ENSO in a hybrid coupled model. Part II: prediction with piggyback data assimilation

Abstract

A hybrid coupled model (HCM) for the tropical Pacific ocean-atmosphere system is employed for ENSO prediction. The HCM consists of the Geophysical Fluid Dynamics Laboratory ocean general circulation model and an empirical atmospheric model. In hindcast experiments, a correlation skill competitive to other prediction models is obtained, so we use this system to examine the effects of several initialization schemes on ENSO prediction. Initialization with wind stress data and initialization with wind stress reconstructed from SST using the atmospheric model give comparable skill levels. In re-estimating the atmospheric model in order to prevent hindcast-period wind information from entering through empirical atmospheric model, we note some sensitivity to the estimation data set, but this is considered to have limited impact for ENSO prediction purposes. Examination of subsurface heat content anomalies in these cases and a case forced only by the difference between observed and reconstructed winds suggests that at the current level of prediction skill, the crucial wind components for initialization are those associated with the slow ENSO mode, rather than with atmospheric internal variability. A “piggyback” suboptimal data assimilation is tested in which the Climate Prediction Center data assimilation product from a related ocean model is used to correct the ocean initial thermal field. This yields improved skill, suggesting that not all ENSO prediction systems need to invest in costly data assimilation efforts, provided the prediction and assimilation models are sufficiently close.