Forecasting regional sea surface temperatures in the tropical Pacific by neural network models, with wind stress and sea level pressure as predictors

We constructed two types of neural network models for forecasting the sea surface temperature anomaly (SSTA) over several standard equatorial Pacific regions (Nino 3, 3.4, 3.5, 4, P2, P4, and P5). The first type used the sea level pressure (SLP) as predictors, while the second one used the wind stress. By ensemble averaging over 20 forecasts with random initial weights, the resulting forecasts were much less noisy than those in our earlier models. The models performed best in the western-central equatorial regions and less well in the eastern boundary regions. At longer leads of 9 - 12 months, the cross-validated skills (1952 - 1993) for the models using the tropical Pacific SLP as predictors were statistically higher than those using the wind stress, Overall, the models using the tropical SLP showed usable cross-validated skills up to 12-month lead. The true out-of-sample forecast performances during the 1982 - 1993 period for the Nino 3.5 SSTA at lead times of 9, 12, and 15 months attained correlation skills of 0.78, 0.80, and 0.75, respectively.