Predictive Skill of Statistical and Dynamical Climate Models in SST Forecasts during the 1997—98 El Niño Episode and the 1998 La Niña Onset

Abstract

Critical reviews of forecasts of ENSO conditions, based on a set of 15 dynamical and statistical models, are given for the 1997?98 El Niño event and the initial stages of the 1998?99 La Niña. While many of the models forecasted some degree of warming one to two seasons prior to the onset of the El Niño in boreal spring of 1997, none predicted its strength until the event was already becoming very strong in late spring. Neither the dynamical nor the statistical models, as groups, performed significantly better than the other during this episode. The best performing statistical models and dynamical models forecast SST anomalies of about +1°C (vs 2.5°?3° observed) in the Niño 3.4 region prior to any observed positive anomalies. The most comprehensive dynamical models performed better than the simple dynamical models. Once the El Niño had developed in mid-1997, a larger set of models was able to forecast its peak in late 1997 and dissipation and reversal to cold conditions in late spring/early summer 1998. Overall, however, skill for these recent two years does not appear greater than that found over an earlier (1982?93) period. In both cases, median model correlation skill averaged over lead times of one to three seasons is near or just above 0.6. Because ENSO extremes usually develop in boreal spring or early summer and persist through the following winter, forecasting impact tendencies in extratropical North America for winter (when impacts are most pronounced) at 5 months of lead time is not difficult, requiring only good observations of the summer ENSO state and knowledge of the winter teleconnections. Because of the strength of the 1997?98 El Niño and the consequent skill of 5-month lead forecasts of U.S. winter 1997?98 impacts, the success of these forecasts was noticed to an unprecedented extent by the general public. However, forecasting impacts in austral winter that occur simultaneously with the initial appearance of an ENSO extreme (e.g., in Chile, Uruguay, Kiribati, Ecuador, and Peru) require forecasting the boreal spring/summer onset of ENSO events themselves at several months of lead time. This latter task is formidable, as evidenced by the fact that formal announcements of an El Niño did not occur until May, leaving little time for users in the above regions to prepare. Verbal summaries of ENSO forecasts issued to users worldwide during the 1997?98 El Niño event contained ambiguities. To address the needs for forecasts to be expressed verbally for nontechnical users and also to be precise enough for meaningful utility and verification, a simple numerically based verbal classification system for describing ENSO-related forecasts is presented.