PREDICTIVE SKILL OF AN NWP SYSTEM IN THE SOUTHERN LOWER STRATOSPHERE

The predictive skill of the Australian Bureau of Meteorology's Global Assimilation and Prediction (GASP) system in the southern lower strato sphere is examined using two different sets of diagnostics: (i) conven tional verification statistics used in numerical weather-prediction st udies (namely, root-mean-square (RMS) error, anomaly correlation, and bias), and (ii) elliptical diagnostics of the polar vortex (defined us ing potential vorticity on isentropic surfaces). Both sets of diagnost ics indicate the same variation in predictive skill for forecasts duri ng October 1994. The stratospheric forecasts are a large improvement o ver persistence even at seven days, with the performance at seven days being comparable to that in the troposphere of three-day forecasts. T here is large daily variability in the forecast scores for seven-day f orecasts, and the days with below-average scores occur when the flow ( vortex) is rapidly changing. Examination of the differences in the ell iptical diagnostics show that the forecast vortex is weaker, less dist urbed (i.e. closer to the pole and less elongated), and rotates faster than the analysed vortex. Consistent with a weaker forecast vortex, t he minimum polar temperature and maximum zonal wind are underpredicted in the forecasts. The verification statistics in the stratosphere hav e a large seasonal variation, although the variation is different for different statistics. The GASP RMS errors are largest (smallest) in la te-spring (summer) whereas both the ratio of GASP to persistence RMS e rror and the anomaly correlation indicate that the performance relativ e to persistence is best (worst) in late-spring (summer).