We report an application of the ensemble Kalman filter (EnKF) and smoother (EnKS) to an intermediate coupled atmosphere-ocean model of Zebiak and Cane, into which the sea surface height (SSH) anomaly observations by TOPEX/ POSEIDON (T/P) altimetry are assimilated. Smoothed estimates of the 54,403 dimensional state are obtained from 1,981 observational points with 2,048 ensemble members. While assimilated data are SSH anomalies alone, the estimated sea surface temperature (SST) anomalies reproduce primary temporal characteristics of the actual SST. The smoothed estimate of the zonal wind anomalies is also consistent with the observation except for the westerly anomalies in the western Pacific.
IntroductionThe El Ni no-Southern Oscillation (ENSO) phenomenon is a dominant climate variation on interannual timescales, and it depends essentially upon coupled interactions of the dynamics of ocean and atmosphere (Neelin et al. 1998). ENSO is characterized by quasiperiodic interannual oscillation of tropical Pacific sea surface temperatures (SSTs) with a dominant period of approximately 4 years. Since ENSO affects not only global climate but ecosystems in and around the tropical Pacific and economies of several countries, successful prediction of ENSO is of great interest from scientific and social points of view (Latif et al. 1998).With a coupled atmosphere-ocean model, Zebiak and Cane (1987) (hereinafter referred to as ZC) have provided the first successful ENSO prediction. The ZC model is a nonlinear anomaly model of intermediate complexity and reproduces an ENSO-like quasiperiodic variation. On the basis of intermediate coupled models, data assimilation studies have been carried out for better prediction and reanalysis of ENSO events. Assimilation methods adopted to the coupled models range from simple schemes of nudging (Chen et al. 1995(Chen et al. , 1998 or direct insertion (Chen et al. 1999) to advanced ones such as the representer method (Bennett et al. 1998(Bennett et al. , 2000, the adjoint method (Lee et al. 2000), a reduced-order Kalman filter (ROKF, Ballabrera-Poy et al. 2001), or the extended Kalman filter (EKF, Sun et al. 2002).Advanced schemes are characterized by their clear assumptions based on the dynamical and statistical implications, which are in contrast to ad hoc assumptions of simple schemes (e.g., Fukumori 2001). However, the advanced schemes listed above could not treat the ZC coupled model without modification or approximation. When the above two variational methods (the representer method and the adjoint method) were applied, the atmospheric component was modified or simplified in order to derive the adjoint equations. The two sequential methods (ROKF and EKF) require a linear approximation of the nonlinear model. It means that the advanced schemes may degrade the model's ability to reproduce nonlinear physical processes.In addition, neglected system noise or Gaussian system noise employed by these works is another problem. If the model dynamics is assumed to be perfect (so-called "...