Disclaimer: My perspective is that of an “educated outsider,” in the sense that my training is in atmospheric science, my expertise is in tornadoes, and I know enough about climate dynamics to teach its basics at the undergraduate, nonmajor level, but I am not a climate dynamics expert, nor have I ever been awarded a grant to study climate change.
Short version
Read the answer to no. 5 in this short piece published in The Conversation:
Long version
The impacts of climate change on extreme weather events receive considerable media attention. Although the influence of global warming on sea levels, droughts, heat waves, and probably even extreme rainfall is fairly clear, it’s not so obvious how tornadoes will be affected.
Let’s start with the past record of tornadoes. Unlike temperature records, which can be traced back hundreds of years using thermometer records, and many thousands of years using proxy data, extensive tornado records only exist going back to roughly 1950, and there are no tornado proxy records. Moreover, records outside of the U.S. are nowhere near as comprehensive as the U.S. record. That said, the U.S. historical tornado record has grave limitations that make its use for establishing tornado trends dubious. The record is derived solely from unsolicited eyewitness reports and post-event damage surveys, both of which suffer from representativeness errors and other unnatural biases. For example, no credible meteorologist believes that the dramatic rise in the annual tornado count since the 1950s is due to global warming, or anything else physical. Rather, the consensus is that the tornado increase is attributable to an increase in reporting. The historical record of tornadoes rated ≥ F1/EF1 [the EF (Enhanced Fujita) scale replaced the F (Fujita) scale in 2007] is flat (link to NCDC plot), which suggests that the increase in tornado count is the result of an increased reporting of F0/EF0 tornadoes. Tornado intensity estimates also are error-prone. The ratings depend on what is actually struck by the tornado, and even where damage exists, the best estimate of the maximum winds is sensitive to the duration and acceleration of the winds, as well as whether the winds are loaded with debris. We know from some serendipitous research-grade mobile Doppler radar observations of tornadoes that even the best damage surveys can misdiagnose wind speeds by over 100 mph.
There are more subtle biases in the tornado record as well. For example, the F3/EF3+ tornado record shows a slight decline since the 1950s (link to NCDC plot), and some have cited this as evidence that global warming is weakening tornadoes (link to New York Times OpEd and response). However, periods of both systematic overrating and underrating of tornadoes are known to exist, making it difficult to know the true trend in intensity, if there is one. Tornadoes preceding the adoption of the Fujita scale in the late 1970s were retrospectively rated based on qualitative damage descriptions in newspaper archives rather than in-person scrutiny of the damage, often by engineers who considered not just the damage at face value, but also the quality of the construction. There’s strong circumstantial evidence to suggest that these tornadoes were systematically overrated—there were many strong tornadoes occurring in environments that today are not associated with strong tornadoes. Either the conditions that spawn strong tornadoes changed in the late 1970s, or the ratings changed. Most believe it is the latter, given that different rating methodologies were used pre- and post-Fujita-scale adoption. Moreover, evidence exists for the underrating of tornadoes in the past decade compared to the 1980s and 1990s. One factor that led to lower ratings was an NWS policy that went into effect in 2003 requiring that a special team of experts evaluate damage done by strong tornadoes. An unforeseen consequence of the policy was a tendency for local NWS offices to assign lower ratings in order to avoid the expense and complexity of involving external evaluators. There have been recent attempts to circumvent the issues in the tornado intensity record by using path width and length as a proxy for intensity (the assumption is that damage path dimensions are more reliable than wind speed estimates), but relating maximum wind speeds to tornado width, longevity, and forward speed (the latter two items control path length) also likely involves large error bars.
Establishing trends in the historical tornado record is also difficult owing to the large interannual variability of tornado occurrences. In the past ten years (2006–2015), the annual number of tornadoes has ranged from 886–1690 (mean 1192, standard deviation 291).
As for what is in store for the remainder of the 21st century, climate models predict, on average, small increases in the energy available for thunderstorms—called CAPE, which stands for convective available potential energy—and small decreases in vertical wind shear within the U.S. The increase in CAPE is largely due to increasing water vapor in the lower atmosphere, and the decrease in shear is due to a weakening pole-to-equator temperature gradient (this, to a large extent, dictates how much shear there is) owing to the polar regions warming faster than equatorial regions. Interestingly, however, the mean decrease in the vertical wind shear is the result of decreases in vertical wind shear on days without significant CAPE. On days with significant CAPE, which are projected to gradually become more common in the climate models, the vertical wind shear is projected to generally change little, which implies an increase in the frequency of environments conducive to severe thunderstorms (Diffenbaugh et al. 2013).
A number of caveats should be noted, however. The first involves the initiation of convection. This process is not explicitly resolved in climate models and is notoriously sensitive to small-scale details in the vertical temperature and moisture profile. If thunderstorms are initiated less frequently in the future, then the increase in the frequency of severe thunderstorm environments will be moot. Interestingly, Harold Brooks at the National Severe Storms Lab has found that past spring months that were warmer than usual (these warm periods could perhaps be the norm in a future, warmer climate) have tended to have fewer, not more, tornadoes. It is tempting to wonder whether this finding could be attributable to fewer storm initiation events during those warm periods (warm periods in Tornado Alley often, though not always, are accompanied by stronger-than-usual capping inversions ~1–2 km above the ground that suppress storm formation).
Although there is no suggestion from climate models that convection initiation will become less likely in the future, some caution in interpreting the heralded increase in severe weather environments is probably warranted given that convection initiation involves processes not explicitly resolved in the climate simulations. Moreover, the handling of convection and its initiation in the climate models (the development of convection must be parameterized in terms of processes that are resolved on the relative coarse grids of climate models) may itself influence the CAPE projections of the models, and therefore also the projections about the future frequency of severe weather environments.
A final caveat concerns the mode of severe weather. Even if severe thunderstorms become more numerous, there remain uncertainties in whether the increase in severe weather will be in the form of increased tornadoes, hail, or straight-line damaging winds. Tornadic supercell thunderstorms are favored in environments that not only possess significant CAPE and vertical wind shear, but also particularly large low-altitude vertical wind shear and relative humidity. Climate models have given some indication that the frequency of strong low-altitude vertical wind shear on significant-CAPE days will slowly increase, but the long-term trends in low-altitude relative humidity remain uncertain. Relative humidity decreases with increasing temperature and decreasing water vapor concentration—both temperature and water vapor concentration are projected to increase as the earth warms.
Summarizing, climate model projections indicate an increase in how often the environment can support severe thunderstorms, but the increase may or may not translate to more tornadoes. Whatever change there might be in tornado frequency (and perhaps intensity) is likely to evolve slowly. The changes are unlikely to be perceivable over one’s lifetime, and equally unlikely to be detected in the historical tornado record given all of its issues.