Improved ENSO Prediction Skill Resulting From Reduced Climate Drift in IAP‐DecPreS: A Comparison of Full‐Field and Anomaly Initializations

摘要

When initiated from observational conditions, coupled climate models used in seasonal predictions generally experience climate drifts. How climate drift affects El Ni?o–Southern Oscillation (ENSO) prediction is important but not clearly understood. Here, we investigate this issue by comparing seasonal hindcasts using two distinct initialization approaches, namely, anomaly and full‐field initializations, based on a climate prediction system named IAP‐DecPreS. The differences between the two approaches are mainly evident in the drift behavior. We find that the hindcasts based on anomaly initialization (Hindcast‐A) have higher ENSO prediction skill compared to those based on full‐field initialization (Hindcast‐F). The climate drifts are largely reduced in the Hindcast‐A as expected. In contrast, the Hindcast‐F features a growing warming of the equatorial central eastern Pacific with increasing lead times. To investigate the impact of drift on the prediction, the 1997/1998 and 2015/2016 El Ni?o cases are analyzed. At a 7‐month lead, the Hindcast‐A reasonably predicts the two events, while the Hindcast‐F shows large errors in both evolution and amplitude. Budget analyses show that the underestimation of warming tendency in the Hindcast‐F is caused by cooling effects of excessive anomalous surface shortwave radiative flux and anomalous temperature advection by mean horizontal currents, both of which are associated with the climate drift. Our results imply that the use of the AI scheme can improve ENSO predictions through a reduction in climate drift in IAP‐DecPreS. The drifts can dynamically influence ENSO predictions, and their impact cannot be thoroughly removed via the empirical bias correction. Thus, reducing drift impacts is necessary. Plain Language Summary Seasonal climate prediction is an initial value problem in nature. The initial conditions for the seasonal prediction based on coupled general circulation model are obtained from model initializations through introducing observational records into the model. There are two distinct model initialization approaches, full‐field initialization (FFI) and anomaly initialization (AI). The FFI offers more accurate initial conditions, but forces model far away from its inherent climatology. In contrast, the AI offers initial conditions with anomalies deviated from the mean states close to those in the observation, but stays model mean state as in the free run. Based on a climate prediction system, IAP‐DecPreS, we show that the AI is superior to the FFI for the ENSO prediction. The major reason is that for the prediction integrations initialized from the FFI, the model drifts toward its preferred state, during which the physical processes responsible for the ENSO evolution inherent in the model are disrupted.