Conditional Tornado Probabilities from Ruc-2 Forecasts

Accurately predicting the timing and location of tornadoes with lead times of more than a few tens of minutes will continue to be difficult, despite recent advances in numerical weather prediction. Tornadoes and the associated thunderstorms are smallscale phenomena, and the theory of chaos strongly suggests that such events will have a very short period of predictability (Lorenz 1969; Islam et al. 1993). However, many researchers have noted that tornadoes are typically associated with large CAPE (high buoyancy) and/or large vertical wind shear (e.g., Rasmussen and Blanchard 1998, and references therein). For lead times several hours or more, we thus aim for a more realistic goal than a precise prediction of timing and location: can we at least provide useful information on the likelihood of tornadoes based on CAPE and shear parameters from commonly available numerical forecast model output? To this end, we demonstrate here a technique to predict the conditional probability a thunderstorm will be tornadic using 12-h forecasts of CAPE and wind shear, or CAPE and helicity from the RUC-2 model (Benjamin et al. 1998). A half year’s worth of prior numerical forecasts will be used to train a Bayesian probabilistic model to forecast conditional tornado likelihood, described below.