Direct verification of forecasts from a very high resolution numerical weather prediction (NWP) model

Summary The University of New South Wales (UNSW) High Resolution numerical weather prediction model (HIRES) is run routinely, on a daily basis, at a horizontal resolution of 25 km. The output is made available to the New South Wales (NSW) regional office of the Australian Bureau of Meteorology in Sydney. HIRES has been used to forecast mean wind direction and speed at a height of 12 metres for a number of events in the past. The opportunity was used in December 1997 to run the model for the annual Sydney to Hobart yacht race area. For the 1997 Sydney to Hobart yacht race the model was run at 25 km horizontal resolution and the output was made available to all competitors on the morning of the race, namely December 26th. It was also decided by the authors to expand the verification to include all observations available both from land and sea within the model domain an well as those available from a moving single point at sea, namely a yacht. After the event, the model was run once at the increased resolution of 10 km, out to 5 days ahead. Both model runs were subjected to detailed verification by one of the authors (KLB) who participated in the race aboard the maxi-yacht Nicorette and who carried out a pre-arranged observational program during the race. Surface synoptic weather maps prepared in the NSW office of the Bureau of Meteorology were also consulted in order to extend the verification scheme. The model predicted winds were verified on a six-hourly basis utilising instrumentation on the yacht as well as surface observations plotted in standard World Meterological Organisation (WMO) format on surface synoptic weather maps. The yacht carried wind sensors situated on top of the mast at a height of 30 metres above the water. The authors were most interested in the accuracy of the wind velocity forecast by the model. It is important to note that forecasting for points over the ocean at widely separated time intervals represents a very difficult challenge. The verifications reveal that the model gave overall forecast guidance of very good to excellent quality and was particularly accurate early in the race, when a Southerly Buster event occurred during the evening of the first day. However, a caveat should be added that this forecast, despite its remarkable accuracy, is not a claim that accurate very high resolution regional prediction is a solved problem. Rather, it is a demonstration that in particular instances current models are now capable of achieving high levels of skill a number of days ahead. Received September 28, 1999 Revised November 25, 1999