Automated Prediction of Surface Wind from Numerical Model Output

Abstract

The Model Output Statistics (MOS) technique has been applied to the prediction of surface winds. Warm and cool season forecasting equations were developed by screening as potential predictors several forecast fields from the National Meteorological Center's Primitive Equation (PE) model. Four additional weather parameters from surface reports were also screened to provide the latest observed conditions for the initial forecast projection. Separate equations for the U and V wind components and wind speed, S, were derived for each of 233 stations for projections of 12 to 48 b. For any given station and projection, the U,V, and S equations were required to use the same predictors. Initially, the first three predictors were forced to be boundary layer U,V, and S forecasts from the PE model. Comparative verification was carried out on independent data for test forecasts at 20 widely distributed stations during warm and cool seasons. In addition, operational guidance forecasts were verified for 92 stations during November 1973 to March 1974. Both of these tests suggested that although the automated method tended to underforecast strong winds, its objective forecasts were generally more accurate than the subjective National Weather Service forecasts. Adjustment and transformation procedures to enable the automated system to produce more strong wind forecasts were also tested.