Statistical short-term forecasting technique for wind and temperature in a coastal tropical location (Townsville)

An objective statistical technique for short-term wind and temperature prediction is developed using linear regression techniques. Potential predictors with immediate relevance to the sea-breeze circulation were emphasised as an integral part of the meteorology of the predictand. Potential predictors were empirically combined to partially account for non-linear relationships. A 31-day operational trial compares the statistical model to manual forecasts, climatology and persistence. It is established that the temperature and wind direction predictions have similar skill to those produced manually by forecasters. The statistical model displayed superior skill over other methods for wind speed prediction. Specifically, for a compilation of three to nine hours, statistical model forecasts for observed winds of greater than 10 kn (5.2 m/s), root mean square errors for wind direction and speed of 20° and 2.2 kn were achieved respectively.