As computer power continues to increase, the grid spacing of numerical weather prediction models correspondingly will decrease to the point that cloud-resolving models will be run on a real-time basis. However, whereas large-scale predictability issues have been studied extensively, very little work has explored the nature of storm-scale predictability. The main objective of this research is to ascertain the length of time for which useful forecasts of various features of the atmosphere at these scales can be expected. These features range from the exact timing, location, mode, and intensity of storms to the nature of the environment (instability, shear, etc.) in which storms may form. The predictability limits for these features will be estimated from large ensemble runs. As a preliminary step the ensembles will be launched from an idealized, horizontally homogeneous control run so that storms may be isolated and the storm environment closely controlled to determine the impact of environmental characteristics on storm predictability. Following that, horizontally nonhomogeneous numerical simulations will also be performed as the interaction between storms and inhomogeneities in the storm environment are likely to affect predictability. All experiments will be performed using the perfect model assumption so that the results will not be limited by the current skill of high-resolution numerical models.
These experiments will provide an upper bound on storm-scale predictability. Knowledge of the inherent predictability limits at these storms is important for directing future research toward physically tractable problems and for instructing forecasters on how to interpret the output of high-resolution numerical models.
