Recent numerical studies on torrential rain during Baiu season in Japan demonstrate that the behavior of a meso-b-scale precipitation band is often well reproduced with a 5km-mesh non-hydrostatic model and the back-building nature of meso-g-scale precipitation cells within the band with a 2km-mesh model. However, how critical is the horizontal resolution of the model for reproducing the behavior of meso-b-scale precipitation systems and individual storms does not seem to have been examined in detail. In order to shed some light on this problem, we have studied sensitivity of a numerically-simulated classic supercell on horizontal grid size, since its basic behavior is relatively well understood. The size of the calculation domain is 128 km x 128km x 16.3km. By placing a thermal bubble at the canter of the calculation domain, the storm was initiated from the horizontally uniform state given by the Del-City storm sounding (Klemp and Whilhelmson, 1978; hereafter referred to as KW78). All the simulations are performed under the same condition except that the horizontal grid size is varied between 1 km and 3 km with an interval of 100 m. It is found that the behavior of the simulated storm changes drastically between 2.5 km and 2.6 km, and between 2.7 km and 2.8 km.

The "standard run" with grid size of 1.0 km reproduced a typical supercell as found by the previous studies (e.g., KW78): the initial storm splits into right- and left-movers. The former continues to intensify and start to have a hook-shaped rainwater pattern and a mesocyclone. It lasts throughout the simulation time of 3 hours with repeating cyclic generation of mesocyclones (e.g., Adlerman et al., 1999). These characteristics of the simulated storm continue to be observed as far as the grid size is less than 2.5km, though spatial patterns of the storm become somewhat smoothed as the grid size increases. When the grid size becomes larger than 2.5 km, however, a drastic change in the storm behavior occurs. After the initial storm is split, the right-mover dissipates but the left-mover lasts throughout the simulation time. The left-mover, however, does not exhibit any indication of a supercell. When the grid size is larger than 2.7 km, the storm does not show any splitting but straightly weakens after its initiation. It is rather surprising that a change of grid size by 100 m causes qualitative change in the storm behavior. The critical grid sizes at which the storm behavior changes drastically would of course vary from model to model. However, it is likely that any meso-scale model would have such critical grid sizes. If this is the case, one has to be deliberate in choosing the grid size when one attempts to predict individual storms or meso-b-scale precipitation system in which convection cells play an important role in maintaining the parent system.