Statistical ENSO forecasting with subsurface ocean temperatures as predictors

Abstract

Listen

Translate article

This study demonstrates the improvement in skill of a statistical ENSO prediction scheme when subsurface ocean information is included. This scheme used depth variations of the thermocline in the tropical Pacific Ocean, represented by changes in the 20°C isotherm depth (Z20), as an ENSO predictor. Using leave-one-out cross-validation, the skill of two types of regression-based prediction schemes for projecting the SST anomalies in the Nino 3 region was evaluated. The ‘subsurface’ scheme has for its predictors, persistence and the second empirical orthogonal function (EOF2) of Z20. The ‘surface’ scheme also included persistence as its predictor but replaced the isotherm EOF2 with the SST EOF2. Separate EOF analysis performed on both the isotherm and SST data within the Indo-Pacific region revealed the first empirical orthogonal function (EOF1) to be closely associated with ENSO. Though less in terms of variance, the EOF2 for each data appeared to be a precursor field of an ENSO event several months earlier. This allowed isotherm EOF2 and SST EOF2 to be a useful predictor variable in the regression schemes. The prediction skill of the two prediction schemes, expressed in terms of skill correlation and explained variance, was compared with persistence that served as the primary skill control measure. Results revealed that for lags greater than three months and up to fifteen months, the ‘subsurface’scheme produced better Nino 3 predictions than the ‘surface’model. Of more importance was the scheme’s ability to produce statistically significant forecasts even across the boreal spring predictability barrier. While this current work focuses on the prediction of ENSO, the use of subsurface ocean temperatures can possibly be extended towards the direct prediction of rainfall and streamflow in Australia.