A new method for determining the reliability of dynamical ENSO predictions

Determination of the reliability of particular ENSO forecasts is of particular importance to end users. Theoretical arguments are developed that indicate that the amplitudes of slowly decaying (or growing) normal modes of the coupled system provide a useful measure of forecast reliability. Historical forecasts from a skillful prediction model together with a series of ensemble predictions from a 'perfect model' experiment are used to demonstrate that these arguments carry over to the practical prediction situation. In such a setting it is found that the amplitude of the dominant normal mode, which strongly resembles the observed ENSO cycle, is a potentially useful index of reliability. The fact that this index was generally lower in the 1970s than the 1980s provides an explanation for why many coupled models performed better in the latter decade. It does not, however, explain the low skill of some coupled models in the early 1990s as the index defined here was then moderate.