Application of a reduced-order Kalman filter to initialize a coupled atmosphere-ocean model: Impact on the prediction of El Nino

A reduced-order Kalman filter is used to assimilate observed fields of the surface wind stress, sea surface temperature, and sea level into the couple d ocean-atmosphere model of Zebiak and Cane. The method projects the Kalman filter equations onto a subspace defined by the eigenvalue decomposition o f the error forecast matrix, allowing its application to high-dimensional s ystems. The Zebiak and Cane model couples a linear, reduced-gravity ocean model wit h a single, vertical-mode atmospheric model. The compatibility between the simplified physics of the model and each observed variable is studied separ ately and together. The results show the ability of the empirical orthogona l functions (EOFs) of the model to represent the simultaneous value of the wind stress, SST, and sea level, when the fields are limited to the latitud e band 10 degreesS-10 degreesN, and when the number of EOFs is greater than the number of statistically significant modes. In this first application of the Kalman filter to a coupled ocean-atmospher e prediction model, the sea level fields are assimilated in terms of the Ke lvin and Rossby modes of the thermocline depth anomaly. An estimation of th e error of these modes is derived from the projection of an estimation of t he sea level error over such modes. The ability of the method to reconstruct the state of the equatorial Pacifi c and to predict its time evolution is shown. The method is quite robust fo r predictions up to 6 months, and able to predict the onset of the 1997 war m event 15 months before its occurrence.