| Summary: | The global responses of two atmospheric general circulation models (AGCM), the National Centers for Environmental Prediction–Medium Range Forecast (NCEP–MRF9) and the University of Hamburg climate model–3 (ECHAM), to simultaneous global SST forcing are examined on a 3-month timescale. Rotated principal components analysis of the model and observations is also used to identify and compare their leading modes of coherent variability. The scope of the present analyses is largely descriptive and does not attempt to explain the differences in model behavior in terms of their formulations. The authors’ main focus is to quantify the simulation skill of the two comprehensive AGCMs on seasonal timescales and compare it to skill obtained using empirical prediction models. Both models are found to exhibit realistic responses to El Niño–Southern Oscillation (ENSO)-related forcing, with the ECHAM response slightly more accurate in the spatial phasing and structure of the atmospheric anomalies. The ECHAM model exhibits realistic atmospheric responses to tropical Pacific SST forcing as well as patterns associated with extratropical internal atmospheric dynamics [e.g., North Atlantic oscillation (NAO) and a high latitude north–south dipole in the Pacific]. It shows a slightly higher signal-to-noise ratio than that found in the real world, while the NCEP model’s signal-to-noise ratio is approximately equal to that in nature. The NCEP model responds with more zonally symmetric atmospheric patterns than observed, although this does not prevent it from forming realistic responses to ENSO over the Pacific–North American region. The NCEP model’s NAO variability is only about half as strong as that observed. In terms of simulation skill with respect to observations, the ECHAM model generally tends to outperform the NCEP model for global 500-hPa geopotential height and surface climate. A decomposition of the observed and model data into rotated principal components indicates that both models reproduce the ENSO-related anomalies in circulation and surface climate of the real atmosphere quite well. The ECHAM model, which handles ENSO variability and impacts slightly better than the NCEP model, shows a larger increment of capability in reproducing other global climate processes. Two linear statistical benchmarks, which are used as skill control measures, sometimes outperform the NCEP model but are more comparable, on average, to the skill of the ECHAM model. Thus as noted in other recent studies, the dynamical models and the statistical models have roughly the same simulation skill and would be expected to have similar forecast skill if the models used forecasted SSTs as their boundary conditions. To first order, the linear component of the relationships appears to be modeled well by the two dynamical models. It is undetermined whether instances of better performance of the dynamical models than the statistical benchmarks are partly attributable to the models’ effective exploitation of nonlinearities in the relationships between tropical SST and global climate. One reason for this inconclusiveness is that evidence for nonlinearities in the present analyses is not compelling. Hence, the question of whether dynamical models have untapped potential to consistently outperform statistical models on the seasonal timescale remains open and may require close examination of each physical formulation in the dynamical models.
|
|---|