付録4-2 基調講演(2)
Dr. M. Kanamitsu、「力学的気候予測に向けて」
"Towards Dynamical Climate Prediction"
Masao Kanamitsu.
In 1981, Charney and Shukla suggested that there is a considerable seasonal predictability in the large scale tropical atmosphere when surface boundary condition, namely the sea surface temperature (sst), is given. This paper is apparently the first to point out that seasonal prediction is possible, at least in the tropics, if the sea surface temperature is known. Just about the same time, the famous El Nino of 1982-83 occurred. This strong warming in the equatorial eastern Pacific ocean caused significant change in the seasonal climate over the global tropics, as expected from Charney and Shukla study. However, the significance of this event is that the effect of El Nino was also felt in extra tropical latitudes away from the tropics, the strongest signal being found over the North American continent, although the signal is still noticeable in other parts of the globe. This remote effect of tropical sst anomaly over the North American continent, known as Pacific North American (PNA) pattern, provided evidence that the seasonal prediction in the extra tropics may also be possible.
The capability of numerical models to simulate tropical and extra tropical response to equatorial sea surface temperature became clear in the international program that compared a number of general circulation models (Atmospheric Model Intercomparison Project, or known as AMIP) held around 1991. The project is designed to compare the atmospheric simulation under the condition that the atmosphere is driven only by the specified observed monthly mean sea surface temperature. A number of models demonstrated that it is possible to simulate tropical and extra tropical response to tropical sst forcing. Encouraged by this result, several institutions started to work on experimental seasonal prediction experiments.
The next important progress in the model application was the recognition of the inherent nature of extra tropical response. The extra tropical atmosphere is dominated by the cyclone scale disturbances, whose effect is to add a large amplitude noise to the stationary response to tropical sst anomaly. The noise exists not only in the real atmosphere but also in the model simulation. In order to reduce this inherent noise, the model integration needs to be done in an 'ensemble' mode, i. e., multiple integrations for the same forecast period need to be performed. The averaging of the ensemble members act as a filter to remove the meteorological noise and provide most likely response of the tropical heating. Furthermore, the fact that meteorological noise exist in the real atmosphere implies that the forecast needs to be probabilistic, not deterministic. The ensemble forecast is the only logical way to provide probabilistic forecast, that can give information of the shift in the probability distribution.