Describing long term seasonal and interannual variability of tornado frequency across the USA and Europe

Tornadoes are weather hazards that can kill people and severely damage infrastructure, whose small size and chaotic dynamics presently limit our capability of a deterministic reproduction of their frequency. Several studies show the role played by meteorological factors such as updraft strength and wind shear. In this study, we apply an empirical approach based on meteorological variables available from the ERA5 reanalysis dataset for the analysis of tornado occurrence over USA and Europe and compare its results to the observed tornado occurrences recorded by the European Severe Weather Database (ESWD) in Europe (https://www.essl.org/cms/) and the Storm Prediction Center (SPC) in the United States (https://www.spc.noaa.gov/wcm/#dat).  The analysis is based on   WS700 (lower troposphere wind shear) and WMAX (maximum updraft parcel vertical velocity associated with Convective Available Potential Energy., or CAPE). These variables are used to estimate the frequency of log10(p) exceedances over crucial thresholds (-3.5 and -4.5) and compare them to observed tornado counts using the methods of Ingrosso et al. (https://doi.org/10.5194/nhess-23-2443-2023). We complement this analysis with an estimate based on the simultaneous occurrence of large values of WMAX and WS700 (HTR, high-tornado-risk WMAX-WS700 conditions). In the United States, HTR WMAX-WS700 occurrences is a powerful indicator of tornado activity, particularly during spring (r = 0.98), with solid yearly correlations (r = 0.91) and significant skill even in winter. Probability thresholds (log₁₀(P) > -4.5, -3.5) reveal strong correlation in spring and annually (r ≈ 0.7-0.8), but weaker or even negative relationships in summer and autumn. Across all seasons, the occurrence-based metric HTR WMAX-WS700 regularly beats the use of probability thresholds in terms of correlation with observed tornadoes frequency, especially in the autumn, when probability-based indicators fail. In Europe, the HTR WMAX-WS700 approach performs less satisfactorily than in the USA (r = 0.50, p = 0.028 at the annual time scale), and probability thresholds perform poorly. Overall, our approach performs best when employing the HTR WMAX-WS700 approach during peak tornado seasons, particularly in the United States, whereas probability thresholds alone are insufficient for meaningful prediction outside of peak periods. These findings emphasize the importance of seasonally adaptive and occurrence-based approaches to tornado risk assessment.

This study was supported by funding from ICSC – Centro Nazionale di Ricerca in High Performance Computing, Big Data and Quantum Computing, as part of the European Union’s NextGenerationEU initiative. Project code: CN_00000033 CUP: C83C22000560007.