Impact of a Stochastic Kinetic Energy Backscatter Scheme on Warm-Season Convection-Allowing Ensemble Forecasts

The stochastic kinetic energy backscatter (SKEB) method adds random perturbations to the wind and temperature fields according to a prescribed power spectrum and a decorrelation time parameter. Previous research has shown a positive impact from the SKEB scheme not only on the skills of global model ensembles (with a grid spacings ranging between 45 km and approximately 52 km), but also in an improved tail of the KE spectrum near the grid scale. However, the benefit of using the SKEB scheme in convection allowing models still remains to be explored.

The SKEB scheme was implemented in the WRF-ARW and tested for 31 cases spanning May 2013 across a large portion of the US. The WRF was configured for storm-scale forecasting with 4 km grid spacing and no convection parameterization. To compare the impact of the SKEB scheme against that of commonly used multi-physics method of ensemble generation, three ensembles, each of size seven members, were configured. Two of the ensembles used identical physics perturbations with one using the SKEB scheme (multi-phys+SKEB) and the other not (multi-phys). The third ensemble contained no physics variation, but used the SKEB scheme with each member using a different random number seed (SKEB). In applying SKEB to convection allowing ensembles, a few sensitivity experiments were conducted to find the most appropriate SKEB parameters.

Verification was performed for a variety of synoptic-scale fields as well as for precipitation. Synoptic-scale fields were verified against 13-km RAP analyses; surface fields were verified against surface observations from in-situ instruments; precipitation was verified against Q3 precipitation analyses from NSSL.

KE spectra computed from the model without SKEB strongly agree with those in Skamarock (2004). The drop away from a -5/3 slope towards the grid scale in the mesoscale reflects the near-grid-scale dissipation and is not considered to be a model deficiency that the SKEB can improve upon since there is no difference in KE spectra between simulations that use the SKEB scheme and those that do not.

Initial results suggest that adding the SKEB scheme to a mixed physics ensemble results in an ensemble that is much more statistically reliable, as spread is significantly increased with minimal change in the RMSE of the ensemble mean. For some fields, SKEB results in a nearly properly dispersive ensemble. Results are mixed for precipitation verification, as the skill scores for various metrics are very similar between the multi-phys+SKEB and multi-phys ensembles. Interestingly, spread in the SKEB ensemble was nearly as large as that in the multi-phys+SKEB ensemble, with rank histograms that had a very similar shape as those from the multi-phys+SKEB ensemble as well. The SKEB ensemble also had similar PQPF verification scores as the multi-phys+SKEB ensemble.

References: Skamarock, W. C., 2004: Evaluating mesoscale NWP models using kinetic energy spectra. Mon. Wea. Rev., 132, 3019–3032.