Statistical forecasting techniques to describe the surface winds in Sydney Harbour

Sydney Harbour was selected as the venue for the sailing events for the 2000 Olympic and Paralympic games. A suite of statistically based prediction methods were derived specifical-ly for these events and provided forecasts for surface wind speed, wind direction, wind gust, and sea-breeze occurrence. The statistical models derived relationships from historical data using linear, logistic and non-linear regression, and also classification and regression trees. A feature of the statistical prediction models was the use of predictors that were selected for their meteorological relevance to the low-level wind struc-ture. Emphasis was placed upon predictors that were influen-tial in defining the sea-breeze circulation in previous studies. During the 1999 Olympic Test Events (from 16-27 September 1999), in a time-lagged trial using observed gra-dient wind data, a regression based statistical tool was veri-fied with root mean square error values of 40 degrees for direction and 2.9 kn for speed (1 kn = 0.51 m s-1). During a period that encompassed the Olympic and Paralympic Games, under real time operational conditions, a synoptical-ly stratified Model Output Statistics (MOS) technique dis-played superior skill over conventional MOS methods.