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

This study applies a domain informed, data-driven decomposition framework to target specific seasonal cycles in tornado time series and isolate long-term trends in the Great Plains and Southeast United States. To target specific seasonal cycles, dominant periodicities in key teleconnections (ENSO, NAO, PNA, AO, PDO, and AMO) were identified using Fourier analysis, followed by cross-wavelet coherence analysis to ensure these teleconnections were coherent with tornado activity before removal from regional tornado counts using multi-seasonal decomposition methods (MSTL and TBATS). Results are consistent with prior findings-showing a decline in Great Plains tornadoes and an increase in the Southeast-but the persistence of these trends after filtering out coherent variability and noise provides robust evidence that the observed trends likely reflect longer-term changes. Additionally, a potential low-frequency cycle in tornado activity is suggested, though its full extent remains unresolved due to the limited length of the observational record. The domain informed, targeted framework developed here offers a novel approach for isolating trends from variability and noise and may be applied to other climate phenomena.