A 16-level general circulation model of the Indian Ocean issued to reproduce interannual variability in ocean circulation from 1987 to 1990 using wind stresses from the United Kingdom Meteorological Office (UKMO) operational forecast model analyses. The results of this integration are compared with those obtained using the Hellerman and Rosenstein <1983> (HR) climatological forcing. The differences in the two wind products are large, as are the resulting surface and subsurface currents. Model results are compared with observations at 55¿E and 73¿E on the equator. The undercurrent using UKMO stresses is more strongly eastward than that using HR, and in the four years studied, the model always simulates an undercurrent in February-March at Gan (73¿E), contrasting with the inconstancy of its appearance in the years sampled by the observations. Interannual differences in the surface currents can be large, as much as 200 cm s-1 in May on the equator. The years 1987 and 1988 are worthy of special consideration, since 1987 was a drought year over India and 1988 had above average rainfall. Therefore, the period 1987-1988 is repeated using stresses from the European Centre for Medium Range Weather Forecasts (ECMWF). There are large differences between the ECMWF and the UKMO stresses and, consequently, in the model currents, where differences can be as much as 80 cm s-1 over a large area of the equatorial zone and in the Somali current region.

At undercurrent depth, the response at Gan (73¿E) is very similar in the first half of the year in all three integrations, but differs significantly in the second half with the UKMO and ECMWF stresses inducing a semiannual response, whereas HR leads to an annual cycle. Although interannual differences are large on a month to month basis, they are not long lasting. Interannual variations in model sea surface temperatures (SST) resulting from variations in the surface stress are poorly related to observed SST anomalies. There are no direct measurements of currents in the 1987-1990 period with which to validate the model simulations. There are, however, satellite measurements of the distortion of the sea surface and so comparison is also made between the simulations of dynamic topography from the model and GEOSAT altimetry data. While there is broad-scale similarity in pattern between the three model simulations and the GEOSAT observations, detailed agreement between the fields is poor. The model results are also compared to the dynamic topography derived from Levitus data. ¿ American Geophysical Union 1993