Datensatz

Zusammenfassung

Two processes limiting the predictability of the El Niño/Southern Oscillation
(ENSO) phenomenon \Mere investigated. First, the perpetual action of fluctua-
tions in wind stress forcing that are not correlated to ENSO itself but that are
an integral part of the tropical atmosphere-ocean system was included into a sim-
plified coupled model of ENSO. The implications of this random element for the
dynamics and predictability were studied. Second, the growth of small errors in
the initial conditions of the coupled model was analysed. This v¡as accomplished
by computing its singular vectors and singular values, i.e. the spatial structures
and the amplification rates of those initial state perturbations that grow most
strongly over a given time interval.
The simplified coupled model of ENSO used consists of an ocean general circu-
lation model coupled to a diagnostic atmosphere model. Following common ter-
minolog¡ such a model is called a Hybrid Coupled Model (HCM). The HCM was
designed to simulate the interannual climate variability of the ocean-atmosphere
system in the tropical Pacific region. The diagnostic atmosphere exploits the sta-
tistical correlation of anomalous sea surface temperature and wind stress present
in observations. Via linear regression both quantities are related in a reduced state
space of their leading Empirical Orthogonal Functions (EOFs).
To study the effect of random perturbations during the forecast, the cou-
pled model was complemented by a stochastic anomalous wind stress field. This
stochastic part was derived from high-pass filtered data. It mimics the observed
statistics of the random perturbations. The singular vectors, on the other hand,
were derived by generating the linearised numerical code of the HCM and the corre-
sponding adjoint code. This was done with the help of an automatic differentiation
tool.
Both above-mentioned processes are important in limiting the predictability of
ENSO. They can be understood as parts of a stochastic dynamical system which
suffers from the imperfect knowledge of its initial state and the unpredictable
components during simulation. It is shown that each ENSO prediction faces a
natural limit of predictability depending on the season and the phase of ENSO at
its start. I conclude that the limit of ENSO predictability is substantially shorter
than the typical ENSO cycle period.