A flash flood occurred on 23 November 1996 in Coffs Harbour on the east coast of Australia. The flood resulted from record rainfall totals of just under 400 mm in 4.5 hours over the Coffs Harbour Creek watershed, which is only about 50 kilometers square. There was one fatality, mobile homes destroyed, and shops and houses flooded. There was an estimated damage bill of AU$30 million. Predicted rainfall totals from a high resolution mesoscale model were comparable with the observed totals.

In attempting to assess the uncertainty involved with single model forecasts, several weather centres including the European Centre for Medium Range Weather Forecasting (EC), the US National Oceanographic and Atmospheric Administration (NOAA) and other institutions, such as Florida State University (FSU), routinely produce ensemble forecasts of meteorological variables including precipitation several days in advance. However, in attempting to produce a possible forecast spread of rainfall for a single event such as the Coffs Harbour flash flood, a scenario approach is used here. This singular procedure is appropriate because the important mechanisms that resulted in the heavy rainfall causing the flash flood have been investigated in detail (Speer et al. 2000). Therefore, the methodology in this study is to generate a range of key initial variables which in turn are used to produce a corresponding spread of possible rainfall forecasts. Twenty-four hour model rainfall forecasts were generated using these perturbed initial conditions which showed that (under current climate conditions) the Coffs Harbour flash flood rainfall could be up to 50% higher than the largest observed totals of 23 November 1996.