Study probes impact of stratosphere on forecasts

Nick Byrne, Inna Polichtchouk (both ECMWF), Ted Shepherd (University of Reading)


To what extent does correctly predicting conditions in the stratosphere enhance sub-seasonal to seasonal forecast skill in the troposphere? And do seasonal forecasts underestimate the predictability of the real world? A new study explores these issues by analysing the performance of ECMWF seasonal forecasts in the southern hemisphere (SH) extratropics. It finds that, during certain ‘windows of opportunity’, the stratosphere can act as an important source of predictability for the troposphere. Moreover, the study finds no evidence for the underestimation of tropospheric predictability in seasonal forecasts.

The role of the stratosphere

The variability of the extra-tropical stratosphere is strongly constrained by the seasonal cycle, with the winter and spring seasons exhibiting the most activity by far. During these seasons, there is generally a band of westerly winds present over the poles, commonly referred to as the stratospheric polar vortex. In the SH, the stratospheric polar vortex gradually descends in altitude as the calendar year progresses, before eventually dissipating with the onset of summer. This dissipation is indicated by a reversal of winds from westerly to easterly and is referred to as the stratospheric polar vortex breakdown event. The descent and the breakdown of the stratospheric polar vortex are of interest to the sub-seasonal to seasonal forecasting community as both events are thought to influence shifts of the tropospheric jet stream to the north or to the south. In particular, in years with an anomalously strong vortex and/or anomalously late polar vortex breakdown, the seasonal equatorward transition of the tropospheric jet stream is delayed, with the opposite behaviour in years with an anomalously weak vortex. Thus, skilful forecasts of these stratospheric events can also make forecasts of the troposphere more skilful.

This hypothesis was explored with the ECMWF System 4 seasonal forecast ensemble, the operational seasonal forecast system at ECMWF when this study was initiated. The ERA-Interim reanalysis was used for validation. Two forecast start dates were considered (1 August and 1 November), which were chosen to roughly coincide with the descent and breakdown events in the SH stratosphere. Forecast skill of the monthly- and weekly-mean tropospheric jet stream (averaged in the east–west direction) was then assessed for these start dates. A subset of the results is shown in the figure. Perhaps most striking is the re-emergence of monthly-mean forecast skill in the troposphere for forecasts initialised on 1 August. This re-emergence of skill is consistent with an influence from the polar vortex descent on the latitude of the tropospheric jet stream. The results suggest that approximately 20–30% of monthly tropospheric jet stream variability at lead times of 3–4 months is predictable based on knowledge of the stratospheric state. Similarly, the 1 November forecasts show results consistent with an influence from the polar vortex breakdown event on the tropospheric jet stream. They suggest that approximately 20–30% of weekly tropospheric jet-stream variability at week 3 and week 4 is predictable based on knowledge of the stratospheric state.

Forecast skill by lead time and pressure level. The left‑hand panel shows the correlation between 31‑day mean ensemble mean winds in the east–west direction from 55°S to 65°S and corresponding values in ERA‐Interim as a function of calendar day and pressure level for forecasts starting on 1 August. Values on the x-axis represent the central date of the 31‑day mean. The right-hand panel shows the same but for 7‑day means for forecasts starting on 1 November. In both panels, shaded contour regions are statistically significant at the 95% level.

The study also found that our knowledge of the stratosphere plays a role in correctly predicting the impact of the El-Niño–Southern Oscillation, marked by anomalously high or low sea-surface temperatures in the equatorial Pacific, on the tropospheric jet stream in the SH.

Do forecasts underestimate tropospheric predictability?

The large ensemble size in ECMWF’s System 4 (50+1 members) provides an opportunity to determine whether seasonal forecasts underestimate tropospheric predictability, i.e. to determine whether the seasonal forecasts are better at forecasting variability in the reanalysis than their relatively weak signal and high levels of noise (large spread) would suggest. This issue – referred to as the ‘signal-to-noise’ paradox – has received much attention in the seasonal forecast community in recent years. The term derives from the apparent mismatch between anomaly correlation values with the reanalysis and the ensemble’s own signal-to-noise ratio. This issue was investigated extensively for the SH extratropics in ECMWF System 4, but no statistically significant mismatch between anomaly correlation and the signal‑to‑noise ratio was found. Thus, there is no evidence to suggest that System 4 underestimates the predictability of SH tropospheric jet stream shifts.

Further information can be found in an article by the authors in JGR Atmospheres, doi:10.1029/2018JD030173.