Reconstructing the frequency of tornado occurrence in the central United States

The recorded frequency of tornado occurrence in the central United States is compared to an independent reconstruction of tornado environments. Radiosonde-based calculations are used to differentiate tornadic environments from severe non tornadic environments during the period 1958 to 2000. Regional and seasonal variation in the calculated values is shown to complicate comparison of storm environments. To facilitate comparison, values are standardized using local cumulative probabilities. This type of scaling reveals that the magnitudes of many sounding-based parameters associated with strong to violent tornadoes are climatologically extreme relative to other types of environment. Furthermore, the scaling of parameter values helps to reveal similarities between weak and strong-to-violent tornado environments. Similarities include low lifting condensation level (LCL) heights and large boundary layer wind shear magnitudes. Multivariate logistic regression is used to estimate the probability that a sounding is tornadic. When the predicted event includes weak, presumably non supercell tornadoes, logistic regression models are not well calibrated. When the predicted event is a sounding associated with supercell tornadoes, however, logistic regression models appear to be well calibrated and the discrimination of events from non events is strong. Evidence is presented that the apparent decrease in strong-to-violent tornado frequency since 1950 implied by the official storm archive is inconsistent with the radiosonde record. Abrupt changes in the bias of the reconstruction, that is, the ratio of the number of tornado soundings predicted to the number observed, are shown to be coincident with changes in storm classification procedures that occurred during the 1970s and the early 1990s. Rather than a decrease in frequency since the 1950s, the reconstruction suggests that supercell tornado frequency has been reasonably stationary until the 1990s, after which some increase in frequency is estimated. While this increase in tornado frequency occurs during a transition to a different set of storm classification practices, the increase does not appear to be an artifact of model calibration or of obvious non-climatic changes in radiosonde observations. Results suggest that variability in Convective Available Potential Energy (CAPE) and LCL have led to the recent increase in tornadic supercell thunderstorm environments suggested by the reconstruction.