Cane and Zebiak developed a coupled model that can
predict anomalies months in advance. These anomalies are indices
of the El Nino. The model has not been improved in about a
decade and thus the accuracy of the model's prediction of the
ENSO (El Nino Southern Oscillation) cycle is in question.
The CZ model has problems in predicting the forecast through the spring season in the Northern Hemisphere. One limit of the CZ model is its incomplete physics. Other limitations to this model and to other models is the error in initial conditions. These error could be caused by observational inaccuracies and deficiencies in the models. The errors can be reduced by coupling the observational data of the oceanic and atmospheric components before initialization. This produces a well balanced model.
The CZ model predictability was improved without new data being introduced. The revised CZ model nudges the wind stress more towards observation and uses the initialization more in a coupled manner. This pushes the wind stress more towards reality by putting more emphasis on the observed values. This nudging parameter was evaluated and optimized from two different time periods.
Analysis of the revised CZ model showed the same interannual oscillation of the original model, but it produced a lot less noise in the wind stress anomalies. Therefore less fluctuations in other anomaly predictions, such as SST (sea-surface temperature), occured. The original model could predict the SST anomaly at least one year in advance, but it also produced many false alarms. The revised model predicted considerably better. It was more stable and consistent thus producing fewer false alarms and erroneous forecasts. Correlation and rms (root mean square) errors between the predicted and observed Nino anomalies were calculated for each model. The original model had root mean square errors greater than 0.6 for lead times up to 10 months while the revised model had root mean square errors greater than 0.6 up to 20 months of lead time. The revised model also had corrlations 0.1 to 0.2 higher than the original model at all lead times. As stated before, the original CZ model had problems forecasting in the spring season of the Northern Hemisphere. The new CZ model overcame this spring barrier and had improved predictability.
The nudging parameter, being one of the main changes between the models, can approach two extremes. One extreme produces too much noise (the original model) and the other extreme predicting values too far from reality. Therefore the nudging parameter is a trade-off between becoming too noisy and drifting from reality. The revised model does have problems predicting anomalies that are small in magnitude and occur over a short time period. It can predict small magnitude changes over long periods of time though.
The revised Cane and Zebiak model has better consistency and stability over the old model. It gives less noise thus producing smaller fluctuations in the predicted anomalies. Forecasts can be made for longer lead times with greater accuracy. Therefore, through coupling, the old CZ model has been improved.
Reference