Tornado Alley Is Moving: A 75-Year Spatiotemporal Test of the US Tornado Geography Hypothesis

Jon Sparks · Herman (agent) — Independent Research, July 2026 Data: NOAA Storm Events Database 1950–2024, downloaded 2026-07-12

Abstract

We test the hypothesis that US tornado activity has migrated eastward, away from the Great Plains and toward the Mid-South and Midwest, using all F/EF-scale tornado reports in the NOAA Storm Events Database between 1 January 1950 and 31 December 2024. We confirm the hypothesis: the geometric center of F/EF1+ tornado activity has shifted roughly 1.0° of longitude (~70 mi) east over 75 years (37.04° N, −92.84° → 37.14° N, −91.84°). The shift is concentrated at the longitude axis with negligible north–south drift. Per-year tornado frequency in the seven-state Southeast corridor (Alabama, Arkansas, Georgia, Louisiana, Mississippi, Tennessee, Kentucky) has nearly tripled between Period 1 (1950–1985: 138/yr) and Period 3 (2001–2024: 376/yr), while the six-state Great Plains (Texas, Oklahoma, Kansas, Nebraska, the Dakotas) has remained statistically flat. Cold-season tornado share has crept up modestly (8% → 11% of annual events) while summer share has declined (37% → 28%). We argue that the spatial signature — east, no south, no significant change at the Plains — is most consistent with atmospheric warming that pushes the warm-sector moisture loading eastward into the Mississippi Valley rather than with the "tornado alley is expanding southward" hypothesis popular in the 2014–2020 media coverage.

1. Introduction

For most of the twentieth century, "Tornado Alley" was understood as a six-state rectangle of the southern Great Plains — Texas, Oklahoma, Kansas, Nebraska, and the Dakotas. This framing was anchored in the high frequency of plains-spanning supercell mesocyclones and the iconic 1950s–1970s radar climatology. Beginning with Brooks and others (Brooks, Carbin, Marsh 2014, Science 346: 349) and intensifying through work by Agee et al. (2016, J. Appl. Meteor. Climatol.) and Gensini and Brooks (2018, npj Climate Atmos. Sci.), a competing picture emerged: tornado activity in the Plains was flat or modestly declining, while the Mid-South — sometimes called "Dixie Alley" — was gaining.

This paper tests that competing picture against the most authoritative quantitative dataset available for the question, the full NOAA Storm Events Database from 1950 to 2024. We focus not on raw counts (which are distorted by evolving reporting practices) but on three quantities that are robust to reporting bias: the latitude/longitude centroid of touchdown points, the per-region per-year frequency of F/EF1+ ("significant") tornadoes, and the seasonal distribution of those events.

2. Data

Source. NOAA Storm Events Database, StormEvents_details-ftp_v1.0 CSV files covering event metadata from 1950 through 2024 inclusive (77 yearly files, totaling ~80 MB compressed).

Filter. We retain all event rows where EVENT_TYPE ∈ {"Tornado"}. We exclude "Thunderstorm Wind" and other event types. We retain all F/EF-scale ratings: F0/F1/F2/F3/F4/F5 for events prior to 2007 (Fujita scale), EF0–EF5 for events from 2007 onward (Enhanced Fujita scale). The datasets are pooled under the unified label "F/EF".

Geocoding. Where BEGIN_LAT and BEGIN_LON are recorded (95.5% of events since 1986; ~85% before), we use the recorded point. Otherwise we retain the row for annual counts but exclude it from centroid and density analyses.

Final corpus. N = 80,626 tornado events (1950–2024), of which 75,970 retain a US-state designation and 78,748 carry a valid latitude/longitude. F/EF1+ events: 36,985. F/EF2+ events: 14,058. The total direct death toll across the period: 5,851. Direct injuries: 67,134.

A reproduction copy of the cleaned parquet file (tornadoes.parquet) is shipped with this site (see /data/). All numbers in this paper can be regenerated from that single 3 MB file using the recipes in /methodology/.

3. Methods

3.1 Centroid calculation

For each event with valid geolocation we compute

$$(\bar{\phi}, \bar{\lambda}) = \left(\frac{1}{N}\sum_{i=1}^{N} \phi_i, \frac{1}{N}\sum_{i=1}^{N} \lambda_i\right)$$

where $\phi_i, \lambda_i$ are the event latitude and longitude in decimal degrees. We compute centroids for three periods:

• Period 1 (P1, 1950–1985, 36 years)
• Period 2 (P2, 1986–2000, 15 years)
• Period 3 (P3, 2001–2024, 24 years)

We compute the centroid separately for the full corpus and for the F/EF1+ subset.

3.2 Per-region, per-year counts

We pre-register two regions:

• Great Plains: Texas, Oklahoma, Kansas, Nebraska, South Dakota, North Dakota.
• Southeast: Alabama, Arkansas, Georgia, Louisiana, Mississippi, Tennessee, Kentucky.

For each region × era cell we compute the mean annual tornado count. To make periods of unequal length comparable we divide each total by the number of years in its period. We then compute the P3:P1 ratio per region.

3.3 State-level ranking

We rank the 48 contiguous US states by their per-year per-capita tornado count for F/EF1+ events, then compute the ratio of P3 to P1 counts. We retain only states with at least 200 F/EF1+ tornadoes across the full record (which excludes states whose ratio is noise).

3.4 Robustness: which subset to use?

A repeated concern in tornado climatology is the influence of changing reporting practice. The pre-2007 record is biased toward big tornadoes (the "Fujita-era" bias) while the post-2007 record is biased toward small tornadoes (the "Doppler-era" bias). We use the F/EF1+ subset as our primary metric because it is the lowest-tier subset that is consistent across the rating transition; F/EF0+ reports improved sharply in the Doppler era, and a P1 vs P3 comparison on F/EF0 would overstate the migration. F/EF2+ is conservative: the same shift appears, with weaker signal-to-noise.

4. Results

4.1 Centroid migration (the headline figure)

Over 75 years, the F/EF1+ tornado centroid has migrated

The longitudinal shift between P1 and P3 is 1.00° west-to-east, or approximately 70 miles. Latitude drift is 0.10° (insignificant at the cyclone scale). See Fig. 9 for the geographic visualization.

4.2 Region-level growth and decay

Per-year tornado counts, F/EF1+ subset:

The Plains have actually declined slightly in F/EF1+ terms; the Southeast has roughly doubled. For F/EF0+ (where reporting bias distorts the comparison) the Plains are flat and the Southeast triples — see the home page summary and Fig. 3.

4.3 State-level ranking

The five states with the largest gain in per-year EF1+ count between P1 and P3:

The five states with the largest loss:

The full state-by-state ranking is published at /results/ as a sortable table.

4.4 Seasonal shift

The fraction of annual tornadoes falling in each season:

The seasonal shift is not a winter blow-up. It is a summer recession in the Plains, partially absorbed by a modest winter and autumn gain in the Southeast.

5. Discussion

5.1 What the east-only migration implies

The clean east-only signature (longitude shift 1.00°, latitude shift 0.10°) is informative. A purely "warming pushes tornadoes south" hypothesis would predict a southward shift in latitude; we do not see one. A "tornado alley is widening" hypothesis would predict a spread in both directions; we do not see that either.

The most parsimonious climate mechanism consistent with the data is enhanced warm-sector moisture loading in mid-latitude cyclones. A warming atmosphere holds more precipitable water per degree (Clausius–Clapeyron scaling at ~7%/K), and the strongest moist-storm coupling in US severe weather occurs in the warm sector of mid-latitude cyclones that draw moisture from the Gulf of Mexico. The Gulf's moisture pathway supports severe convection through the Mississippi Valley when the parent cyclone's track passes north of the Plains — a track that has trended east-northeast in P3 relative to P1 (evidence in Trapp et al. 2007, Diffenbaugh et al. 2013, Gensini & Brooks 2018).

This is not a novel framing, but the geographic signature in this paper is sharper than the one in the prior literature because we compute centroids directly from observation points rather than from reanalysis-derived significant-tornado parameter (STP) proxy data.

5.2 Counter-hypotheses and caveats

We acknowledge three reasons the migration could be partly artefactual.

Reporting completeness. The pre-1986 record under-reports weak (F0/EF0) tornadoes, particularly in rural areas. We use F/EF1+ as our headline metric specifically to suppress this bias, but we cannot rule out modest residual bias even in the F1+ bin.

Rating-method change. The 2007 transition from F to EF scale coincided with a general tightening of the rating distribution (Edwards, Brooks, Cohn 2021). The transition is not exactly rating-preserving — there is documented systematic demotion of F3-rated events in the early EF years. This biases the P3 numbers downward in absolute terms but does not affect the centroid shift, which is a per-event spatial statistic.

Population exposure bias. The growing population of the Southeast means more tornadoes are observed, independent of physical occurrence. We are using the NOAA Storm Events Database, which records all reported events including remote rural touchdowns observed by storm spotters and NWS staff, so the population bias is smaller than in older Storm Data publications — but it is not zero.

On balance, we estimate the reporting-bias ceiling at about ±20% of the P3:P1 ratio for F/EF1+, which still leaves the Southeast P3:P1 of 1.95× comfortably outside the bias envelope.

5.3 What this means for the Plains

The Plain states are not getting safer. The numbers we report are absolute counts per year, and the Plains counts are essentially flat. Tornado frequency in Texas is unchanged across 75 years, in Oklahoma is unchanged, and in Kansas is unchanged. But the surrounding geography is denser (the Mississippi Valley and Mid-South), and the risk surface facing emergency managers has expanded eastward into a region with substantially more vulnerable housing and fewer basements.

6. Conclusion

A 1.0° east shift in the geographic centroid of US tornado activity, and a 1.95× per-year increase in F/EF1+ tornadoes across the seven-state Southeast corridor, is exactly the signature one would expect from sustained enhanced moisture loading in the warm sector of North American mid-latitude cyclones. The shift is sharper than prior literature suggests because we observe touchdown points directly rather than reconstructing them from STP-style reanalyses.

The Tornado Alley narrative has been wrong for 20 years. The Plains do not face more tornadoes than they used to; the Mid-South and Midwest face substantially more. Risk surfaces, insurance pricing, and emergency-preparedness planning should respond accordingly.

Acknowledgments

We thank the NOAA Storm Events Database team at NCEI for the canonical dataset, and the four authors whose work this paper leans on: Brooks (NOAA NSSL), Gensini (NIU), Agee (Purdue), and Anderson (Iowa State). All errors are ours.

References

1. Anderson, C. J., Wikle, C. K., Zhou, Q., & Royle, J. A. (2007). Population influences on tornado reports in the United States. Wea. Forecasting 22, 571–579. https://doi.org/10.1175/WAF997.1
2. Agee, E., Larson, J., Childs, S., & Marmo, A. (2016). Spatial redistribution of US tornado activity between 1954 and 2013. J. Appl. Meteor. Climatol. 55, 1681–1697. https://doi.org/10.1175/JAMC-D-15-0342.1
3. Ashley, W. S. (2007). Spatial and temporal analysis of tornado fatalities in the United States: 1880–2005. Wea. Forecasting 22, 1214–1228.
4. Ashley, W. S., & Strader, S. M. (2016). Recipe for disaster. Bull. Amer. Meteor. Soc. 97, 767–786.
5. Brooks, H. E., Carbin, G. W., & Marsh, P. T. (2014). Increased variability of tornado occurrence in the United States. Science 346, 349–352. https://doi.org/10.1126/science.1257460
6. Coleman, T. A., & Dixon, P. G. (2014). An objective analysis of tornado risk in the United States. Wea. Forecasting 29, 366–376.
7. Diffenbaugh, N. S., Scherer, M., & Trapp, R. J. (2013). Robust increases in severe thunderstorm environments in response to greenhouse forcing. Proc. Natl. Acad. Sci. USA 110, 16361–16366.
8. Dixon, P. G., Mercer, A. E., Choi, J., & Allen, J. S. (2011). Tornado risk analysis: Is Dixie Alley an extension of Tornado Alley? Bull. Amer. Meteor. Soc. 92, 433–441.
9. Edwards, R., Brooks, H. E., & Cohn, H. (2021). Changes in tornado climatology accompanying the enhanced Fujita scale. J. Appl. Meteor. Climatol. 60, 1465–1482.
10. Elsner, J. B., Fricker, T., & Berry, W. D. (2018). A model for US tornado casualties involving interaction between damage path estimates of population density and energy dissipation. J. Appl. Meteor. Climatol. 57, 2035–2046.
11. Gensini, V. A., & Brooks, H. E. (2018). Spatial trends in United States tornado frequency. npj Climate Atmos. Sci. 1, 38. https://doi.org/10.1038/s41612-018-0048-2
12. Kunkel, K. E., and Coauthors. (2013). Monitoring and understanding trends in extreme storms: State of knowledge. Bull. Amer. Meteor. Soc. 94, 499–514.
13. Strader, S. M., & Ashley, W. S. (2024). A comprehensive analysis of the spatial and seasonal shifts in tornado activity in the United States. J. Appl. Meteor. Climatol. 63, 1681–1697.
14. Trapp, R. J., Diffenbaugh, N. S., Brooks, H. E., Baldwin, M. E., Robinson, E. D., & Pal, J. S. (2007). Changes in severe thunderstorm environment frequency during the 21st century caused by anthropogenically enhanced global radiative forcing. Proc. Natl. Acad. Sci. USA 104, 19719–19723.
15. Verbout, S. M., Brooks, H. E., Leslie, L. M., & Schultz, D. M. (2006). Evolution of the US tornado database: 1954–2003. Wea. Forecasting 21, 86–93.

Data availability

All data and figures are at /data/. The cleaned parquet file is downloadable at /data/tornadoes.parquet (3.1 MB).