Creating physically coherent and spatially correlated perturbations to initialize high-resolution ensembles of simulated convection

The use of ensembles for numerical weather prediction has become common during the last decade. For global models, the generation of initial-condition perturbations has a number of well-tested methodologies. In ensembles that predict convective storms explicitly (i.e., Delta x<4$$ \Delta x km), the generation of physically realistic perturbations is less well posed. This study introduces a technique to generate physically coherent and spatially correlated (PCSC) initial-condition perturbations that are calibrated to the environment. Ensembles of idealized CM1 simulations, initialized with either PCSC perturbations (EXP_PCSC), spatially coherent random perturbations (EXP_3KM), or Gaussian white-noise random perturbations (EXP_WHITE), are run for both a linear convective line of storms and a single "supercell" storm to demonstrate the utility of this new perturbation technique in diverse environments. PCSC perturbations are extracted from high-resolution simulations of boundary-layer turbulence and random perturbations are calibrated to be the same in magnitude as PCSC perturbations. EXP_PCSC simulations spawn turbulence fastest in this study. The simulated turbulence is more robust than in other experiments more than 1 hr into the simulation, because horizontal convective rolls enhance power at the largest scales. Random perturbations are slow to generate turbulence; this problem is exacerbated when the base model state flow is nonturbulent. Due to robust turbulence, EXP_PCSC ensemble spread increases fastest during the first simulation hour and remains largest throughout the remainder of the simulations. Although EXP_PCSC spread is the largest, the sensitivity of convection to the initial perturbations varies at different times in the storm life cycle. Storms appear more sensitive to perturbations added near the time of convective initiation.