There are at least two significant reasons why trying to predict tornadoes directly from the ENSO signal is problematic:

1) because the tornado database comes only from reported tornadoes, it is thought that many more tornadoes occur than are reported, especially from sparsely populated areas. In addition, not all tornado and high wind reports are thoroughly investigated, and not all tornadoes and high wind damage can be surveyed to determine the true nature of the damage. Tornado damage and strength assessment are subjective, and indeed the amount of damage depends on what areas the tornadoes hit, not necessarily the character of tornadoes themselves. These problems do not affect the storminess database which is objectively obtained.

2) The physical process of tornadogenesis in the dry season generally involves the spinup of a tornado from a thunderstorm mesocyclone. The fact that a tornado may or may not spinup is determined by processes on a very small space-time scale and can theoretically have nothing to do with sea surface temperatures in the tropical Pacific Ocean. Indeed, why some mesocyclones produce tornadoes and some do not is not well-understood. Anecdotal evidence from NEXRAD doppler radar observations at Melbourne over the past 10 years indicates that most mesocyclones do not, in fact, produce tornadoes. On the other hand, the synoptic environment that is favorable for tornadogenesis and severe weather is well known and understood. This is why storminess is perhaps a better measure of seasonal tornado potential, and a more useful proxy for assessing societal impact of ENSO.