Using convection-allowing models to produce forecast guidance for severe thunderstorm hazards via a “surrogate severe” approach

To assist forecasters in producing predictions of severe thunderstorm hazards (i.e. wind, hail, and tornadoes), convection-allowing models (CAMs) provide useful guidance about convective storm mode (Done et al. 2004), and thus, likely convective hazards. CAMs have the potential to provide more detailed information about the placement and intensity of these hazards, even though they do not explicitly predict all of them. Specifically, since CAMs are capable of resolving features as small as individual convective elements, it is possible that extremes in model fields associated with convection might be useful predictors of observed severe phenomena. For example, the presence of intense rotating updrafts at CAM grid points could be used as a surrogate for the locations of severe weather reports. If the relationship between the locations of the “surrogates” and observed severe phenomena is robust, unique guidance products can be developed, along with potential application in future warn-on-forecast systems (Stensrud et al. 2009).

To explore this idea, we examined a collection of model fields produced from the 4-km 36-hour WRF-ARW simulation run daily at 00 UTC at the National Severe Storms Laboratory (NSSL) during Spring 2008. Fields based on simulated model reflectivity, vertical velocity, 10 meter wind speed, and updraft helicity (a diagnostic field to identify supercellular storms) were chosen, as preliminary assessments suggested that they were likely to be useful indicators of convective intensity in the model. For each field, surrogate severe storm reports (SSRs) were placed at grid points where a parameter threshold was exceeded at any point during a 24-hour period corresponding to the 12 to 36 h forecast period (12 UTC to 12 UTC). To produce a guidance product akin to SPC probabilistic Convective Outlooks, a Gaussian smoother was applied to the grid of SSRs to create a density (or probability). This product is referred to as a surrogate severe probability forecast (SSPF).

Sobash et al. (2009) established a proof-of-concept and highlighted some of the challenges associated with this procedure. In this work, an objective and quantitative comparison between the SSPF guidance and observed storm reports was undertaken. First, the ability of the SSRs to reproduce the observed spatial behavior of observed storm reports was documented. Additional research questions included 1) what threshold yields the most skillful SSPF guidance for each surrogate field and 2) what surrogate field provides the most skillful guidance overall? Traditional verification measures for probabilistic forecasts, including forecast reliability and resolution, were used to investigate these questions. Although a deterministic CAM was used in this work, the procedure can be extended to an ensemble of CAMs. An ensemble-based application of this work will be highlighted in another presentation at the conference.