Improving EnKF-Based Initialization for ENSO Prediction Using a Hybrid Adaptive Method

Abstract

Abstract Probabilistic forecasts, which are usually initialized by an ensemble Kalman filter (EnKF), are known to be better than deterministic (or one member) forecasts for the El Niño–Southern Oscillation (ENSO) phenomenon. Because of sampling errors caused by a finite ensemble and the errors related to model biases associated with the physical parameterizations, dynamic core, model resolution, and so on, a state-of-the-art inflation method is commonly used in the standard EnKF to increase the prior variance so as to avoid filter divergence. However, the optimal inflation factor is almost prohibitive in reality because of vast computational cost. An adaptive EnKF and multigrid analysis hybrid approach without inflation is presented to compensate for the abovementioned limitations of the standard EnKF. In this study, the adaptive approach is applied to an intermediate coupled model for ENSO prediction. Gridded observations of daily-mean sea surface temperature (SST) anomalies from the Advanced Very High Resolution Radiometer (AVHRR) during January 1982–December 2012 are assimilated into the model to initialize a 2-yr ENSO hindcast. Results show that compared to the standard EnKF that uses multiplicative variance inflation, the adaptive method can reduce analysis errors by 63% for both the daily SST anomaly and the Niño-1+2 SST anomaly. The prediction skill of Niño-1+2 SST anomaly is consistently enhanced, especially for phase forecast. For SST anomaly forecasting, the advantage of the adaptive method mainly occurs in the eastern equatorial Pacific and the northern boundary of the intermediate coupled model.