The theme of my thesis research is to perform breeding experiments with NASA/NSIPP coupled general circulation model (CGCM) in order to obtain ENSO-related growing modes for ensemble perturbations. We show for the first time that the breeding method is an effective diagnostic tool for studying the coupled ENSO-related instabilities in a coupled ocean-atmosphere general circulation model that includes physical and dynamical processes of many different time scales. We also show for the first time that it is feasible to utilize the coupled bred vectors (BV) as a way to construct perturbations for ensemble forecasts for ENSO prediction using an operational coupled climate prediction model.

The results of the thesis research show that coupled breeding can detect a precursor signal associated with ENSO events. Bred vectors are characterized by air-sea coupled features and they are very sensitive to ENSO phases and background season. This indicates that bred vectors can effectively project on the seasonal-to-interannual instabilities by growing upon the slowly varying coupled instability. These results are robust: bred vectors obtained from both the NASA and NCEP coupled systems exhibit similarities in many fields, even in atmospheric teleconnected regions.

We show that bred vectors have a structure similar to the one-month forecast error (analysis increment). The BV growth rate and the one-month forecast error show similar low frequency variations. Both of their subsurface temperatures have large-scale variability near the depth of thermocline. Evidence shows that bred vectors capture the eastern movement of the analysis increment (one-month forecast error) along the equatorial Pacific during 1997-1998 El Niño evolution. The results suggest that one-month forecast error in NSIPP CGCM is dominated by dynamical errors whose shape can be captured by bred vectors, especially when the BV growth rate is large.

These results suggest that bred vectors should be effective coupled perturbations for ensemble ENSO predictions, compensating for the lack of coupled ENSO-related perturbations in current operational ensembles. The similarity between the bred vectors and the one month forecast errors suggests that bred vectors can capture "errors of the month" and could also be applied to improve oceanic data assimilation by providing information on the month-to-month background variability.