensemble integrations will be made from recent initial conditions where the full set of available ocean observations is assimilated, and where the ocean is specified entirely from atmospheric forcing.


The basic scientific question addressed in WP4.2 is how strongly are decadal forecasts of the THC impacted by the ocean observing systems?  These include satellite derived remotely sensed data, hydrographic and current data collected within the Argo programme and time series observations in the energetic boundary current.
To address these questions, decadal ensemble integrations (using the models ins WP4.1 above) will be made from a very recent ocean-atmosphere analysis (initial date from 2007 for example)
I.   with the full set of ocean observations assimilated
II.  with no ocean observations assimilated
III. with observations at various sampling densities. This will provide information on the required density of Argo floats to optimise decadal THC prediction.

The decadal ensemble integrations will be evaluated in perfect-model predictability mode. That is, the difference fields from ensemble integrations initialised under I and II will be first assessed to study the impact of the ocean observing system on potential forecast skill. The timescale after which this difference is no longer significant compared with the internal ensemble variability will determine the timescale within which the initial ocean observations have a potential impact on forecast skill. Similarly, comparing III with II, one will be able to try to estimate an optimal density of Argo observations, in the sense that with densities higher than this, the extra observations have little impact on improving potential forecast skill, but with densities lower than this, potential forecast skill noticeably drops off.

Lead: Doug Smith (MET-O)
Participants: ECMWF, UREAD, IFM-GEOMAR, MET-O, MPG-M, KNMI