How Predictable Is

JULY 2003 AMERICAN METEOROLOGICAL SOCIETY | A recent analysis of 12 statistical and dynamical models used for El Niño predictions finds that at the long (1–2 yr) and even medium (6–11 months) ranges there were “no models that provided useful and skillful forecasts for the entirety of the 1997/98 El Niño” (Landsea and Knaff 2000). Most of the models were wrong in predicting the timing of the onset and/or demise of El Niño, and unable to predict the full duration and even one-half of the actual amplitude of the event. An earlier study by Barnston et al. (1999) reached similarly discouraging conclusions. Regarding El Niño of 2002, the forecasts again cover a very broad spectrum of possibilities (Kirtman 2002). Why is it so difficult to predict El Niño? How predictable is this phenomenon? There is agreement that the continual Southern Oscillation—El Niño in its warm phase—depends on ocean–atmosphere interactions that amount to a positive feedback. (The winds induce sea surface temperature patterns that in turn affect the winds.) The intensity of the feedbacks is a matter of considerable debate. Figure 1 shows a range of possibilities, from a strongly damped oscillation (Fig. 1a), to a self-sustaining and approximately linear oscillation (Fig. 1b) to an oscillation that is so unstable that nonlinearities introduce irregularities (Fig. 1c). In the latter case, instabilities amplify errors in the initial conditions and thus limit predictability. If reality were to correspond to this parameter range then the predictability of El Niño would be similar to the predictability of weather. An appropriate approach is then an ensemble of calculations, each one starting from a slightly different initial condition. Cane et al. (1986), Zebiak and Cane (1987), Zebiak (1989), Tziperman et al. (1995), and others explore this range of parameters. Figure 1a shows a very different range of parameters, that of a highly damped oscillation. Now random atmospheric disturbances are always responsible for initiating developments. Predictability is once again very limited, not because of errors in initial conditions as in the case of weather, but rather because of the random disturbances. (Penland and Sardeshmukh 1995; Penland 1996; Thompson and Battisti 2000, 2001, explore this parameter range.) To insist that westerly wind bursts, which occur at ranHOW PREDICTABLE IS EL NIÑO?