170 Mojib Latif, AxeI Timmermann, Anselm Grotzner, Christian Eckert, Reinhard Voss
anomalous surf ace stress is important to the generation of the quasi-decadal variability, the anomalous fresh water flux is important to the generation of the interdecadal variability. In both types of variability, however, the air-sea coupling
appears to be rather weak, and the resulting modes are rather strongly damped, as
can be inferred from the spectral analyses presented above.
9.5 Predictability ofinterdecadal variability in the North
Atlantic
After having discussed some aspects of the dynamics of the decadal and interdecadal variabilities in the North Atlantic, we would like to address now the question of the predictability at decadal time scales. We can distinguish three basic
cases from the above discussion (Fig. 9.17). We consider first the 'pure' stochastic
climate model. In this case, the predictability limit of oceanic quantities is given by
their persistence (autocorrelation), which ranges typically from a few months (for
surf ace quantities) to many years (for subsurface quantities). The predictability of
the atmosphere at time scales beyond the predictability limit of individual weather
phenomena, which is of the order of about fourteen days, is not affected, since no
feedback from the ocean to the atmosphere is considered (Fig. 9.17a). The second
case is the stochastically forced 'ocean-only' mode. Enhanced predictability relative
to persistence is found for oceanic quantities around the resonance frequency,
while atmospheric quantities cannot be predicted with higher skill relative to the
'pure' stochastic climate model (Fig. 9.17b). Thus, if an oceanic mode exists with a
decadal period, the variations in the ocean associated with this mode will be predictable at decadal time scales. Finally, third, we consider the stochastically forced
coupled ocean-atmosphere mode. Both atmospheric and oceanic quantities exhibit
enhanced predictability (again relative to persistence) at the resonance frequency
(Fig. 9.17c). An example ofthis latter case is the ENSO phenomenon, for which
both oceanic (e.g. eastem equatorial SST anomalies) and atmospheric quantities (e.
g. the Southem Oscillation Index) are predictable at lead times of about one year
(for a recent review on ENSO predictability see e.g. Latif et al. (1998»). If coupled
modes with decadal periods exist in the North Atlantic, both the atmosphere and
the ocean will exhibit enhanced predictability at decadal time scales in the North
Atlantic region.
We have described two such modes from our coupled model simulation, and it
will depend critically on the degree of damping, whether the coupled nature of the
mode will affect significantly the predictability of the atmosphere at decadal time
scales.
Thus, the predictability characteristics of the low-frequency variability in the
North Atlantic can provide important informations about its underlying dynamics.
We conducted therefore an ensemble of predictability experiments with our coupled model. Before we describe the results of the predictability experiments, we
discuss briefly the spectral characteristics of the three key quantities (SST, over-
anomalous surf ace stress is important to the generation of the quasi-decadal variability, the anomalous fresh water flux is important to the generation of the interdecadal variability. In both types of variability, however, the air-sea coupling
appears to be rather weak, and the resulting modes are rather strongly damped, as
can be inferred from the spectral analyses presented above.
9.5 Predictability ofinterdecadal variability in the North
Atlantic
After having discussed some aspects of the dynamics of the decadal and interdecadal variabilities in the North Atlantic, we would like to address now the question of the predictability at decadal time scales. We can distinguish three basic
cases from the above discussion (Fig. 9.17). We consider first the 'pure' stochastic
climate model. In this case, the predictability limit of oceanic quantities is given by
their persistence (autocorrelation), which ranges typically from a few months (for
surf ace quantities) to many years (for subsurface quantities). The predictability of
the atmosphere at time scales beyond the predictability limit of individual weather
phenomena, which is of the order of about fourteen days, is not affected, since no
feedback from the ocean to the atmosphere is considered (Fig. 9.17a). The second
case is the stochastically forced 'ocean-only' mode. Enhanced predictability relative
to persistence is found for oceanic quantities around the resonance frequency,
while atmospheric quantities cannot be predicted with higher skill relative to the
'pure' stochastic climate model (Fig. 9.17b). Thus, if an oceanic mode exists with a
decadal period, the variations in the ocean associated with this mode will be predictable at decadal time scales. Finally, third, we consider the stochastically forced
coupled ocean-atmosphere mode. Both atmospheric and oceanic quantities exhibit
enhanced predictability (again relative to persistence) at the resonance frequency
(Fig. 9.17c). An example ofthis latter case is the ENSO phenomenon, for which
both oceanic (e.g. eastem equatorial SST anomalies) and atmospheric quantities (e.
g. the Southem Oscillation Index) are predictable at lead times of about one year
(for a recent review on ENSO predictability see e.g. Latif et al. (1998»). If coupled
modes with decadal periods exist in the North Atlantic, both the atmosphere and
the ocean will exhibit enhanced predictability at decadal time scales in the North
Atlantic region.
We have described two such modes from our coupled model simulation, and it
will depend critically on the degree of damping, whether the coupled nature of the
mode will affect significantly the predictability of the atmosphere at decadal time
scales.
Thus, the predictability characteristics of the low-frequency variability in the
North Atlantic can provide important informations about its underlying dynamics.
We conducted therefore an ensemble of predictability experiments with our coupled model. Before we describe the results of the predictability experiments, we
discuss briefly the spectral characteristics of the three key quantities (SST, over-