Verification of Some Numerical Models for Operationally Predicting Mesoscale Winds Aloft

Abstract

Four numerical models are described for predicting mesoscale winds aloft for a 6-hr period. These models are tested statistically against persistence as the control forecast and against predictions made by operational forecasters. Mesoscale winds aloft data were used to initialize the models and to verify the predictions on an hourly basis. The model yielding the smallest r.m.s. vector errors (RMSVEs) was the one based upon the most physics, which included advection, ageostrophic acceleration, vertical mixing, and friction. Horizontal advection was the most important term in reducing the RMSVEs, followed by ageostrophic acceleration, vertical advection, surface friction, and vertical mixing. From a comparison of the mean absolute errors, based upon approximately 72 independent wind-profile predictions made by operational forecasters, by the most complete model, and by persistence, it is concluded that the model is the best wind predictor in the free air. In the boundary layer, the results tend to favor the forecaster for direction predictions. The speed predictions showed no overall superiority in any of these three models.