Predicting ENSO Using Multiple Linear Regression

An evaluation of two multiple linear regression models for predicting Niño 3.4 sea surface temperature anomalies is presented. The first model is trained using monthly data from all months while the second model is trained using monthly data that varies with a three month sliding window. Both models use the monthly Niño 3.4 sea surface temperature, 850-hPa East Pacific trade wind, 850-hPa West Pacific trade wind, warm water volume, and Central Tropical Pacific indices from the Climate Prediction Center as predictors. They also use the second principal component described in Xue et al. (1994) as a predictor. Forecasts are verified using the Pearson correlation and root mean square error metrics. It is found that the first model begins to outperform the second model after six months lead time.