204
about two weeks, by inevitable growing errors in the atmosphere-there is
no way to deterministically predict the motions of the atmosphere beyond
this limit of deterministic predictability no matter how small the initial
error.
The second type of predictability is governed by the slowly varying
boundary conditions to which the atmosphere responds. If we can predict
SST, or soil moisture, or ice and snow cover, then we will be able to know
the statistics of the atmosphere in equilibrium with these boundary conditions. The system is initialized by initializing those factors that affect the
boundary condition: in practice, this is only possible for the SST boundary
conditions and those parts of the ocean that affect the surface in a given
prediction time must be initialized. This is the type of prediction that has
been so successful for SST in the tropical Pacific characteristic of ENSO
(e.g. Latif et al., 1994): only the upper ocean is initialized since only the
near-surface ocean will affect SST on predictions times of order one year.
Because the maritime tropical atmosphere is so tightly coupled to SST,
skillful forecasts of rainfall around the tropical Pacific basin can be made
in terms of the SST predictions.
The third type of predictability is that typified by the predictions made
for greenhouse warming. No aspect of the climate system is initialized
but scenarios for the radiatively active constituents of the atmosphere are
specified as function of time and the statistics of the entire atmosphereocean- land system are predicted in response to the changing atmospheric
constituents.
Thus for prediction of the first type, we predict what the precise state
of the atmosphere is for, say, next Wednesday. For prediction of the second
type, we might predict the March averaged rainfall over Peru nine months
in advance. For the third type of prediction, all we can say is that summer
temperatures over the US will generally be warmer, but we can't offer
a prediction for the specific month, August 2050, say. In this type of
prediction, the magnitude and phase of the natural cycles are not predicted
for a specific time. Only in the second type of prediction can we capture the
magnitude and phase of a decadal-to-centennial cycle and we will therefore
confine ourselves to that type of prediction.
In order for the coupled atmosphere-ocean system to be predictable we
must be able to predict the slowly varying boundary conditions and the
nature of the mechanism determines the possible range of predictability.
When the mechanism for low frequency variability is due to random forcing
about two weeks, by inevitable growing errors in the atmosphere-there is
no way to deterministically predict the motions of the atmosphere beyond
this limit of deterministic predictability no matter how small the initial
error.
The second type of predictability is governed by the slowly varying
boundary conditions to which the atmosphere responds. If we can predict
SST, or soil moisture, or ice and snow cover, then we will be able to know
the statistics of the atmosphere in equilibrium with these boundary conditions. The system is initialized by initializing those factors that affect the
boundary condition: in practice, this is only possible for the SST boundary
conditions and those parts of the ocean that affect the surface in a given
prediction time must be initialized. This is the type of prediction that has
been so successful for SST in the tropical Pacific characteristic of ENSO
(e.g. Latif et al., 1994): only the upper ocean is initialized since only the
near-surface ocean will affect SST on predictions times of order one year.
Because the maritime tropical atmosphere is so tightly coupled to SST,
skillful forecasts of rainfall around the tropical Pacific basin can be made
in terms of the SST predictions.
The third type of predictability is that typified by the predictions made
for greenhouse warming. No aspect of the climate system is initialized
but scenarios for the radiatively active constituents of the atmosphere are
specified as function of time and the statistics of the entire atmosphereocean- land system are predicted in response to the changing atmospheric
constituents.
Thus for prediction of the first type, we predict what the precise state
of the atmosphere is for, say, next Wednesday. For prediction of the second
type, we might predict the March averaged rainfall over Peru nine months
in advance. For the third type of prediction, all we can say is that summer
temperatures over the US will generally be warmer, but we can't offer
a prediction for the specific month, August 2050, say. In this type of
prediction, the magnitude and phase of the natural cycles are not predicted
for a specific time. Only in the second type of prediction can we capture the
magnitude and phase of a decadal-to-centennial cycle and we will therefore
confine ourselves to that type of prediction.
In order for the coupled atmosphere-ocean system to be predictable we
must be able to predict the slowly varying boundary conditions and the
nature of the mechanism determines the possible range of predictability.
When the mechanism for low frequency variability is due to random forcing