ENSO Predictions with Coupled Ocean Atmosphere Models
323
purely statistical models or in hybrid coupled models is that most of the tunable
parameters have strong physical constraints. Parameters in these models are not
tuned for best prediction skill but to give the best representation of the physical
laws. Thus, a change of the mean state of different feedbacks over a certain period
are in principle not a problem for these models.
However, a lot of other problems are encountered with coupled general circulation models. First of aH they are extreme1y expensive in terms of computer time
which makes it difficult to tune these models and to obtain time series long enough
for reliable statistics. Second, most of the coupled models show a more or less
severe climate drift. This means that the model drifts into a new and unrealistic c1imatology after coupling. Typical c1imate drift values in terms of SST are between
0.1 and 1 degree centi grade per year. The drift is due to an imbalance of the energy
fluxes that are exchanged between the models. Therefore flux corrections are
applied to many coupled models. Another approach to avoid climate drift is to couple only anomalies relative to a mean annual cycle which is determined separately
for the ocean and atmosphere model. However, both methods are questionable if
the model is used for ENSO studies. Flux correction methods generally reduce or
even suppress interannual variability, and anomaly coupling ignores non-linear
interactions between the ENSO cyc1e and the annual cyc1e and cannot handle a
change ofthe mean state. Therefore mode1s without any flux correction are usually
used for ENSO studies and forecasts. Since the integration time for ENSO studies
is typicaHy of the order of one to two decades and for forecasts about one year, a
c1imate drift of about 1 degree centi grade in 10 year, as it is, for instance, found in
the ECHO model can be accepted. To achieve such a relatively low climate drift,
most ENSO mode1s do not exchange informations at high latitudes, like poleward
of SO"N/S. In these regions c1imatological boundary conditions are prescribed for
the atmosphere and ocean. The reason is that during the formation and melting of
sea ice strongly nonlinear processes play an important role, and errors in the
description of these processes may lead to a very severe climate drift.
Nowadays ENSO prediction skills are achieved with coupled general circulation
models that are comparable with the results from intermediate or hybrid coupled
models.
16.4 ENSO predictions
16.4.1 Initialization methods
A very important issue in carrying out an ENSO forecast is the initialization of
the forecast system. ENSO prediction is like numeric al weather prediction an initial value problem. The further evolution of the system highly depends on the initial state, and thus an initial state as realistic as possible is crucial for a successful
forecast. In section 16.3.2 it was already pointed out that the atmosphere in the
tropics is strongly forced by sea surface temperature anomalies and that the memory of the system resides entirely in the upper ocean. Thus, the initialization of the
323
purely statistical models or in hybrid coupled models is that most of the tunable
parameters have strong physical constraints. Parameters in these models are not
tuned for best prediction skill but to give the best representation of the physical
laws. Thus, a change of the mean state of different feedbacks over a certain period
are in principle not a problem for these models.
However, a lot of other problems are encountered with coupled general circulation models. First of aH they are extreme1y expensive in terms of computer time
which makes it difficult to tune these models and to obtain time series long enough
for reliable statistics. Second, most of the coupled models show a more or less
severe climate drift. This means that the model drifts into a new and unrealistic c1imatology after coupling. Typical c1imate drift values in terms of SST are between
0.1 and 1 degree centi grade per year. The drift is due to an imbalance of the energy
fluxes that are exchanged between the models. Therefore flux corrections are
applied to many coupled models. Another approach to avoid climate drift is to couple only anomalies relative to a mean annual cycle which is determined separately
for the ocean and atmosphere model. However, both methods are questionable if
the model is used for ENSO studies. Flux correction methods generally reduce or
even suppress interannual variability, and anomaly coupling ignores non-linear
interactions between the ENSO cyc1e and the annual cyc1e and cannot handle a
change ofthe mean state. Therefore mode1s without any flux correction are usually
used for ENSO studies and forecasts. Since the integration time for ENSO studies
is typicaHy of the order of one to two decades and for forecasts about one year, a
c1imate drift of about 1 degree centi grade in 10 year, as it is, for instance, found in
the ECHO model can be accepted. To achieve such a relatively low climate drift,
most ENSO mode1s do not exchange informations at high latitudes, like poleward
of SO"N/S. In these regions c1imatological boundary conditions are prescribed for
the atmosphere and ocean. The reason is that during the formation and melting of
sea ice strongly nonlinear processes play an important role, and errors in the
description of these processes may lead to a very severe climate drift.
Nowadays ENSO prediction skills are achieved with coupled general circulation
models that are comparable with the results from intermediate or hybrid coupled
models.
16.4 ENSO predictions
16.4.1 Initialization methods
A very important issue in carrying out an ENSO forecast is the initialization of
the forecast system. ENSO prediction is like numeric al weather prediction an initial value problem. The further evolution of the system highly depends on the initial state, and thus an initial state as realistic as possible is crucial for a successful
forecast. In section 16.3.2 it was already pointed out that the atmosphere in the
tropics is strongly forced by sea surface temperature anomalies and that the memory of the system resides entirely in the upper ocean. Thus, the initialization of the