Decadal prediction: aim and approach at SMHI - Rossby Centre

Decadal predictions are increasingly interesting for decision makers that plan with a 10-30 years time horizon in mind. Since this time period is often characterized by a weaker forced climate change signal, a big effort is currently put into getting more accurate initialization of slow-varying components of the climate system in order to improve the skills of the internal variability.

Initial value problems for transient systems have limited predictability even in the hypothetical conditions of a perfect modeling system due to the growth of errors generated by inherent initial state uncertainties (Lorenz, 1963). Some quasi-oscillatory slower phenomena (like El Nino) can be skillfully predicted (regarding monthly or seasonal means) with a lead time of 6-12 months. On decadal scales where both initial conditions and external forcing contribute to generate variability that provide sources/sinks of predictability, the upper limit for the predictability is not known.

Observed phenomena that could potentially contribute to prediction skill

Decadal climate variability (DCV, arising from: internal climate variability, natural external forcing (solar and lunar cycles, volcanic activity) and anthropogenic forcing), as well as its attributed impacts have been analyzed in observational data. Coherent patterns of DCV, isolated in surface temperature (sea and land) and rainfall on land have been associated with these forcing components and in some cases physical mechanisms have been proposed, all these phenomena representing potential sources of decadal climate predictability.

Simulated DCV

State-of-the-art models are able to simulate some key features of observed DCV modes. On global and continental scales the decadal surface temperature variability in coupled models (and to a lesser degree also that for precipitation) has proven realistic.

The same is the case for the multi-decadal variations of SST in the Atlantic sector similar to the observed AMO (Atlantic Multidecadal Oscillation, hemispheric pattern of multidecadal variability in surface temperature, centered on the North Atlantic basin, associated often with fluctuations in the Atlantic meridional overturning circulation – AMOC) and for the decadal SST variations in Pacific that are simulated in strong resemblance with observed features.

Models also reproduce many of the climate impacts from these variations such as Northern Hemisphere average surface temperature or North American and Western European summertime climate variations. Some of the simulated features (strength, period) differ among the models, however commonly proposed mechanisms for DCV may be retrieved, as: interaction with internal variability on shorter time scales (as El Nino and La Nina), large-scale coupled atmosphere-ocean interactions, uncoupled atmospheric variability features traced trough ocean slower exchange components.

Although features like decadal spectral peaks are represented by models, their non-stationarity makes a challenging ongoing research on their usage as a potential source for multiyear to decadal climate predictability.

Current modeling of decadal scales

Decadal climate variations come from both: internal processes and external forcing, thus their prediction is a mixed initial-boundary conditions problem. Studies on the initial condition problem indicate that internal climate variations may be predictable on decadal timescales (mainly on the North Atlantic, North Pacific and Southern oceans: Latif, 2006) while radiative forcing have been shown to increase decadal predictability in surface temperature.

Their relative importance depends on the region and spatial scales considered. Recent studies demonstrated enhanced skill from the initialisation of upper ocean heat content (shown to bring enhanced skill on global scale) and from the initialisation of the dynamical component such as the Atlantic MOC (that increases the skill in predicting North Atlantic SST). Differences between various modeling systems skills come from different initialisation techniques, the effects from the minimization of systematic model biases, and from the representation of various uncertainties (of the initial state, modeling and forcing).

Anomaly initialisation

The actual recognition of decadal prediction importance has lead to an increased effort in the research community. Phase 5 of the Climate Model Intercomparison Project (CMIP5) includes experiments on decadal prediction and potential predictability.

A series of 10-year hindcast experiments are to be produced in order to investigate the theoretical limits of decadal predictability, the actual modeling performance and the role of initialisation. In addition to these hindcasts, an actual 10- and 30-year prediction for the coming decade(s) will also be performed, including the effects of external forcing from greenhouse gases and aerosols.

Modeling groups will use different models and initialisation techniques in order to allow for broad analysis and intercomparison. The Rossby Centre contributes to the decadal prediction experiments with the EC-EARTH model (Hazeleger et al, 2010). The research related to decadal predictions is funded by the EU FP7 project COMBINE.

Figs. 1-4 show an example of the results obtained with the EC-EARTH model with anomaly initialization for the ocean and sea-ice. This technique implies models initialisation with observed anomalies added to model climate in order to avoid model drift towards its climatological state. However differences in climates need accurate specification as they may contribute to biases resulted from both: linear and non-linear interactions between mean and perturbed states. Also matching different subsystem’s climate is an important issue in representing a correct initial phase interaction.

A more realistic representation of initial state uncertainties will be achieved through ensemble simulations that start from slightly modified initial states. The ensembles methods used may spread from simple (e.g. perturbing ocean initial state) to more sophisticate (e.g. identify perturbations that amplify fastest and lead to an optimal error growth).

Decadal prediction, RC news 2, 2010
Fig1: Decadal annual mean SST anomalies to climate, scaled by its standard deviation over the 10 years (1991-2000): model simulated anomalies (green) versus ERA anomalies (pink), averaged over the North Atlantic area, represented for years 1992-2000 (simulated year 2 to 10 on Ox axis). Fig2: same as Fig.1, but averages are over the global domain Fig.3. and Fig.4: same as respectively Fig1 and Fig2 but for monthly mean anomalies, running mean averaged over 3 months, for years 1 to 10 of simulation (1991-2000). Enlarge Image


Decadal prediction attempts need to be carefully tested in hindcast studies. The task is challenging due to many reasons: data scarcity and time series length, natural variability representation, forced signal separation from background signal attributable to internal variability, etc.

Although models are continuously improved and societal needs push for more accurate climate predictions for the near future, decadal predictions are still in its infancy and a common research effort is needed in order to provide valuable results.


Hurrell J. W. et al. “Decadal climate prediction: Opportunities and Challenges” Community white paper, sept.2009.

Latif M, Collins M, Pohlmann H, Keenlyside N (2006) A Review of Predictability Studies of Atlantic Sector Climate on Decadal Timescales.J Clim 19 (23): 5971-5987, doi:10.1175/JCLI3945.1

Lorenz E. N. , 1963: "Deterministic nonperiodic flow", Journal of Atmospheric Science 20, 131-140.

Meehl G. A. et al. “Decadal prediction: Can it be skillful ?” BAMS, AMS, Oct.2009, 1467-1485.