Forecasting ENSO Events: A Neural Network–Extended EOF Approach

The authors constructed neural network models to forecast the sea surface temperature anomalies (SSTA) for three regions: Niño 4, Niño 3.5, and Niño 3, representing the western-central, the central, and the eastern-central parts of the equatorial Pacific Ocean, respectively. The inputs were the extended empirical orthogonal functions (EEOF) of the sea level pressure (SLP) field that covered the tropical Indian and Pacific Oceans and evolved for a duration of 1 yr. The EEOFs greatly reduced the size of the neural networks from those of the authors’ earlier papers using EOFs. The Niño 4 region appeared to be the best forecasted region, with useful skills up to a year lead time for the 1982–93 forecast period. By network pruning analysis and spectral analysis, four important inputs were identified: modes 1, 2, and 6 of the SLP EEOFs and the SSTA persistence. Mode 1 characterized the low-frequency oscillation (LFO, with 4–5-yr period), and was seen as the typical ENSO signal, while mode 2, with a period of 2–5 yr, characterized the quasi-biennial oscillation (QBO) plus the LFO. Mode 6 was dominated by decadal and interdecadal variations. Thus, forecasting ENSO required information from the QBO, and the decadal–interdecadal oscillations. The nonlinearity of the networks tended to increase with lead time and to become stronger for the eastern regions of the equatorial Pacific Ocean.