A major forecasting problem at weather centers worldwide is in the provision of short-range numerical weather prediction (NWP) guidance for convection, especially if the convection is expected to reach the severe stage. The Sydney basin and surrounding area is a natural laboratory for thunderstorms, particularly during the warm season (September to April), when moisture availability from the adjacent Pacific Ocean is high. The chance of large hail or damaging winds during this period is greatly increased. Notable recent severe storms include the 14 April 1999 hailstorm and the 18 March 1990 hailstorm but there are many other days when conditions appear favorable for severe convection. Air-mass boundaries, such as frontal or sea-breeze convergence lines, often occur to the west and southwest of the Sydney basin but convective initiation does not always eventuate. We therefore believe that attempting to account for initial condition uncertainty will become increasingly important in developing improved short-range NWP guidance for convection (or no convection). To this end, the authors have been undertaking a program of short-range ensemble forecasting (SREF) for several years in relation to severe and non-severe weather events.

In this study we look at one day in particular, December 8, 2000, when severe convection was expected in the Sydney area by forecasters but no convection occurred. The expectation was that there would be sufficient buoyancy for afternoon convection to occur and to reach the severe stage, similar to the day before. However, extensive cumulus development in the morning was eroded by strong vertical wind shear. Afternoon convection failed to materialize, despite a pre-existing boundary between moist northeast winds and drier northwest winds on the ranges just to the southwest of the Sydney area. The environment became virtually cloud-free in the afternoon, assisted by decreasing moisture in the low to middle tropospheric levels.

A possible range of initial NWP model states was generated to account for initial condition uncertainty present with this type of forecast situation. Our methodology involved producing initial states, based on perturbing the observations of horizontal wind speed and moisture in the low to mid-tropospheric levels, and hence a range of forecasts for December 8, 2000.