Tornado Alley is drifting south-southeast, not racing east

Seventy-six years of significant US tornadoes show a slow, real migration toward the Southeast, and a Great Plains in genuine decline. The popular "moving east" figure is three-to-seven times too fast.

TerraPulse paper DP-002. Pre-registration: docs/scope-tornado-alley-east.md (frozen 2026-06-18). Data, scripts, and results.json in this workspace. Status: draft.

Abstract

Press coverage and several research summaries describe "Tornado Alley" as shifting eastward at roughly 10 to 20 km per year. We test that claim against the full Storm Prediction Center tornado record, 1950 to 2025, using only significant tornadoes (Enhanced Fujita scale 2 and above, N = 13,400). Restricting to strong tornadoes removes the dominant confound in the record: the four-fold inflation of weak-tornado reports caused by improved detection. We track the path-length-weighted geographic centroid of significant-tornado activity each year and fit its trajectory with cluster-bootstrap uncertainty resampled by convective day.

The centroid moves east at 2.7 km per year (95% confidence interval 1.5 to 3.6) and south at 2.3 km per year (95% confidence interval 3.3 to 1.4). The two components are statistically equal: over 76 years the center of significant-tornado activity has moved about 203 km east and 179 km south, a near-diagonal drift toward the Southeast, not a due-east march. The measured eastward rate is three to seven times slower than the circulated 10 to 20 km per year figure.

Decomposing the shift by region shows it is genuine and mechanistic, not a reporting artifact. The Southeast ("Dixie Alley": AR, LA, MS, AL, TN, GA, KY, MO) raised its share of significant tornadoes from 25% to 40% while holding a roughly flat absolute count. The Great Plains (TX, OK, KS, NE) fell in both share (36% to 24%) and absolute count (74 to 32 significant tornadoes per year). Because national significant-tornado counts also declined over this period, we tested whether the Plains drop is merely that national decline: it is not. One region rose while the other fell, which a uniform national trend cannot produce. The eastward signal is robust to the choice of tornado location point, to leaving out any single decade, and to the start year. We find no clean single change-point; the drift is gradual.

1. Why this question matters

"Is Tornado Alley moving east?" is one of the most-shared claims in severe-weather science. It has appeared in major newspapers and is supported in part by peer-reviewed work on tornado environments. It matters because the Southeast is more vulnerable than the Great Plains to the same tornado: higher population density, more mobile homes, more nocturnal tornadoes, and more tree cover that hides them. A real migration of activity toward the Southeast moves tornadoes toward the people least equipped to survive them.

But the headline number that circulates, on the order of 10 to 20 km per year, has outrun the evidence. We set out to measure the migration of significant-tornado activity directly, at the full 76-year scale of the US record, with the methods pre-registered before we looked at the result.

2. Data

We use the Storm Prediction Center Tornado History database, 1950 to 2025, as ingested into TerraPulse (73,458 tornadoes; source spc_tornado_history). Each record carries a start location, a track length, a state, and an Enhanced Fujita rating. We take the start point as the tornado's location and weight each tornado's contribution to the annual centroid by its track length, because long-track tornadoes are the least likely to be missed in any era.

We restrict the primary analysis to significant tornadoes, rated EF2 or stronger (N = 13,400 after cleaning). This is the central design choice. The US tornado record is dominated by a detection explosion: recorded EF0 tornadoes rose from about 4,200 in 1950 to 1975 to about 17,300 in 2000 to 2025, a four-fold increase driven by spotter networks, Doppler radar, and smartphones, not by more tornadoes. Over the same period, EF2 counts fell. Strong tornadoes were essentially never missed, even in 1950, so their geography is trustworthy across the whole record. We report EF1-and-above (N = 38,565) and all-tornado (N = 71,832) trajectories as a sensitivity ladder, never as the headline.

We exclude 1,546 tornadoes with an unknown rating (all modern) and 80 tornadoes outside a contiguous-US bounding box or with missing coordinates.

3. Methods

All methods were frozen in the pre-registration before extraction. For each year we compute:

the path-length-weighted geographic centroid (latitude, longitude) of significant tornadoes, with an unweighted count centroid as a sensitivity check;
a variance ellipse (semi-major axis, semi-minor axis, orientation) describing the spread and tilt of activity;
the share and absolute count of significant tornadoes in the Great Plains, the Southeast, and the rest of the country.

We fit the centroid trajectory with ordinary least squares over the full record and in rolling 30-year windows, converting longitude degrees to kilometers at the mean centroid latitude so the eastward rate can be compared directly to the published figure.

For uncertainty we use a cluster bootstrap resampled by convective day, because tornadoes on the same day belong to the same storm system and are not independent. The 13,400 significant tornadoes collapse to 5,044 independent storm-days, a 2.7-fold reduction in effective sample size. We resample whole days (1,000 draws) to obtain confidence intervals on every annual centroid and on the trend slopes. The naive tornado-level bootstrap is reported alongside so the cost of clustering is explicit: it understates the eastward interval by about a third.

4. Results

4.1 The migration is south-southeast, and slow

The center of significant-tornado activity moved east at 2.67 km per year (95% CI 1.54 to 3.62) and south at 2.35 km per year (95% CI 3.31 to 1.36) over 1950 to 2025. Both intervals exclude zero. The two components are the same size. Integrated over 76 years the centroid moved roughly 203 km east and 179 km south, a bearing about 41 degrees south of due east. The correct one-line description is "drifting toward the Southeast," not "moving east."

The southward component is as robust as the eastward one. Leaving out any single decade, the southward rate stays between 1.8 and 3.1 km per year and never crosses zero; moving the start year from 1950 to 1960 only strengthens it. A paper that reported only the eastward number would be hiding half of a clean, two-dimensional signal.

4.2 The popular figure is three-to-seven times too fast

The circulated rate of 10 to 20 km per year is not supported. Our measured eastward rate of 2.67 km per year sits a factor of 3.7 to 7.5 below it; even the upper bound of our confidence interval (3.62 km per year) is less than 40% of the lower bound of the popular range. We measure the centroid of significant-tornado activity; if the circulated figure refers to a different object, such as the eastern edge of tornado-favorable environments, the two are not the same quantity. Either way, the center of where strong tornadoes actually touch down is moving far more slowly than the headline suggests.

4.3 The mechanism: Southeast expansion riding on Plains decline

The shift decomposes cleanly into two regional stories.

The Great Plains is declining in both share and absolute number. Significant tornadoes per year fell from 74 (first 15 years) to 32 (last 15 years); the Plains share of US significant tornadoes fell from 36% to 24%.
The Southeast is expanding in share while holding steady in absolute number. Its share rose from 25% to 40%, on a roughly flat absolute count (50 to 58 per year).

This pattern survives the most important objection. National significant-tornado counts also declined over the record (to about 0.69 times their early level), partly because of how the Enhanced Fujita scale, introduced in 2007, rates strong tornadoes. If the Plains decline were merely this national deflation, every region would fall together. Instead, the Southeast rose while the Plains fell about 1.6 times faster than the national trend. A uniform national decline cannot make one region go up and another go down. The decomposition is the spine of this paper, and it is robust to reasonable changes in which states are assigned to each region.

4.4 No clean regime shift

A single-change-point search on the longitude trajectory returns 1963 for significant tornadoes but 1983 for the EF1-and-above and all-tornado populations, and a step model explains only about three percentage points more of the year-to-year variance than a straight line (R-squared 0.15 versus 0.12). We therefore report no clean single change-point: the drift is gradual and continuous, not a sudden jump. Reporting a specific break year would be reading structure into noise.

4.5 Sensitivity ladder

The eastward rate is stable across populations: EF2+ 2.67 km per year (CI 1.54 to 3.62), EF1+ 3.28 (CI 2.32 to 4.12), all tornadoes 3.20 (CI 2.33 to 4.06). Using the track midpoint or end point instead of the start point changes the eastward rate by less than 0.2 km per year (2.67 to 2.51 to 2.34), so the result is not an artifact of the location convention. None of the populations, and none of the location choices, approaches the popular 10 to 20 km per year figure.

5. Discussion

The headline that Tornado Alley is "moving east" is directionally right and quantitatively wrong. The center of significant-tornado activity is migrating, but slowly, diagonally toward the Southeast, and at a fraction of the circulated speed. The more important finding is the mechanism: this is not a rigid alley sliding east across the map. It is the Great Plains losing strong tornadoes in absolute terms while the Southeast gains share, two regional trends that happen to move the national centroid southeast.

That distinction has consequences. The Southeast already suffers disproportionate tornado fatalities for a given storm strength. A shift of activity toward it, even a slow one, compounds an existing vulnerability. The public-safety story is not "the alley is racing east"; it is "the country's strongest tornadoes are concentrating where they do the most harm."

6. Limitations

The pre-1953 record is thin and its early-year centroids carry wide confidence intervals, which we show rather than smooth away. Enhanced Fujita ratings for pre-2007 tornadoes are retrospective mappings from the older Fujita scale; we use the published ratings as-is and note that the 2007 scale change contributes to the national decline in significant-tornado counts. We measure activity centroids, not environmental favorability, and we do not apply a population-density correction to the EF2+ headline, on the argument that strong-tornado reporting is near-complete throughout the record. All numbers are reproducible from the pre-registered scripts with a fixed random seed.

7. Conclusion

Across 76 years of US significant tornadoes, the center of activity has drifted about 203 km east and 179 km south, a slow south-southeast migration toward the Southeast. The Great Plains is in genuine decline; the Southeast is expanding its share. The widely circulated "10 to 20 km per year eastward" figure overstates the real rate by three-to-seven-fold. The honest picture is slower, more diagonal, and more concerning for the people in its path than the headline version.

Companion artifacts to build: animated year-by-year centroid-and-ellipse trajectory map; the Plains-versus-Southeast share crossover chart; the sensitivity-ladder figure.