Abstract
In this study, the sensitivity of the continental seasonal climate to initial conditions is estimated from an ensemble of decadal simulations of an atmospheric general circulation model with the same specifications of radiative forcings and monthly ocean boundary conditions, but with different initial states of atmosphere and land. As measures of the “reproducibility” of continental climate for different initial conditions, spatiotemporal correlations are computed across paired realizations of 11 model land surface variables in which the seasonal cycle is either included or excluded—the former case being pertinent to climate simulation and the latter to seasonal prediction.
It is found that the land surface variables that include the seasonal cycle are impacted only marginally by changes in initial conditions; moreover, their seasonal climatologies exhibit high spatial reproducibility. In contrast, the reproducibility of a seasonal land surface anomaly is generally low, although it is substantially higher in the Tropics; its spatial reproducibility also markedly fluctuates in tandem with warm and cold phases of the El Niño–Southern Oscillation. However, the overall degree of reproducibility depends on the particular land surface anomaly considered. It is also shown that the predictability of a land surface anomaly implied by its reproducibility statistics is consistent with what is inferred from more conventional predictability metrics. Implications of these results for climate model intercomparison projects and for operational forecasts of seasonal continental climate also are elaborated.
Corresponding author address: Thomas J. Phillips, Program for Climate Model Diagnosis and Intercomparison, Lawrence Livermore National Laboratory, L-103, Livermore, CA 94551. Email: phillips14@llnl.gov
