Comparison of Error Growth Characteristics in Convection-Permitting Ensembles

Convection-permitting ensemble forecasts are known to be overly-confident and under-dispersive. Here, we examine the sensitivity of warm-season convection-permitting ensemble precipitation forecasts over the conterminous US to different initial condition perturbation methods designed to promote error growth. These methods include correlated random perturbations and flow-dependent perturbations sampled from the NCEP Short-Range Ensemble Forecasts (SREF) and an ensemble Kalman filter data assimilation system. A scale decomposition of kinetic energy perturbation growth and precipitation perturbation energy were used to characterize the multi-scale error growth properties of these initial condition perturbation methods.

Findings indicate that the random perturbations have the fastest kinetic energy error growth at all scales over the first 12 hours, as compared to using the flow-dependent perturbations. Beyond 12 h, perturbation errors from all three methods begin to converge. By 36 h, differences among the three perturbation errors collapse, indicating a loss of information of the different initial perturbations. The relationship between initial condition perturbation strategy, and precipitation forecast skill and spread-skill relationship, will also be discussed. Results from this study will help guide the development of alternative perturbation methods to promote error growth.