Transition characteristics and predictability of ENSO diversity

Transition characteristics and predictability of ENSO diversity are examined using sea surface temperature (SST) observation (1870-2012 HadISST V1.1) and 4000-year pre-industrial control simulation from GFDL CM2.1 GCM. Using a phase space spanned by the first two principal components of monthly Tropical Pacific SST anomalies, the continuum of SST variability is divided into five domains: Eastern Pacific/ Central Pacific El Niņo (EPEN/CPEN) and Eastern Pacific/Central Pacific La Niņa (EPLN/CPLN) as well as neutral patterns. Domain transition probabilities across various time range are summarized for grouped patterns within one domain or subgroup for each season. The transition probability tailored for given SST anomaly pattern is applied to predict which ENSO domain it may develop to, with according likelihood. Optimal initial patterns to predict each domain are also diagnosed. Prediction horizon to distinguish EP/CP flavor of ENSO is restricted to several months. CPLNs and EPENs are more predictable than EPLNs and CPENs. A series of transition diagram are introduced, which highlight several SST variability characteristics including local persistence, zonal propagation and EN-LN phase change. They provide measures of GCM performance/bias on the transition characteristics of ENSO diversity. In observation, SST variability normally propagate westward and ENs/LNs mainly peak in boreal winter. Transition details differ with seasonal timing, e.g., EPLNs appearing in summer (JJA)/autumn (SON) tend to propagate westward to become CPLNs while EPLNs that appear in winter (DJF)/spring (MAM) generally decay locally. Observed ENs typically first appear as EPENs and later move westward decaying as CPENs. Some portion of ENs that develop locally as CPENs often peak around autumn. In the simulation, LNs primarily propagate westward and peak in winter, which is consistent with the observation. As to ENs, they tend to first appear as CPENs in winter and propagate eastward peaking as EPENs in summer, which is different from observation. In addition, extreme EPENs in the observation are often tightly followed by EPLNs while in the simulation they are followed by CPLNs. This study is potentially applicable for model comparison and diagnostics.