A pre-registered forecast of the spatial shift, plus two follow-on structural tests on seasonality and intensity mix. All three are reproducible from the parquet file under /data/.
predicted · longitudinal centroid increase ≥ 0.5° east
Across 75 years of F/EF1+ tornadoes, the latitude/longitude-weighted geometric center of US tornado activity has migrated roughly 1° of longitude east. The vertical drift is negligible (~0.1°), so the signature is a clean east-of-the-Mississippi shift.
predicted · per-year Southeast count ≥ 2× P1
In the seven-state Southeast corridor — Alabama, Arkansas, Georgia, Louisiana, Mississippi, Tennessee, Kentucky — tornado frequency per year has nearly tripled, while the Great Plains has stayed roughly flat. The geographic shape of "Tornado Alley" is more usefully drawn as a corridor along the Mississippi River than as a Plains rectangle.
predicted · cold-season share ≥ 15% in P3
Winter (Dec–Feb) tornado share has crept up from 8% to 11% of annual activity — a statistically real but modest broadening. Spring is now nearly half of all events (47%, up from 45%) while summer has receded from 37% to 28%. The expected cold-season blow-up did not appear, but the warm-season cooling is a stronger signal than headlines suggest.
The single most striking feature of 75 years of US tornado data is also its simplest summary: take the average latitude and average longitude of every F/EF1+ tornado touch-down in two 35-year windows, and the second window's center sits about 70 miles east of the first's.
The shift is concentrated at the longitude axis. Latitude drifts only 0.10° north. The clean east-only signature rules out the simplest "warming tilts the jet stream further south" narrative and instead points at atmospheric moisture loading in the warm sector of mid-latitude systems — a hypothesis consistent with the climate-change literature (see paper §6).
The classic Tornado Alley — a six-state rectangle covering Texas, Oklahoma, Kansas, Nebraska, and the Dakotas — held its own. Per-year tornado counts are statistically indistinguishable between 1950–1985 and 2001–2026 (an increase of about 6% that's well inside the year-to-year noise).
The seven-state Dixie Alley tell a completely different story. Tornado frequency per year has nearly tripled, from 138 to 376. The Southeast now matches the Great Plains in volume, with the addition of an "out-of-season" (winter and autumn) signature that the Plains does not generate.
Crucially, the gain in the Southeast is in the absolute count of significant (EF1+) tornadoes, not just in the small and arguably-overreported EF0 bins. The shift is real.
Eight hex-bin density maps showing every reported F/EF0+ tornado touch-down by decade. The bright spots move. The 1950s–1970s read as a "Plains drench." By the 2000s and 2010s, the densest clusters run along the Mississippi Valley and the Mid-South.
The ratio-of-ratios: P3 (2001–2026) per-year mean ÷ P1 (1950–1985) per-year mean, filtered to F/EF1+ and only over states with at least 200 significant tornadoes across the full record.
The biggest gainer is Kentucky (2.81×); biggest loser Wyoming (0.50×). The table tells the migration story more precisely than any aggregate metric.
| Top 5 gainers | P1/yr | P3/yr | Ratio | |
|---|---|---|---|---|
| Kentucky | EF1+ tornadoes | 8.7 | 24.5 | 2.81× |
| Mississippi | EF1+ tornadoes | 20.9 | 53.0 | 2.54× |
| Virginia | EF1+ tornadoes | 4.2 | 10.1 | 2.4× |
| New York | EF1+ tornadoes | 2.5 | 5.8 | 2.28× |
| Alabama | EF1+ tornadoes | 20.6 | 44.4 | 2.16× |
| Top 5 losers | P1/yr | P3/yr | Ratio | |
|---|---|---|---|---|
| Wyoming | EF1+ tornadoes | 4.7 | 2.4 | 0.5× |
| Colorado | EF1+ tornadoes | 10.9 | 5.6 | 0.51× |
| Michigan | EF1+ tornadoes | 13.1 | 7.0 | 0.53× |
| Texas | EF1+ tornadoes | 73.3 | 42.7 | 0.58× |
| Florida | EF1+ tornadoes | 24.0 | 14.8 | 0.62× |
NOAA and academic meteorology have spent two decades publishing maps that look like Texas. The data argues those maps should now look like a corridor along the Mississippi River — starting at the Arkansas/Louisiana border, running through Tennessee, Kentucky, and into southern Indiana and Ohio. The activity hasn't disappeared; it has moved.
What this paper offers: a precise quantitative description of that shift (a 70-mile east centroid migration, a 2.7× Southeast gain, a still-flat Plains), grounded in the entire NOAA Storm Events record from 1950 to 2024 and reproduced from the data shipped with this site.
Annual-trend series, seasonal split, intensity mix, casualty timeline, and the centroid migration arrow. All built in matplotlib from the same parquet.
