Variability of North Pacific SST, on both interannual and decadal timescales, is well modeled as the sum of direct forcing by El Nino-Southern Oscillation (ENSO), "re-emergence" of North Pacific SST anomalies in subsequent winters, and white noise. Thus, over a series of winter/springs, the North Pacific integrates the effect of ENSO, resulting in a power spectrum of the Pacific Decadal Oscillation (PDO) which has more power than ENSO itself for periods greater than about 10 years. An observational analysis yielding the above results is presented, and it is shown that a low-order Linear Inverse Model (LIM; Penland and Sardeshmukh 1995) which includes forcing by ENSO is very skillful at forecasting PDO one year ahead (for cross-validated forecasts over the years 1900-2001, anomaly correlation=.86) and can reproduce the PDO power spectrum on both interannual and decadal time scales.

Power spectra of the leading modes of both Tropical and North Pacific SST variability of ten different coupled models from the Coupled Model Intercomparison Project (CMIP2) are compared to confidence intervals derived from the observational LIM. In addition, a LIM analysis is performed on output for each coupled model. Using the simple model formulation from above, differences in the simulated North Pacific decadal variability are directly related to spatiotemporal differences in the simulated ENSO variability, which are in turn well captured by the LIMs.