Predictability of Stratospheric Sudden Warming Events in Grims Model

The predictability of stratospheric sudden warming (SSW) events is examined by integrating an operational NWP model, GRIMs, for four displacement and four split SSW events. The model is initialized every 6 hours starting from 25 days before each SSW onset date, and integrated up to 30 days with time-varying surface boundary conditions. SSW prediction skill is then quantified by combining mean squared skill score (MSSS), mean squared error (MSE), and anomaly correlation coefficient (ACC) of 10-hPa geopotential height fields. In consistent with previous findings, the displacement SSW events (11 days) are better predicted than the split SSW events (9 days). A lower prediction skill of the latters is partly caused by anomalously weak wavenumber-two (s=2) wave activity in the lower stratosphere. Such an underestimation is largely due to the misrepresentation of wavenumber-one (s=1)-induced polar vortex modification. This result indicates that s=1 wave activity plays a critical role in simulating not only the displacement but also the split SSW events. A series of sensitivity tests further show that SSW prediction is more sensitive to the initial condition than the boundary condition. Sensitivity to the gravity wave parameterization is rather minor. These results suggest that more realistic background flow, through comprehensive stratospheric data assimilation, may be crucial to improve SSW prediction skill in an operational NWP model.