Model errors of an intermediate model and their effects on realistic predictions of El Nino diversity

In the present study, a nonlinear forcing singular vector (NFSV)-based data assimilation approach is adopted to quantify the model errors of an intermediate coupled model (ICM) and their effects on El Nino predictions. Then, the tendency errors of the NFSV structure (NFSV-TEs) that represent the combined effect of different kinds of model errors are determined in terms of the sea surface temperature (SST) anomaly component in the El Nino predictions. The NFSV-TEs exhibit large values over the eastern equatorial Pacific and on model boundaries, indicating that large model errors exist there. In addition, two dominant NFSV-TEs are found: one is E-type that NFSV-TEs are mostly located in the far eastern Pacific, and the other is the D-type that presents positive anomalies in the eastern equatorial Pacific (EQeast-TEs) and negative anomalies in the central equatorial Pacific (EQcenter-TEs). The D-type NFSV-TEs often occur during realistic predictions of El Nino events. Simulations using the ICM equipped with the NFSV-TEs are then implemented to study the effect of NFSV-TEs on the ENSO predictions. It is shown that the ICM forced by the EQeast-TEs shows better performance in reducing prediction errors and systematic bias, while the EQcenter-TEs forcings are superior to the EQeast-TEs forcings in reproducing the horizontal distribution of the SST anomaly, especially in identifying the difference between the central tropical (CP) and eastern tropical (EP) El Nino. This is because EQcenter-TEs forcings can adjust not only the wind but also the ocean processes to yield realistic air-sea conditions favouring CP-El Nino formulations. Therefore, to make a better prediction of CP-El Nino, the model uncertainties occurring in the central tropical Pacific should be considered preferentially and finally removed in realistic predictions of El Nino diversity.