ALBERT

Description: Numerical and statistical predictions of simplified models are linearly combined in a sensitivity study to identify the crucial parameters that are needed to enhance predictability of the El Nino-Southern Oscillation (ENSO) phenomenon. The results indicate that the ENSO prediction skill of the simplified models can be improved. The profit achieved strongly depends on the phase information that is utilized by the forecast combination and is inherent in predictions of a quasi-periodic process such as ENSO. The simplest forecast combination that still yields useful forecasts at longer lead times of about half of the ENSO period ( 18 - 24 months) is the combination of two persistence forecast schemes. For the prediction period 1982 - 2003, that is the persistence of a sea surface temperature anomaly (SSTA) index area at 60degreesS, 180degreesW and the Nino-3 index SSTA. The level of skill improvement critically depends on the prediction schemes and prediction period, as well as on the period from which the combination weights are derived. Differences between combination forecast and hindcast are minimized if the statistical weights are derived from a time period that is characterized by an ENSO statistic that is close to the prediction period. In this study the prediction period has simply been halved for the sake of simplicity to derive the statistical weights, which is sufficient for predicting the 1980s and 1990s ( with each other), but not for predicting, for example, the 1970s. The suppressed 1976/77 El Nino event makes the periodic occurrence less regular compared to the other decades. However, this forecast combination technique can be applied in a much more elaborate way