K. Bryan et S. Griffies, NORTH-ATLANTIC CLIMATE VARIABILITY ON DEC ADAL TIME-SCALES - IS IT PREDICTABLE, Izvestia Akademii nauk. Rossijskaa akademia nauk. Fizika atmosfery iokeana, 32(5), 1996, pp. 591-599
The predictability of Atlantic climate variability is an important sci
entific as well as practical problem. Methods of determining predictab
ility through ''identical twin'' numerical experiments with models hav
e been worked out for the atmosphere. These methods have also been app
lied to models of the El Nino/Southern Oscillation. Delworth, Manabe a
nd Stouffer (1993) have demonstrated a 40-50 year period of climate va
riability in the GFDL model. The simulated variability involves an Atl
antic thermohaline conveyor belt and produces quite realistic surface
temperature anomalies. Although the GFDL model does not reproduce all
of the observed characteristics of the historical data for Atlantic cl
imate variability, it appears to be the best available model at the pr
esent time for predictability studies. Ensemble experiments have been
carried out in which an initial error is inserted in the atmospheric c
omponent of the GFDL model. The solutions indicate a rapid error growt
h in the atmospheric component. This error then spreads at a slower ra
te to the ocean component. A measure of the predictability is the time
at which the variance within the ensemble becomes 50% of the climatol
ogical variance. Using the maximum amplitude of the North Atlantic the
rmohaline circulation as an index, experiments indicate that the 50% p
oint is reached in about 2-5 years. Tests using empirical orthogonal f
unctions of the meridional transport stream function and the dynamic t
opography indicated greater predictability. The variance of the yearly
averaged amplitudes in the ensembles reach 50% of saturation in 10-20
years for the first two empirical orthogonal functions. For compariso
n, the dominant time scale of the model variability is 40-60 years. Th
e results imply that the predictability is very sensitive to both spat
ial and temporal smoothing and this has to be taken into account in th
e design of monitoring systems.