Madden-Julian oscillation influences United States springtime tornado and hail frequency

Severe convective storms (SCSs—tornadoes, severe hail, and damaging winds) cause substantial annual losses in the US, and improving their subseasonal prediction is of high interest. The Madden-Julian Oscillation (MJO), a 30–90 day eastward-migrating convective disturbance from the Indian Ocean into the Indo-Pacific warm pool, can force Rossby wave trains that modulate North American weather, including extremes of temperature, precipitation, atmospheric rivers, snowstorms, hurricanes, and SCSs. While correlations between MJO phase and US tornado/hail occurrence have been reported, results vary by MJO index and method, and physical pathway variability (e.g., event strength, propagation) is not well characterized. This study examines the MJO–SCS connection with attention to MJO lifecycle components, testing whether and how tornado and hail probabilities are modulated following periods of strongest MJO upper-level divergence and how different MJO "flavors" (location, strength, speed) yield distinct teleconnection pathways relevant to extended-range prediction.

Multiple indices characterize the MJO, notably the Real-time Multivariate MJO (RMM) that combines OLR and zonal winds, and the OLR-based MJO Index (OMI) focused on deep convection. Prior work found phase-dependent relationships: increased April tornado days during RMM phases 6 and 8 and May increases during phases 5 and 8; violent spring outbreaks more than twice as likely during RMM phase 2; and enhanced tornado-day frequency in RMM phase 2 in some analyses, while OMI-based results were less robust, highlighting methodological sensitivity. Strong concurrent correlations between anomalous hail frequency and RMM phase have been noted. The MJO has been leveraged for “forecasts of opportunity,” e.g., statistically significant central Great Plains tornado/hail increases 3–4 weeks after OMI phase 8 and several weeks after phases 1–2, with the late-May 2019 US tornado outbreak anticipated weeks in advance via MJO, elevated atmospheric angular momentum (AAM), jet breakdown, and favorable synoptic regimes. However, many studies emphasized concurrent relationships and did not incorporate event-specific MJO characteristics such as strength and propagation speed that may affect lagged responses.

• Study period/season: 1979–2019 boreal spring (March–June; MAMJ) to coincide with climatological peaks in tornado and hail.
• MJO characterization: Used CPC pentad MJO indices representing the negative projection of 200-hPa velocity potential (χ200) anomalies onto ten time-lagged patterns of the leading EEOF of pentad χ200. Indices are defined at longitudes 20°E–10°W and track enhanced/suppressed upper-level divergence. Active MJO events were identified when minimum pentad index ≤ −0.5, with the minimum at or west of 160°E, and the date/longitude of minimum (proxy for maximum 200-hPa divergence/convection) cataloged. Events were classified by whether they propagated across the Maritime Continent (MT/PAC) or experienced the MC barrier effect and remained in/near the Indian Ocean (IO). OMI phase-space positions were used for comparison and visualization of event evolution.
• Severe convection target metric: Practically Perfect Hindcasts (PPH) of tornado and hail (Gaussian-smoothed statistical point process based on actual reports) on the NCEP 211 grid (~80 km). Defined PPH Enhanced (ENH) risk days as tornado PPH ≥ 10% and/or hail PPH ≥ 30%. Also examined Moderate (MDT) risk days (tornado PPH ≥ 30% and/or hail PPH ≥ 60%). Computed probabilities within centered 3-day windows for the US (all grid points east of 106°W), Great Plains (GP: 106–91°W), and Eastern US (E: east of 91°W). Removed a 1979–2019 climatology of 3-day window probabilities to obtain anomalies.
• Environmental fields: From NARR, computed Significant Tornado Parameter (STP) and Supercell Composite Parameter (SCP) using 3-hourly fields summed 1200–1200 UTC to capture the diurnal SCS cycle. From ERA5, daily (1200–1200 UTC) 500-hPa geopotential height (Z500) anomalies were derived relative to 1979–2019 climatology. Earth-relative Northern Hemisphere AAM was calculated from ERA5 6-hourly zonal wind and surface pressure on native levels and averaged 1200–1200 UTC.
• Compositing/lag analysis: For each active event, composited anomalies of PPH ENH risk probabilities in centered 3-day windows for 28 days following the minimum pentad index (day 0), also examining Z500, STP+SCP sums, and AAM. Separated analyses for IO vs MT/PAC events.
• Event decomposition: Among MT/PAC events (N=53), defined three clusters: Cluster 1 (weak events, −1.0 < minimum pentad index ≤ −0.5; N=13). Clusters 2 and 3 derived via k-means (Lloyd’s algorithm) using pentad indices from day −14 to +14 around the minimum (N=18 and N=22, respectively), representing different strengths and propagation speeds.
• Significance testing: 1000-iteration bootstrap for daily PPH probability anomalies; two-tailed Student’s t-test for Z500 anomalies; Mann–Whitney U-test for STP/SCP and PPH anomalies due to non-Gaussian distributions. Significance threshold p < 0.05.

• Overall MJO influence: Across 100 active MJO events (1979–2019 MAMJ), statistically significant increases in US tornado and hail probabilities were found 3–4 weeks after the period of strongest MJO upper-level divergence for the 53 events that propagated past the Maritime Continent (MT/PAC). In contrast, 47 events that did not cross the MC (IO; barrier effect) showed few significant increases in US severe weather probabilities.
• IO (barrier-effect) events: Minimum pentad indices typically near 60°E with events concluding within ~12 days. Significant increases in Eastern US PPH ENH risk probabilities peaked at days 4–5 after the minimum; decreases in US probabilities at days 17–19. OMI phase evolved from phase 8/1 at day 0 toward phase 6 by day 28.
• MT/PAC events: Strengthened while traversing the Indo-Pacific warm pool, with OMI phase space progressing through phases 4–1. Northern Hemisphere AAM increased significantly at days 6–14 following the minimum; this coincided with a significant decrease in US PPH ENH risk probability, followed by significant increases in GP at days 19–21 and Eastern US at days 22–25. Similar modulations held for PPH Moderate risk days.
• Synoptic pathway (MT/PAC composites): A Rossby wave train linked the tropical-central Pacific to the western US by day 14, featuring a significant negative Z500 anomaly over the central Pacific and positive Z500 over the eastern US. By day 21, a migrating Z500 trough over the central US coincided with significant increases in STP/SCP and PPH ENH risk probabilities over the southern Great Plains and Southeast. By day 28, the trough shifted east with subsidence and offshore flow reducing probabilities over the eastern US.
• Event strength dependence: Stronger MJO events (pentad amplitude ≥ 1.0) produced larger increases in PPH ENH probabilities, with a composite Z500 cyclone centered over Oklahoma at day 21 and increased probabilities in the warm sector.
• Clustered MT/PAC flavors:
  • Cluster 1 (weak; N=13): No significant AAM response until day ~20; thereafter significant increases in US PPH ENH probabilities days 22–28. Pattern evolution showed a western US anticyclone at day 14 shifting northeast by day 21, yielding a favorable western-trough/eastern-ridge setup and increased GP probabilities.
  • Cluster 2 (fast-propagating; N=18): Strong negative pentad indices from 160°E to 80°W, propagation ~21° lon/day; OMI phases 4→2. Significant increases in ENH risk probabilities over the eastern US at days 23–24, with a western US anticyclone and eastern US trough around day 21 and enhanced probabilities over the Southeast east of the 500-hPa thermal trough axis.
  • Cluster 3 (active, slower; N=22): Strong convection confined 120°E–140°E, propagation ~13.5° lon/day; OMI phases 3→8. Significant AAM increases days 3–12; significant increases in ENH risk days for GP at days 16–22 and Eastern US at days 20–25. An eastern US anticyclone developed by day ~10, persisted through day 21, and supported increased SCS probabilities, especially focused on the Ark-La-Tex region by day 21.
• Case example (April–May 2019): A strong MJO event initiated in late April with peak upper-level divergence on 3 May 2019. AAM peaked near +10 on 3 May and declined over the next four weeks. The subsequent 3–4 week period yielded 14 days with PPH ENH-level risks, with significant peaks in PPH grid points and summed STP+SCP during mid-to-late May under a favorable western trough/eastern ridge regime.

The study demonstrates that the MJO’s lifecycle—and specifically the timing of maximum upper-level divergence—provides a physically grounded predictor of subsequent US tornado and hail activity on subseasonal time scales. Events that propagate across the Maritime Continent (MT/PAC) generate robust extratropical teleconnections, including Rossby wave trains and AAM fluctuations, that modulate North American synoptic patterns and severe-weather ingredients. The timing and magnitude of enhanced tornado and hail probabilities depend on MJO propagation characteristics and strength. Distinct clusters of MT/PAC events yield different lag structures and regional emphases (Great Plains versus Eastern US), identifying windows of enhanced predictability (weeks 3–4) that can be exploited as forecasts of opportunity. Conversely, IO events that experience the MC barrier effect show limited or earlier, less robust impacts on US severe weather. These findings refine the MJO–SCS linkage from simple phase-based associations to lifecycle-aware, pathway-dependent modulations, improving interpretability and potential operational utility in extended-range severe weather outlooks.

By cataloging and clustering boreal spring MJO events using pentad χ200 indices and linking them to practically perfect hindcasts of US tornado and hail, the study identifies robust, physically consistent pathways by which the MJO modulates severe weather probabilities on 2–4 week lead times. Key contributions include: (1) demonstrating significant increases in US tornado/hail probabilities 3–4 weeks after maximum MJO divergence for MT/PAC events, (2) contrasting these with limited impacts from IO barrier-effect events, and (3) revealing three distinct MT/PAC MJO flavors with unique timing and regional signatures in severe weather risk. These insights support the development of hybrid dynamical–statistical subseasonal prediction systems that weight probabilities by observed MJO characteristics. Future work should expand samples using climate-model simulations to address limited observational event counts, explore additional modes of variability interacting with the MJO, and assess operational integration to improve extended-range severe weather outlooks.

The primary limitation is sample size: only 41 boreal spring seasons (1979–2019) and 100 active events, with further subdivision by propagation pathway and cluster reducing statistical power. Results focus on MAMJ and may not generalize to other seasons. Affiliated analyses rely on reanalyses (NARR, ERA5) and PPH reconstructions, which, despite demonstrated utility, carry inherent uncertainties. The clustering approach may not capture the full diversity of MJO behavior that would emerge in much larger samples. Future studies using long climate-model simulations could provide more robust statistics and explore interactions with other climate modes.