Sensitivity of the NCEP Regional Spectral Model to Domain Size and Nesting Strategy

Abstract

This paper evaluates the performance of the National Centers for Environmental Prediction (NCEP) Regional Spectral Model (RSM) based on the sensitivities of different model domain sizes and horizontal resolutions. The perturbation method and the spectral computation in the NCEP RSM construct the nesting strategy as a ?domain nesting? in physical space as well as a ?spectral nesting? in spectral space, instead of the conventional ?lateral boundary nesting? as used in most regional models. The NCEP RSM has the same model structure, dynamics, and physics as its outer coarse-resolution global model, and it also has a terrain blending along the lateral boundary at the initial time. Both together result in a smooth lateral boundary behavior through one-way nesting. An optimal lateral boundary relaxation reduces the influence of lateral boundary error and generates more areas with small-scale features. The treatment of numerical stabilities, such as a semi-implicit time scheme, time filter, and horizontal diffusion, are applied in perturbation without recomputing or disturbing the large-scale waves. The combination of the aforementioned methods is the uniqueness of the NCEP RSM, which demonstrates the capabilities to conserve the large-scale waves, resolve the mesoscale features, and minimize the lateral boundary errors. A case of winter cyclogenesis with propagation of the synoptic-scale disturbances through the lateral boundaries was selected to investigate the sensitivities of the NCEP RSM based on different nesting strategies. The results from the experiment over a quarter-sphere domain with similar resolutions between RSM and T126 global model demonstrated that the domain nesting was a success, because the lateral boundary error and perturbation were negligibly small. The experiments in a 48-km resolution with different sizes of the model domain had mixed results. The continental domain had the best performance but inclined to generate erroneous large-scale waves that degraded its performance after 60 h. The results of the regional and subregional domains were proximity to their base field, T126, in terms of root-mean-square differences. They had similar mesoscale features in a 48-km horizontal resolution regardless of the different model domain sizes. The results from the experiments with nesting in different coarse grids over the radar-range domain imply that it can use either a T126 or subregional domain as its base field for similar performances. Nevertheless, more mesoscale features were obtained by the experiment with the base field at higher resolution. The results from the experiments, with 30-day integration, reveal that the performance of the experiment in the subregional domain was much closer to its base field than that in the continental domain. It indicates that the predictability of the global model is the predictability of the NCEP RSM in the regional domain; however, the regional domain could generate higher-resolution features than its base field. This successful long-range integration with the regional domain is because the lateral boundary errors are relatively small and the large-scale waves are preserved through the domain and spectral nesting.