The regionality and seasonality of tornado trends in the United States

Nearly a decade ago, Brooks et al. documented a dichotomous trend in U.S. tornado activity: fewer tornado days per year but more tornado outbreak days per year when aggregated across the contiguous United States (CONUS). This suggested increased clustering of tornadoes in time. Building on that finding, the present study asks when during the calendar year and where geographically such changes in tornado days and tornado outbreak days have occurred, given that preparedness, vulnerability, and socio-economic conditions vary seasonally and regionally (e.g., Southern Great Plains versus the Southeast U.S.). A further question is whether the trends identified a decade ago have continued into the most recent decade. This is pertinent in the context of anthropogenic climate change (ACC), which is projected to increase the frequency of environments supportive of intense, potentially tornadic thunderstorms, but also in light of possible multi-decadal modulation by internal climate variability and teleconnections. Using the NOAA Storm Prediction Center (SPC) tornado reports over 1960–2022, the study extends previous analyses by nine years to address these interrelated questions with finer seasonal and regional granularity.

Prior work (Brooks et al., 2014) identified increased variability in U.S. tornado occurrence with fewer tornado days but more outbreak days, motivating investigation into temporal clustering. Subsequent studies examined spatial trends and regional shifts in tornado frequency (e.g., Gensini and Brooks, 2018; Moore, 2017; Moore and DeBoer, 2019; Agee et al., 2016; Moore, 2018), including earlier peak tornado activity dates in parts of the Great Plains (Long and Stoy, 2014) and enhanced risk in the Southeast (Dixon et al., 2011). The definition of tornado outbreaks has evolved and lacks a universal standard (Cwik et al., 2021), and evidence suggests more tornadoes in the most extreme outbreaks (Tippett et al., 2016). Projections under ACC indicate robust increases in severe thunderstorm environments (Del Genio et al., 2007; Trapp et al., 2007; Diffenbaugh et al., 2013; Hoogewind et al., 2017). Internal variability and teleconnections, including ENSO and North Atlantic SST variability, have been linked to regional tornado outbreaks and environmental predictability (Lee et al., 2016; Niloufar et al., 2021; Tippett et al., 2022). Changes in the jet stream and planetary wave dynamics have been proposed as ACC-related mechanisms (Mann et al., 2017; Trapp and Hoogewind, 2018). Reporting practices and dataset limitations also affect observed trends (Trapp, 2013).

Data: NOAA Storm Prediction Center severe weather database of georeferenced U.S. tornado reports (CONUS), 1960–2022, including time, date, and (E)F rating. EF0 tornadoes were excluded due to rating uncertainty and minimal impacts; the analysis focuses on EF1+ reports. Primary metrics: • Tornado day: any calendar day with at least one EF1+ tornado report. • Tornado outbreak day: any calendar day with more than 30 EF1+ tornadoes. To address the small annual sample sizes produced by the >30 threshold, outbreak-day trends were recomputed using lower thresholds of >20 and >10 EF1+ tornadoes. Temporal disaggregation: Annual tornado days were disaggregated by month to assess seasonal dependencies, following approaches in prior work for increased temporal granularity. Seasons: Warm season defined as April–July; cool season as November–February, justified by seasonality diagnostics and prior literature. Spatial domains: Analyses performed for U.S. states and for two regional domains: Southern Great Plains and Southeast (as in Fig. 3), to assess regional contributions by season. Period stratification: Three periods were compared—1960–1979 (period 1), 1980–1999 (period 2), and 2000–2022 (period 3)—for frequencies of days exceeding multiple tornado-count thresholds (>10, 15, 20, >25). Statistical analysis: • Linear trends in counts of tornado days and outbreak days computed with p-values for significance. • Monthly linear trends evaluated with associated p-values. • State-level warm/cool-season trends in tornado days reported as gains/losses per decade, with significance indicated. • Two-sample t-tests evaluated changes in frequencies across periods per threshold and season, with significant increases/decreases marked. Variability analysis: Interannual variability assessed via annual differences of EF1+ tornado reports for CONUS and the two regions/seasons. Volatility quantified by a 15-year rolling standard deviation of annual differences, following Tippett (2014), and sensitivity to anomalous years (e.g., 2011) tested. Reproducibility: Data available from SPC; analysis code available on GitHub (provided by authors).

• National trends (1960–2022): − Tornado days: long-term linear trend is negative, with a statistically significant decrease of 1.03 days per year. − Tornado outbreak days: long-term linear trend is positive, with a statistically significant increase of 0.04 days per year (using >30 EF1+ threshold). − Using lower outbreak thresholds (>20 and >10 EF1+), trends remain positive and statistically significant. • Recent decades: − Since 2000, tornado days show a weak, statistically insignificant positive trend of 0.11 per year (p=0.73), influenced by high activity in 2017. − Since 2000, outbreak days (>30 EF1+) show a very weak, insignificant positive trend of 0.002 per year (p=0.97), affected by a lack of outbreaks in the late 2010s. − A trend reversal in outbreak days in the most recent decade is evident for the >20 threshold; the >10 threshold shows strong multi-decadal variability without a clear reversal. • Seasonal disaggregation of tornado days: − Significant negative trends in tornado days occur from March through September, with especially large losses in June–August. − No significant trends October–February. − Peak U.S. tornado-day probability shifted earlier by about three weeks, from 14 June (1960–1980) to 24 May (2000–2021). • Regional and seasonal patterns: − Warm season (Apr–Jul): • Texas exhibits the largest state-level decrease: −3.31 tornado days per decade (statistically significant). • Broad decreases in warm-season tornado days across the southern and northern Great Plains; EF1+ report trends are also negative there. • In the Southeast, warm-season EF1+ reports have increased over the past three decades, implying a contribution to positive outbreak-day trends. − Cool season (Nov–Feb): • CONUS shows no overall trend in tornado days, but Southeast states generally show increases in tornado days and EF1+ reports (amplified over the last three decades). • No state has a statistically significant decrease in tornado days in the cool season. • “Fewer days, more tornadoes” vs. “fewer days, fewer tornadoes”: − Southeast: • Warm season: tornado days decreased from period 1 to 2 and from 2 to 3, while frequencies of days with many tornadoes (>10, 15, 20, >25) significantly increased from period 2 to 3. • Cool season: tornado days show no significant overall trend, but days with many tornadoes significantly increased. • Conclusion: Southeast contributes to “fewer days, more tornadoes.” − Southern Great Plains: • Warm season: tornado days significantly decreased from period 1→2 and 2→3; limited increases only for >15 threshold from period 2→3; overall EF1+ reports trend negative. • Cool season: no significant increases in days with many tornadoes; EF1+ reports trend negative. • Conclusion: Southern Great Plains exhibits “fewer days, fewer tornadoes.” • Variability and volatility: − Interannual variability is larger in warm seasons for both regions. − CONUS volatility (15-year rolling SD of annual differences) increases sharply in the 2000s, largely due to Southeast warm-season activity and the anomalously active year 2011; removing 2011 yields a more gradual, multi-decadal cycle (decrease in 1970s, increase in recent decades). • Overall: Indications that the historical dichotomy (fewer tornado days, more outbreak days) has relaxed in the most recent decade, warranting caution and further study of drivers.

Temporal and geospatial disaggregation shows that long-term decreases in annual tornado days since 1960 primarily occur during boreal spring and summer, most dramatically in June–August, with prominent warm-season losses in Texas and the broader Southern Great Plains. These changes have shifted seasonality, moving the peak tornado probability from mid-June to late May. In contrast, both warm- and cool-season activity in the Southeast U.S. exhibit significant increases in the frequency of days with many tornadoes (outbreaks), supporting a “fewer days, more tornadoes” tendency, regardless of whether outbreaks are defined using >10, >20, or >30 tornado thresholds. Given the high vulnerability in the Southeast, improving understanding and predictability of outbreaks is particularly important. There are indications that the dichotomous national trends in tornado days and outbreak days have relaxed over the most recent decade. Whether this reflects internal climate variability alone or interactions with anthropogenic climate change remains an open question. Potential mechanisms include ACC-driven changes in large-scale circulation (e.g., jet stream strength and variability tied to meridional temperature gradients) and modulation by multi-decadal climate modes. The sensitivity of volatility to single anomalous years underscores the need for caution when attributing recent trend changes.

This study extends national tornado-trend analyses through 2022 with seasonal and regional resolution, revealing that warm-season tornado days have significantly declined—especially in the Southern Great Plains—while the Southeast shows increases in days with many tornadoes in both warm and cool seasons. Collectively, the Southeast exhibits a “fewer days, more tornadoes” pattern, whereas the Southern Great Plains shows “fewer days, fewer tornadoes.” The historical dichotomy between fewer tornado days and more outbreak days appears to have relaxed in the last decade, but attribution remains uncertain. Future work should: • Quantify the relative roles of internal climate variability versus anthropogenic forcing in modulating tornado activity. • Explore physical mechanisms linking large-scale circulation changes to regional tornado environments. • Enhance predictability and risk assessment of high-impact outbreaks, particularly in vulnerable Southeast communities. • Complement report-based analyses with meteorological proxies and model-based diagnostics to mitigate reporting biases and extend insight beyond the observational record.

Findings rely on report-based tornado datasets that have known biases, nonstationary reporting practices, and relatively short record lengths. EF0 events were excluded due to rating uncertainty, with a small risk that some strong tornadoes were misclassified as EF0; this is expected to have minimal impact. Outbreak definitions (e.g., >30 EF1+) yield small annual sample sizes, limiting statistical power and increasing sensitivity to year-to-year fluctuations. Volatility results are sensitive to anomalous years (e.g., 2011), highlighting the need for caution in interpretation and attribution.