Predictability and variability of a coupled ocean-atmosphere model

Abstract

This paper presents a summary of our research on the predictability and variability of a coupled ocean-atmosphere model (Cane et al., 1986; Zebiak and Cane, 1987). The detailed description of our work including the modeling experiments and the results are being presented elsewhere in the form of two scientific papers. In the first paper (Goswami and Shukla, 1990a) we have investigated the predictability of a coupled ocean-atmosphere model and in the second paper (Goswami and Shukla, 1990b) we have proposed a mechanism for aperiodic variability in a coupled ocean-atmosphere model.We first integrated the model for 24 years with prescribed wind stress forcing from observations beginning with January, 1964. We refer to this as the control run. We then used the initial conditions from this simulations to integrate the coupled ocean-atmosphere model. We first integrated the coupled model for a period of 36 months for 181 separate initial conditions corresponding to the period January 1970 through January 1985 of the control run. We have compared the forecasts of SST by the coupled model with the observed and the control simulation. We have noted some systematic errors in the model suggesting that forecasts can be further improved by removing the systematic errors. We find that the SST forecasts with the coupled model are better than persistence for the first three months. It is also worth noting that the root mean square error between the control run initial conditions and observations are comparable to the standard deviations of the observations themselves.We have also integrated the coupled model for 15 years by slightly perturbing the surface winds initially. Using a large ensemble of such identical twin experiments we have found that the growth of small initial errors in this coupled model is characterized by two well separated time scales. The fast time scale gives an error doubling time of 5 months and the slow scale gives an error doubling time of about 15 months. We are encouraged by the prospects of extended range predictions using coupled models because of the existence of the slow time scale, however, in order to realize the potential predictability of the coupled, model it would be essential to control the fast time scale error growth.We have also investigated the possible mechanisms responsible for the aperiodic behavior of this model. Sensitivity of the coupled model's variability to the nonlinearily associated with the coupling processes is studied. The atmospheric heating associated with the anomalous low level convergence (convergence feedback) seems to play an important role in producing the model's aperiodic variability. We show that this feedback has a strong seasonality due to its dependence on the seasonal mean convergence. In the absence of the convergence feedback, the standard parameters of the Cane and Zebiak model give a periodic variability with a periodicity of about 4 years. This feedback produces a broadening of the basic low frequency spectrum through the introduction of a high frequency component.