The assimilation of satellite and in situ data in the Mediterranean Forecast System (MFS) is based on an optimal interpolation scheme which uses empirical orthogonal functions (EOFs) to represent vertical modes of the background-error correlation matrix. In this study we present a new methodology to estimate, and the calculation of, these multivariate EOFs. The new EOFs are considered time and space varying (seasonal time-scales and subregional). They examine the vertical-error cross-variance between temperature, salinity, barotropic stream function and sea-level anomaly. These EOFs are used to assimilate four years of along-track sea-level anomaly data. The validation of MFS analyses and forecasts using the assimilation system diagnostics and the comparison with independent observations show that, in relation to an old operational scheme which was using only one EOF, the use of several multivariate EOFs significantly improves the accuracy of analyses and forecasts in the Mediterranean. © Royal Meteorological Society, 2005.

cited By 21