Limited surface impacts of the January 2021 sudden stratospheric warming

The study addresses how a sudden stratospheric warming (SSW) on January 5, 2021 influenced surface weather on subseasonal timescales. Subseasonal forecasting (10–30 day lead) is an emerging frontier with recognized windows of opportunity, yet attributing predictability to initial states of the atmosphere, ocean, sea ice, and land remains challenging. SSWs, occurring roughly every two winters in the Northern Hemisphere, are dynamically driven stratospheric events that can be predicted and are often associated with a surface pattern resembling the negative phase of the Northern Annular Mode (NAM). Despite frequent attribution of post-SSW surface anomalies to stratospheric influences, the mechanisms of stratosphere–troposphere coupling and their impacts on surface predictability are not well understood. This paper aims to unambiguously quantify the surface impacts of the January 2021 SSW and assess the roles of tropospheric circulation, stratospheric variability, and surface boundary forcings using controlled ensemble forecasting experiments.

Prior work shows that SSWs can be driven by enhanced upward-propagating planetary waves and other stratospheric processes, and often precede surface patterns of warmth over northern North America and the Middle East and cold over Eurasia, consistent with a descending negative NAM. Studies indicate enhanced forecast skill following SSWs in certain regions and time windows, but the extent and mechanisms of downward influence remain debated. Some analyses of the January 2021 event argued for strong stratospheric projection onto surface weather and highlighted polar vortex stretching and planetary wave reflection as key precursors to cold outbreaks. However, distinguishing causality versus correlation is difficult, and experiments that constrain the stratosphere may inadvertently embed two-way feedbacks, complicating attribution. The present study builds on this literature by experimentally separating the roles of tropospheric and stratospheric initial states and their coupling.

Model and forecast framework: The CESM2(WACCM6) subseasonal Earth system prediction framework (atmosphere extending to ~140 km; coupled ocean, sea ice, and land) produces weekly 21-member ensemble forecasts. Initial conditions are generated by nudging winds and temperatures toward NASA FP-IT reanalysis in the week prior to initialization to spin up atmospheric chemistry and Earth system components. Experimental design (Jan 4, 2021 initializations): Four 21-member ensembles were run: (1) standard forecast with accurate initial conditions; (2) scrambled tropospheric initial conditions with accurate stratospheric initial conditions; (3) scrambled stratospheric initial conditions with accurate tropospheric initial conditions (producing no SSW); and (4) scrambled atmospheric initial conditions using an atmospheric state from January 2, 2017 (a year with no major SSW). Scrambling is achieved during spin-up via tapered nudging to create divergence from observations in the targeted atmospheric layer while maintaining one-way coupling from the correctly initialized portion. Additional forecasts around the February cold outbreak: Ensembles initialized on February 1, 2021 (standard; scrambled stratosphere with accurate troposphere) and February 8, 2021 (standard; scrambled troposphere with accurate stratosphere; scrambled stratosphere with accurate troposphere) were conducted to evaluate predictability of the North American cold event. Verification and diagnostics: Verification used MERRA2 reanalysis (0.5°×0.66° grid). Surface (2-m) temperature and geopotential height anomalies were computed relative to 1999–2019 daily climatologies (lead-dependent for forecasts), with 120-day triangular smoothing applied to climatologies. Geopotential height anomalies were standardized by daily means and standard deviations (lead-dependent). The squared anomaly correlation coefficient (latitude-weighted) communicated shared spatial variance and correlation sign. Polar cap averages used 60–90°N; midlatitude averages used 45–75°N; North America land average used 25–70°N, 70–135°W over land only. Tropopause was diagnosed from lapse rate (<2 K/km criterion). Wave activity flux (WAF) diagnostics followed quasi-geostrophic formulations for stationary waves with appropriate vector scaling; both zonal-mean and meridional-mean perspectives and streamline visualizations were used to infer propagation and reflection. Model components: atmosphere (finite-volume 1.25°×0.9°), ocean POP2, sea ice CICE5.1.2, land CLM5. Ensemble generation used random field perturbations. Nudging configurations for scrambling: for scrambled troposphere, nudging tapered from 1 h at 20 km to no nudging at 15 km (branch from Dec 21, 2020); for scrambled stratosphere, nudging tapered from 1 h at 9 km to no nudging at 12 km (branch from Nov 16, 2020).

• Weeks 1–2 after the January 5, 2021 SSW: Observed surface temperatures (MERRA2) showed the characteristic NAM-like pattern. The standard forecast explained 72% of the spatial variance but with some biases. Scrambling the troposphere destroyed skill and anticorrelated with observations and the standard forecast, whereas scrambling the stratosphere (no SSW) yielded a forecast indistinguishable from the standard. Conclusion: surface temperatures in weeks 1–2 were driven entirely by the tropospheric circulation, independent of the SSW.
• Weeks 3–4: The NAM-like pattern persisted with shifts (cold into Siberia, warmth over NE Canada). The standard forecast explained 30% of spatial variance. The scrambled stratosphere forecast (no SSW) best matched observations and captured most of the standard forecast’s variance, whereas scrambled troposphere showed partial resemblance but misplaced North American anomalies. Ocean/sea-ice/land showed little direct contribution in weeks 3–4. Overall, the SSW explained none of the spatial variance in weeks 1–2 and at most about one-third in weeks 3–4.
• Stratosphere–troposphere coupling: Positive polar cap geopotential height anomalies descended after the SSW, but experiments indicate the troposphere helped drive lower-stratospheric anomalies and their persistence via feedbacks. With scrambled troposphere, stratospheric anomalies remained aloft; with scrambled stratosphere, the troposphere forced lower-stratospheric anomalies. Feedbacks likely sustained anomalies in the standard forecast. Overall, insufficient evidence that the SSW or stratosphere substantially impacted surface circulation in the month after the event.
• February 12–18, 2021 North American cold outbreak: The observed North America land-average anomaly was −5.5 °C, the coldest since 1980. Forecasts initialized January 4 (standard and scrambled stratosphere) did not predict extreme cold and were not significantly different from the climatology. Forecasts initialized February 1 and 8 (standard and scrambled stratosphere) predicted mean anomalies ~−3 to −3.5 °C, comparable to the second-coldest event; at least one member in each ensemble was colder than −5 °C; all Feb 8 members predicted anomalous cold. The Feb 8 scrambled troposphere forecast predicted +3 °C (second-warmest-like), highlighting the primacy of tropospheric initial conditions for this event.
• Vortex stretching and wave reflection: Polar vortex stretching signatures and wave reflection appeared in observations and were partially reproduced in forecasts, but were not necessary for the cold outbreak. WAF analyses showed that downward-reflected stratospheric wave activity did not penetrate the tropopause to feed the North American trough; relevant WAF convergence in the upper troposphere/lower stratosphere originated from below. The cold outbreak would have occurred with or without wave reflection.
• Additional notes: In scrambled stratosphere ensembles, no SSW occurred (0/21 members), confirming the experimental separation of SSW influence from tropospheric drivers.

The experiments demonstrate that, for the January 2021 case, post-SSW surface temperature variability was dominated by the existing tropospheric circulation and its coupling with the lower stratosphere, rather than by the SSW itself. The common interpretation of a downward-propagating negative NAM is not supported as the primary causal mechanism; instead, the troposphere induced and helped sustain lower-stratospheric anomalies after the SSW, consistent with a feedback framework. The February 2021 North American cold event was not a deterministic outcome of the SSW, nor did vortex stretching and wave reflection play a proximal role; correct tropospheric initialization within one week was key to capturing the event. These findings refine attribution narratives and suggest that stratospheric signals may often modulate persistence via feedbacks rather than directly force surface extremes. The results contrast with prior analyses that inferred strong stratospheric control for this event, underscoring the value of experimental designs that separate tropospheric and stratospheric influences.

Using initial condition scrambling within a state-of-the-art Earth system prediction framework, the study shows that the January 5, 2021 SSW had limited direct impact on surface temperatures over the subsequent month, and did not cause the record North American cold of mid-February 2021. Tropospheric circulation and two-way coupling with the lower stratosphere were the dominant contributors, with stratospheric anomalies acting as feedbacks that may enhance persistence rather than as primary drivers. The work introduces and validates initial-condition scrambling as a practical tool for event-scale attribution, complementing statistical methods and reducing causal over-interpretation. Future research should apply this framework to larger samples of SSWs and other extremes, quantify conditions under which vortex stretching crosses thresholds relevant for cold outbreaks, examine nonlinearity in wave–mean flow interactions, and assess model dependence across forecasting systems.

• Case-study focus: Results are based on a single SSW and one major cold outbreak; generalizability requires analysis of more events.
• Model dependence: Conclusions rely on CESM2(WACCM6) physics and coupling; other models may exhibit different sensitivities or coupling strengths.
• Lead-time challenges: Forecast skill degrades at 4–5 week leads; lack of deterministic prediction for the February event is partly due to intrinsic predictability limits.
• Diagnostic limitations: Wave activity flux interpretations can be misleading under nonlinearity; zonal/meridional averaging may obscure regional features.
• Experimental design: Scrambling via tapered nudging introduces one-way coupling during spin-up and is not purely random; while intentional, this may influence the exact manifestation of coupling and persistence.