that the eddies in the Agulbas retroflection region are strongly related to the formation of the Benguela Current and hence weak eddy activity in the model may have led to poor reproduction of this current.
Comparison of the eddy kinetic energy derived from the buoy data in this study with that of the SC model shows that the eddy kinetic energy of the model is much less than that from the buoy data. For example, the energy level obtained from the buoy data in the Agulbas retroflection region reaches 2000cm2s-2 while that in the SC model is 500cm2s-2. This result is consistent with that of Garraffo et al. (1992) and Wilkin and Morrow (1994). They concluded that the peak energy from the model result is less than that from the altimetry by a factor of 4. Wilkin and Morrow (1994) further showed that in the region of the low eddy energy, model results indicate the much less energy level by a factor of 10, similar to the result of comparison between the WOCE model and the altimetry in the North Atlantic (Stammer and Boning, 1992).
4 Assimilation model
As indicated by many previous studies (e.g., Qiu et al., 1991) one of the major problems for use in monitoring mesoscale to large scale ocean circulation and in improving its predictability through numerical modeling is the lack of the accurate geoid information on length scale of a few hundred kilometers. This has limited most studies based on altimetric data to focus on either the time-varying components of the SSH or the basin scale circulation patterns (Stammer and Wunsch, 1994). For many oceanographic applications, an accurate mean SSH with length scale of several hundred kilometers is required. This is particularly true for the western boundary current and its extension regions where spatial changes of the mean SSH are comparable to those of the time-varying ones.
To make a better determination of the mean SSH on mesoscales, the model of Willebrand et al. (1990) and Marshall (1985) will he combined and extended in this study. The mean SSH field from the climatological hydrographic data or the model average is corrected by assimilating velocities derived from the drifting buoy data in addition to the altimetric data to overcome deficiencies in these previous studies. When the velocity and SSH data are assimilated, the ageostrophic component of the velocity is considered. By using this assimilation model, an absolute SSH can be obtained not only in regions where the drifting buoy data are available but in the nearby regions. The error of mean SSH field is expected to be corrected by the velocity data over all wave lengths of our concern.
To estimate the absolute SSH from drifting buoy and altimetric data, we adopt the identical twin approach, which uses the model-simulated results as the "observations" instead of real measurements (hereafter the "model-simulated data" are called "observation data"). To carry out several case studies, the model is set as simply as possible; a 1.5-layer, primitive equation model is used, in which the time varying interface depths are regarded as the altimetric data (SSH). The assimilation scheme of optimal interpolation for multivariate (Daley, 1991) is adopted, which is formulated with an assumption that the error field of the velocity data and that of the SSH data are geostrophically related. It should be noted that geostrophic relationship is assumed for only the error field and hence the velocity field derived from the assimilation includes the ageostrophic components. The method of successive correction of the mean SSH field follows Marshall (1985) which is similar to the optimal interpolation.
4.1 Numerical model
The model basin is a rectangular ocean of 40° width in both latitude and longitude that idealizes the western part of the North Pacific. The governing equations are familiar ones of a 1.5-layer reduced gravity model with steady wind stresses forcing a double-gyre circulation. The values of parameters used are summarized in Table 1. After a 22-year integration, the model state is
