Improving NWS Convection-Allowing Hazardous Weather Ensemble Forecasts through Optimizing Multi-Scale Initial Condition (IC) Perturbations

Reliable and sharp storm scale ensemble forecasts (SSEFs; i.e., grid spacing of ~1-4 km) require proper sampling of initial condition (IC) errors from multiple scales. Common methods of generating IC perturbations for SSEFs include downscaling perturbations from a coarser resolution ensemble and adding these downscaled perturbations to the convective scale control analysis, or even just directly downscaling the coarser resolution ensemble members. While such methods are easy to implement, convective scale perturbations that have been shown to contribute to upscale growing forecast errors are excluded from the IC perturbations. An alternative approach is to take advantage of the ensemble perturbations generated as a byproduct of ensemble-based data assimilation on the convection-permitting grid to initialize the forecast ensemble.

While issues relating to the optimal design of SSEF IC perturbations have been previously studied, the results have varied along with domain size, data assimilation configuration, grid spacing ratio at the lateral boundaries and how well other sources of forecast uncertainty are accounted for in the ensemble design. In this study we aim to advance SSEF design at the National Weather Service by implementing and evaluating the multi-scale IC perturbation methods using near-operational settings where the operational model and near operational GSI based multi-scale DA system ingesting both large scale, mesoscale and convective scale radar data are used.

A series of three systematic experiments focused on high impact convective events are conducted using a HRRRE-like domain, GSI-based data assimilation of conventional and radar observations, and a multi-physics with SKEB configuration. First, the skill of forecasts initialized with multi-scale IC perturbations is compared to the skill of forecasts initialized with IC perturbations downscaled from SREF and GEFS ensembles. Second, the impact of blending the two IC perturbation sources at different scales in order to improve IC and LBC perturbation consistency is evaluated. Third, the multi-scale and downscaled IC perturbations are filtered to resolve identical scales. The filtering allows us to evaluate how well each method samples the meso- to synoptic-scale analysis uncertainty, while controlling for the upscale growth of convective scale perturbations.