Moore (2025) Targeted decomposition of tornado records reveals long-term trends in the Great Plains and Southeast United States

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-10-22
• Authors: Todd W. Moore
• DOI: 10.1038/s41598-025-20868-z

Research Groups

• Department of Geosciences, Middle Tennessee State University, Murfreesboro, TN, USA

Short Summary

This study developed a novel domain-informed decomposition framework to isolate long-term trends in tornado activity in the Great Plains and Southeast United States, revealing a robust decline in the Great Plains and an increase in the Southeast after explicitly filtering out teleconnection-related variability and noise.

Objective

• To develop and apply a domain-informed, data-driven decomposition framework to isolate and statistically test long-term trends in (E)F1+ tornado activity in the Great Plains and Southeast United States, after removing seasonal cycles and noise related to key teleconnections.

Study Configuration

• Spatial Scale: Great Plains (30–41°N, 94–105°W) and Southeast United States (30–41°N, 82–93°W).
• Temporal Scale: January 1960 to December 2022 (63 years).

Methodology and Data

• Models used:
  • Fourier analysis (using periodogram function in R's TSA package) to identify dominant periods in teleconnection time series.
  • Cross-wavelet coherence analysis (using R's WaveletComp package) to assess statistical coherence between teleconnections and tornado activity.
  • Multi-seasonal decomposition methods: MSTL (Multiple Seasonal-Trend decomposition using LOESS) and TBATS (Trigonometric Box-Cox Autoregressive Trend Seasonal) (from R's forecast package) to extract targeted seasonal cycles, noise, and long-term trends.
  • Statistical trend tests: Mann-Kendall trend test (from R's Kendall package) and Theil-Sen estimator (from R's zyp package) to quantify and assess the significance of trends.
• Data sources:
  • (E)F1+ tornado dataset from the Storm Prediction Center (SPC).
  • Monthly teleconnection indices (ENSO, NAO, AMO, PNA, PDO, AO) from the National Oceanic and Atmospheric Administration’s (NOAA) Physical Science Laboratory (PSL).

Main Results

• Dominant periodicities in teleconnection indices ranged from intra-annual to multidecadal cycles (e.g., 17, 42, 70, 55, 59, 64, 154, 384 months).
• Key cycles (e.g., 42, 64, 154, 384 months) showed statistically significant coherence with tornado counts in the Great Plains and Southeast regions.
• Both MSTL and TBATS decomposition methods consistently identified a dominant 12-month cycle and multi-year cycles (e.g., 42, 64, 154 months), with multi-year cycles being more pronounced in the Southeast.
• Seasonal components in the Southeast showed growing amplitudes over time, suggesting an increasing influence of these seasonal cycles and teleconnections on tornado activity in the region.
• After removing variability and noise, the long-term trend components from both decomposition methods showed a decline in the Great Plains and an increase in the Southeast.
• Annually aggregated trends indicated a linear decrease in the Great Plains of approximately -1 tornado per year (MSTL: -1 tornado/year, TBATS: -1.1 tornadoes/year; both p < 0.001).
• Annually aggregated trends indicated a linear increase in the Southeast of approximately 1 tornado per year (MSTL: 1.5 tornadoes/year, TBATS: 1 tornado/year; both p < 0.001).
• The trend component in the Southeast hinted at a ~40-year cycle, which, when explicitly included in the decomposition, smoothed the trend and increased the rate of increase (e.g., 2.1 tornadoes/year; p < 0.001).
• Accounting for longer cycles suggested that the increasing trend in the Southeast flattens after the early 2000s, implying some apparent trend in raw data might be due to cyclical variability.

Contributions

• Developed a novel domain-informed, data-driven decomposition framework that integrates climate knowledge (teleconnection periodicities) with statistical methods to precisely isolate long-term trends from variability and noise in complex time series.
• Provided robust evidence for previously observed regional tornado trends (decline in Great Plains, increase in Southeast) by demonstrating their persistence after explicitly filtering out coherent variability and noise associated with multiple teleconnections.
• Identified and incorporated specific teleconnection-related periodicities (e.g., 12, 42, 64, 154, 384, 480 months) into the decomposition, ensuring that extracted trends are physically grounded.
• Suggested the presence of a potential low-frequency cycle (~40 years) in Southeast tornado activity, offering new avenues for future research despite limitations of the observational record.
• Demonstrated the flexibility and broad applicability of the developed framework to other severe weather and climate phenomena, and potentially other knowledge domains with complex time series.

Funding

The author received no funding for this research.

Citation

@article{Moore2025Targeted,
  author = {Moore, Todd W.},
  title = {Targeted decomposition of tornado records reveals long-term trends in the Great Plains and Southeast United States},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-20868-z},
  url = {https://doi.org/10.1038/s41598-025-20868-z}
}