Selected monthly 500- and 700-mb mean-error patterns from the National Meteorological Center's extended numerical predictions to 144 hr are shown and discussed. The NMC model (barotropic after 48 hr) consistently underforecasts the amplitude of most troughs and ridges. The pattern of error is usually established by 48 hr with the magnitude increasing with time. Also presented are charts (for 1968) of seasonal mean 700-mb error for the 5-day averaged numerical forecasts centered 4 days in advance (D + 4). The large-scale mean errors vary with season and are also highly correlated with the observed height anomaly.

Average error patterns for special extended runs of the NMC six-level primitive equation (PE) model are compared with those produced by the standard operational model. Except for a negative height bias in the PE model at low latitudes, especially in the warm season, there is very little difference at 500 mb.

The hemispheric 5-day mean upper level prognostic charts prepared from the NMC's numerical output have shown gradual improvement during recent years, as measured by RE skill, a statistic based on “reduction of error.” Comparing D + 4 mean forecasts made by direct averaging of the NMC numerical output with those generated by the Extended Forecast Division's (EFD) flow model shows that the former have maintained a consistent and gradually increasing advantage over the latter, except for the southeastern United States.

Removal of the more systematic large-scale numerical prediction errors, either by use of a running average error or, better yet, by circulation typing, can lead to improved circulation forecasts.

Journal:: Monthly Weather Review

Volume/Issue:: Volume 98: Issue 5

DOI:: https://doi.org/10.1175/1520-0493(1970)098<0385:EOMDAD>2.3.CO;2

Published Online:: May 1970

Abstract

Selected monthly 500- and 700-mb mean-error patterns from the National Meteorological Center's extended numerical predictions to 144 hr are shown and discussed. The NMC model (barotropic after 48 hr) consistently underforecasts the amplitude of most troughs and ridges. The pattern of error is usually established by 48 hr with the magnitude increasing with time. Also presented are charts (for 1968) of seasonal mean 700-mb error for the 5-day averaged numerical forecasts centered 4 days in advance (D + 4). The large-scale mean errors vary with season and are also highly correlated with the observed height anomaly.

Average error patterns for special extended runs of the NMC six-level primitive equation (PE) model are compared with those produced by the standard operational model. Except for a negative height bias in the PE model at low latitudes, especially in the warm season, there is very little difference at 500 mb.

The hemispheric 5-day mean upper level prognostic charts prepared from the NMC's numerical output have shown gradual improvement during recent years, as measured by RE skill, a statistic based on “reduction of error.” Comparing D + 4 mean forecasts made by direct averaging of the NMC numerical output with those generated by the Extended Forecast Division's (EFD) flow model shows that the former have maintained a consistent and gradually increasing advantage over the latter, except for the southeastern United States.

Removal of the more systematic large-scale numerical prediction errors, either by use of a running average error or, better yet, by circulation typing, can lead to improved circulation forecasts.