Mesoscale NWP forecasts are made for all of Australia twice daily using the Australian Bureau of Meteorology's LAPS NWP system. The grid spacing of this model is approximately 14km. Increasingly the NWP forecasts are used to provide direct prediction of the timing of major wind changes, severe and tornadic thunderstorm environments, and similar "major impact" weather events. One of the major strengths of these mesoscale forecasts is the ability to predict unusual or extreme behaviour, and thus to provide an alert to forecasters.

The standard verification statistics long used in NWP, such as root-mean-square error, skill scores, etc, do not provide a measure of skill of the NWP systems in predicting these events, yet such information is a vital component in the forecasters' decision-making process.

One of the difficulties of verifying such forecasts is in determining whether the NWP model has forecast an event, and whether an event has occurred, and this issue will be discussed in the context of two major Australian forecasting challenges. The first is that of the timing of major wind changes during the southeastern Australia fire season, and the other in predicting cool-season tornado environments in southern Australia. In each case the verification strategy will be described and the performance of the model forecasts documented.