Using Ensemble Sensitivity Analysis to Identify Storm-Scale Characteristics Associated with Tornadic Potential in High-Resolution Idealized Supercells

While a majority of all tornadoes are associated with supercell thunderstorms, only about 25% of all supercells actually go on to produce tornadoes. Predicting which supercells will spawn a tornado has proven to be difficult for researchers and forecasters alike. While environmental conditions certainly play a role in supercell formation and tornadogenesis, studies have also shone light on the importance of storm-scale properties (e.g., potential temperature within the rear-flank downdraft) that may influence tornado potential. It is proposed here that the importance of such storm-scale features can be identified through the use of ensemble sensitivity analysis (ESA) on high-resolution idealized simulated supercells.

In ESA, a forecast metric (e.g., updraft helicity, maximum lowest-level vertical vorticity) is related to a model state variable (e.g., virtual potential temperature) at an earlier time through ensemble statistics. ESA has been shown to effectively identify sensitivity in the synoptic scales, but has also been applied to the mesoscale and, most recently, the convective scale. While the inherent non-linearity of these smaller scales make ESA application and interpretation more difficult, robust signals can be identified for relatively short lead times.

In this study, ESA is applied to an ensemble of 51 idealized supercells simulated in Cloud Model Version 1 (CM1). To create the ensemble, a control simulation is run using the May 24, 2011 Rapid Update Cycle (RUC) sounding used in Orf et al. (2016) to initialize a homogeneous environment. The supercell produced one hour into the model is then used to initialize 50 simulations with slightly perturbed versions of this sounding. Even with relatively small perturbations in the initial sounding, the ensemble contains both nontornadic and strongly tornadic members. ESA is employed to identify areas of sensitivity that best relate the model state to the production of strong vertical vorticity (or lack thereof) within these storms. Sensitivity signals for lowest-level vertical vorticity are evident within the rear-flank downdraft and along the left-flank convergence boundary, with sensitivity values varying across different thermodynamic state variables (e.g., equivalent potential temperature versus virtual potential temperature). The sensitivity of lowest-level vertical vorticity to the vertical profile of thermodynamic variables within the rear flank and forward flank is also presented, providing a look at how surface vertical vorticity is affected by thermodynamic characteristics above the surface, which is not easily observed.