Mesoscale convective systems (MCSs) are a common phenomenon across the world during the warm season. These organized lines of thunderstorms, often followed by a trailing stratiform precipitation region, remain a challenge to forecast accurately even though they represent a main source of warm season rainfall. Part of this problem is no doubt related to the challenges of convective initiation, but part of the problem also is related to the inherent predictability of these systems. Numerical simulations, in both two and three dimensions, are produced using the National Severe Storms Laboratory Collaborative Model for Mesoscale Atmospheric Simulations (NCOMMAS) to explore the predictability of MCSs. The model uses horizontally homogeneous initial conditions, as is typical of idealized cloud-scale model simulations, and also uses a line of warm bubbles to start deep convection. However, these homogeneous initial conditions are perturbed to represent the uncertainty in the initial environments in which MCSs form. Over 100 perturbed two-dimensional model runs are produced for a suite of increasingly large environmental perturbation magnitudes and compared to the control run without perturbations. Results from these two-dimensional model simulations suggest that MCSs can be predicted with only 60% certainty for environmental uncertainties representative of 24-h forecasts from current operational models. This means that even if convective initiation is predicted accurately, the uncertainty in how the convection evolves over the subsequent hours is large. These results also indicate the accuracy required in specifying the mesoscale environmental in order to increase our confidence in the numerical forecasts. Results from three-dimensional simulations, where the perturbations are again based upon errors in the forecast model fields, also are discussed. While the results are perhaps not quite so negative, the resulting divergence of the solutions is large. Sensitivities to the metric used to define a MCS also are discussed.

While Ed Lorenz did not explore questions of mesoscale predictability, his work on predictability laid the foundation upon which the present study is based.