Using modern time series analysis techniques to predict ENSO events from the SOI time series

Abstract. We analyze the monthly 1866-2000 Southern Oscillation Index (SOI) data to determine:

1) whether the SOI data are sufficiently noise-free that useful predictions can be made from them, and

2) in particular, whether future ENSO events can be predicted from the SOI data. 

The "Hilbert-EMD" technique is used to aid the analysis. This new frequency-time algorithm, based on the Hilbert transform, may be applied to time series for which the conventional assumptions of linearity and stationarity may not apply.

With the aid of the EMD procedure, a cleaner representation of ENSO dynamics is obtained from the SOI data. A polynomial function is then used to predict SOI values. Using only the data from January 1866 through December 1996, this prediction correctly indicated a warm event in 1997-1998 and a cold event in 1999. Using all the data (through December 2000), this prediction shows no strong ENSO events (positive or negative) during the time period January 2001 through December 2004.