SYSTEMATIC ERRORS IN OPERATIONAL BAROCLINIC PROGNOSES AT THE NATIONAL METEOROLOGICAL CENTER

Abstract

This paper reviews the performance of the short-range, operational national weather prediction over the Northern Hemisphere at the National Meteorological Center. Mean monthly 500-mb error charts, valid at 1200 gmt only, for four midseason months of 1964 through 1968 are presented from 36-hr prognoses made using the National Meteorological Center's operational three-level and primitive equation models. Systematic errors found to persist in both models show that, in the mean, the amplitude of troughs and ridges is underforecast. This tendency shows up in mean monthly and in daily error charts as positive errors in troughs and negative errors in ridges. Very large-scale errors (i.e., in waves 1 through 5 and so on when used in this context) occur in forecasts from both models. The causes of these very large-scale errors remains obscure. However, the causes of certain smaller scale systematic errors have been diagnosed. In particular, the large 500-mb negative error over western Canada, ahead of the Gulf of Alaska trough, has been diagnosed as caused by deficiencies in amplitude of the terrain in the models over western Canada. Corrective adjustments in the model mountains made in September 1968 have decreased errors over western Canada. Comparison of errors in the three-level and later primitive equation 500-mb prognoses indicates that an improvement in skill has occurred over North America. Preliminary study of the vertical distribution of errors from 1000 mb to 200 mb in September and October 1968 indicates that the primitive equation model 1) underforecasts the strength of the mean thermal wind particularly above 500 mb, and 2) contains a negative bias in the depth of 1000 mb (surface) Lows east of the American Rockies.