Tornadoes tend to occur within transient synoptic scale systems, so as the location of such systems is altered by climate conditions, it follows that tornado frequency will also be altered. We have devised a hierarchical, Bayesian regression to examine interannual covariability of tornado frequency and climate indices. The primary motivation for this approach is the need to have a non-gaussian, non-linear regression model with the flexibility to account for spatial correlation of climate index effects.

We have generated a regular grid of annual tornado counts by counting the number of tornado reports that occurred within 50km boxes overlayed on the eastern two-thirds of the U.S. We have regressed annual tornado counts against Nino-3.4 index (ENSO), North Pacific Index (NPI), and North Atlantic Oscillation index (NAO), including interaction terms between ENSO and NPI and ENSO and NAO. Preliminary modeling indicates negative (positive) association between annual tornado counts and ENSO (NPI) over much of the eastern U.S. (western Plain states). The signal between ENSO and tornado counts is strongly modulated by NPI such that positive association between the NPI-ENSO interaction term and tornado counts occurs in the eastern U.S.

At the conference, we will discuss physical analyses in support of these results and refinements to our regression approach.