Exploring the use of CAM Ensembles for Next-day Tornado Prediction

Three diagnostic fields were examined to assess their ability to act as surrogates for tornadoes within the NCAR convection-allowing ensemble system run daily since Spring 2015. The diagnostics included mid-level (2 km – 5 km AGL) updraft helicity (UH25), low-level (0 – 3 km AGL) updraft helicity (UH03), and low-level (1 km AGL) vertical relative vorticity (RVORT1). The relationship between these diagnostics and the simulated near-storm environments revealed that simulated storms possessing large magnitudes of RVORT1 tended to occur within environments similar to those identified as conducive for the development of tornadic supercells in proximity sounding-based studies (e.g., low lifted condensation levels and strong low-level shear). This result motivated the use of low-level rotation (e.g., UH03 or RVORT1) as a direct surrogate for tornadoes, instead of using UH25 combined with environmental information. The behavior of UH25 and UH03 among the simulated storms was also explored; UH03 only exceeded UH25 in storms occurring within low-CAPE/high-shear environments, while UH03 rarely exceeded UH25 in traditional supercell environments.

Next-day ensemble surrogate severe probability forecasts (E-SSPFs) for tornadoes were generated using these diagnostics for a large collection of ensemble forecasts from Spring 2015 and 2016. E-SSPFs for tornadoes using RVORT1 and UH03 were more skillful than E-SSPFs using UH25. The UH25 E-SSPFs possessed little skill, regardless of threshold or smoothing length scale. All E-SSPFs suffered from poor sharpness at skillful scales, with few forecast probabilities greater than 40%. The implications of these results for the prediction of tornadoes with convection-allowing models will be discussed.