ESTIMATING THE EFFECT OF STOCHASTIC WIND STRESS FORCING ON ENSO IRREGULARITY

One open question in El Nino-Southern Oscillation (ENSO) simulation an d predictability is the role of random forcing by atmospheric variabil ity with short correlation times, on coupled variability with interann ual timescales. The discussion of this question requires a quantitativ e assessment of the stochastic component of the wind stress forcing. S elf-consistent estimates of this noise (the stochastic forcing) can be made quite naturally in an empirical atmospheric model that uses a st atistical estimate of the relationship between sea surface temperature (SST) and wind stress anomaly patterns as the deterministic feedback between the ocean and the atmosphere. The authors use such an empirica l model as the atmospheric component of a hybrid coupled model, couple d to the GFDL ocean general circulation model. The authors define as r esidual the fraction of the Florida State University wind stress not e xplained by the empirical atmosphere run from observed SST and a noise product is constructed by random picks among monthly maps of this res idual. The impact of included or excluded noise is assessed with sever al ensembles of simulations. The model is run in coupled regimes where , in the absence sf noise, it is perfectly periodic: in the presence o f prescribed seasonal variability, the model is strongly frequency loc ked on a 2-yr period; in annual average conditions it has a somewhat l onger inherent ENSO period (30 months). Addition of noise brings an ir regular behavior that is considerably richer in spatial patterns as we ll as in temporal structures. The broadening of the model ENSO spectra l peak is roughly comparable to observed. The tendency to frequency lo ck to subharmonic resonances of the seasonal cycle tends to increase t he broadening and to emphasize lower frequencies. An inclination to ph ase lock to preferred seasons persists even in the presence of noise-i nduced irregularity. Natural uncoupled atmospheric variability is thus a strong candidate for explaining the observed aperiodicity in ENSO t ime series. Model-model hindcast experiments also suggest the importan ce of atmospheric noise in setting limits to ENSO predictability.