Fifty years of ocean data (sea surface temperatures) and 100 years of atmospheric data (mean sea level pressure) are used to examine indices associated with the detection of El Niņo events. Typically the criterion for an El Niņo episode has been for some index or indices to meet certain thresholds over a period of some months, often five. The onset of an El Niņo event, however, can be quite rapid, as in 1997, and El Niņo related impacts are often experienced within a few months of the onset. Thus, in an operational environment a five month lag in declaring an El Niņo event is seen as inconvenient. Here we present results where indices are compared to thresholds over a three month period, making such a method for declaring an El Niņo suitable for an operational definition.

The statistical analysis is based on a simple contingency table, using a dichotomous forecast of an El Niņo event occurring or not occurring. Probability of detections and false alarm ratios are calculated. Results of objective probability experiments suggest that a definition based on ocean SSTs in the NINO3 region correctly identifies each of the generally accepted strong and moderate El Niņo events of the last 50 years (a probability of detection equal to one), along with a reasonable representation of duration periods (onset and termination times). The probability of detecting the entire duration of each of the El Niņo events given that the threshold is reached is in the order of 95%, with false alarm rates typically less than 10%. A similar method is then used for providing simple forecast probabilities for the onset of an El Niņo period. A listing of historical El Niņo events used in this study, including onset and termination times is also presented.