An Assessment of the Quality of Forecast Trajectories

Abstract

Forecast and “analysis” (reference) trajectories were computed from six sites over North America at three altitudes (500, 1000, and 1500 m above ground) twice a day for a one-year period using Nested Grid Model wind fields. The reference meteorology was a series of short-term forecasts. Absolute error (distance between reference and forecast trajectory), relative error (absolute error divided by forecast trajectory travel distance), and the angle between the reference and forecast trajectory were also computed. The mean relative error for all the forecast trajectories for a travel time of 36 h is about 35%; the 90th percentile of the relative error is about 65%. The forecast is slightly biased to the left of the reference early in the forecast period. Absolute error and travel distance both are larger in winter than summer, so that the relative error is generally constant throughout the year. Differences in mean error among the three starting altitudes, among the six origin sites, and between the two origin times are insignificant when compared to the variation in errors for a collection of trajectories at a given origin. The forecast trajectories were objectively classified through a cluster analysis, which groups trajectories by direction and travel distance. For all clusters, by season, origin site, and altitude, differences between the. minimum and maximum cluster-mean relative errors were about a factor of 2-3. Individual forecast trajectories composing clusters with the minimum relative error (about 20%) tended to originate within stronger, steady flow either ahead of or behind a cold front. Maximum relative error (about 45%) was associated with forecast trajectories originating in regions of generally slow wind fields such as under a high pressure system or near stationary or slowly moving fronts.