Assessing the skill of the Pacific Decadal Oscillation (PDO) in a decadal prediction experiment

Abstract

A modified approach to the assessment of the prediction skill of “modes of variability” is proposed and applied to a decadal prediction experiment. In particular, the skill of predicting the Pacific Decadal Oscillation (PDO) is investigated. The approach depends on separately calculating the EOFs of the observations, the ensemble of forecasts, and an ensemble of simulations made with the same model and external forcing. The skill of predicting and simulating the spatial structure of the modes is captured by comparing forecast and simulated EOFs with the observation-based EOFs. This is in contrast to the case where forecasts and simulations are expanded in observation-based EOFs, or other structure functions, which gives no direct information about the model-based EOF structures themselves. The skill of predicting the temporal evolution of EOFs is separately captured by comparing the associated expansion functions. Finally, the contribution of the modes to the overall prediction skill is obtained by weighting the spatial and temporal skills with the variances involved. The behaviour of the first mode, identified as the PDO, is given particular attention. Perhaps not unexpectedly, the EOF structure of the forecasts more closely resembles that of the simulations than that of the observations, but both reproduce the structure of the observed PDO quite well with spatial correlations near 0.8. The temporal correlation of the expansion functions is near 0.7 for year 1 forecasts and declines toward zero subsequently. The overall correlation skill for the North Pacific is dominated by the PDO with a small contribution from the second mode and none from the third mode.