Reducing Model Error Effects in El Niño–Southern Oscillation Prediction Using Ensemble Coupled Data Assimilation

Abstract

Model error is an important source of uncertainty that significantly reduces the accuracy of El Niño–Southern Oscillation (ENSO) prediction. In this study, ensemble coupled data assimilation was employed to estimate the tendency error of the fifth-generation Lamont–Doherty Earth observation (LDEO5) model, which represented the comprehensive effect of different sources of errors. Then, the estimated tendency error was applied to an ensemble prediction system for ENSO prediction. Assimilation experiments showed that tendency error estimation yielded better analysis than state estimation only. With tendency error estimation, simulated state variables such as zonal wind stress anomalies and subsurface temperature anomalies in the Niño3.4 region and upper layer depth anomalies along the equator showed good agreement with their reanalyzed counterparts. The ensemble ENSO prediction system with tendency error estimation demonstrated significantly better prediction skill than the ensemble system without tendency error estimation or the original LDEO5 model, especially for long lead times. The tendency error estimation improved the prediction skill for El Niño more than for La Niña. This study provides a promising approach to further improve prediction skill by reducing model error effects in an ensemble prediction.

Keywords

• El Niño–Southern Oscillation
• model error effect
• ensemble adjustment Kalman filter
• data assimilation
• simulation and prediction