Predictability of Sudden Stratospheric Warming Events

Abstract

The predictability of sudden stratospheric warming (SSW) events, such as those that occurred on January 5, 2004 and January 7, 2013, are examined by integrating an operational weather prediction model. The model is initialized from 25 days before the onset of a SSW event to 20 days after, and integrated for 30 days. The prediction skill is then evaluated by computing both mean squared error and anomaly correlation coefficient with respect to the reanalysis data. The 2004 SSW event, which was characterized as a vortex displacement event, was preconditioned by the vertical propagation of a wavenumber-one wave in the lower stratosphere, while the 2013 SSW event, a vortex split event, was driven by the amplification of wavenumber-one followed by wavenumber-two waves. These two different events were reliably predicted with a maximum lead-time of nine days and seven days, respectively. However, their nature is quite different. For the 2004 displacement SSW event, zonal-mean errors accounted for a great part of the forecast errors, whereas the forecast errors of the 2013 split SSW event were highly influenced by the zonally asymmetric components. Comparatively a large magnitude of zonally asymmetric errors in the split SSW were mainly caused by an underestimated amplification of zonal-wavenumber two waves in the lower stratosphere. This result indicates that the predictability of SSW events may be dependent on the morphology of the polar vortex. To confirm this finding, the five most recent displacement and split SSW events were further investigated in this paper. On average, the model shows a better prediction skill for the displacement SSW events than the split SSW events with about one-day longer lead-time. Moreover, all displacement events show a smaller ratio of zonally asymmetric errors to zonal-mean errors than the split events. In addition, underestimated-vertical propagation of wavenumber two waves explains a large zonally asymmetric errors in all split cases.