In this study, the Southern Oscillation Index (SOI) is decomposed in terms of intrinsic mode functions (IMF) using empirical mode decomposition (EMD) method. The EMD is a recently developed method that can resolve natural oscillation patterns hidden in a time series. The method is capable of analyzing nonlinear, nonstationary climate data that always cause problems to many linear statistical methods and their users.

Our results show that the EMD method, in general, provides a good tool to look into many characteristics of interannual variability in the tropics, such as varying periods of different ENSO events, phase locking of modes to annual cycle, etc. Especially noteworthy is a multi-decadal oscillatory trend of SOI that resembles the global temperature anomaly over the last century. That oscillatory trend contributes very significantly to ENSO events after 1984, especially the long-lasting ENSO event in early 1990s. However, when that oscillatory trend of SOI is further decomposed, we find that the amplitude of real trend is very small, while the amplitude of a multi-decadal oscillation reaches an unprecedently large value. Since the amplitude of this oscillation pattern was large in 1870s, reached its minimum around 1920, and has been growing ever since, it is not ready for us to predict its future based on current information.