The Advent of Convection-Permitting Forecast Models: Now What?

In the past decade, numerical models have become routinely integrated using a horizontal mesh spacing below 5 km. These so-called “convection-permitting” models are now being integrated for global forecasts and research problems. This talk explores the use of such models to date and examines their limitations, especially in light of fundamental predictability limits, neglected processes, data volume, and computational cost. With resolved turbulent scales being a daunting prospect for prediction models even in the next decade, several avenues of investigation appear pressing. The development of scale-insensitive parameterizations of cloud systems is needed for variable-resolution or nested domains that attempt to focus resolution where it is needed. Data assimilation, to produce initial conditions and quantify systematic model errors in convection-permitting models, is an active and growing area, especially for regions where satellite radiances are the dominant observations. This effort for predictive time scales of a few hours or less runs headlong into extremely rapid error growth at convective scales, making progress toward numerical nowcasts difficult. Finally, extraction of user-relevant information from convection-permitting forecasts represents a considerable challenge that will require advances in algorithmic, statistical and human-based detection.