Improving Storm Scale Ensemble Forecast Performance by Optimizing Multi-Scale Initial Condition Perturbations.

An important, and still open, question for SSEF design is how to optimally perturb the Initial Conditions (ICs) such that the differences among ensemble members at the initial time adequately sample the analysis errors contributing to subsequent forecast errors. The accuracy of convective scale forecasts depends on processes on convective scales, as well as the synoptic and mesoscale environments supporting such processes. Therefore, reliable and sharp convective scale probabilistic forecasting requires proper sampling of the fast-growing IC errors across a broad range of scales. Building on past work in the context of perfect model OSSEs, this study uses 10 convectively active cases and an operationally feasible ensemble configuration that is similar to the operational HRRRE. Different IC perturbation methods are extensively tested in order to optimize ensemble forecast performance and transition the results from this research effort into operations. Particular emphasis is placed on the relative merits of downscaling ICs from a coarser ensemble vs. implementing a multi-scale ensemble DA system to provide the IC perturbations, the potential advantage of blending both types of ensemble perturbations, and the interactions with other sources of ensemble diversity.

It is found that generating the ICs with the multi-scale DA system does have forecast advantages over downscaling from a coarser ensemble, the convective scales of the IC perturbations are important in addition to the larger perturbation scales, in some ways blending can further improve forecast performance, and the results can be somewhat sensitive to how the physics errors are sampled in the ensemble.