Forecast Skill and Low-Frequency Variability in NMC DERF90 Experiments

Abstract

A series of 90-day integrations by a low-resolution version (T40) of the National Meteorological Center's global spectral model was analyzed for its performance as well as its low-frequency variability behavior. In particular, 5-day mean 500-mb forecasts with leads up to 88 days were examined and compared with the observations. The forecast mean height decreased rapidly as forecast lead increased. A severe negative bias of the mean height in the Tropics was caused by a negative temperature bias and a drop of the surface pressure of about 2 mb. The forecast variance also dropped rapidly to a minimum of 75% of the atmospheric standard deviation before being stabilized at day 18. The model could not maintain large anomalous flows from the atmospheric initial conditions. However, it is quite capable of generating and maintaining large anomalies after drifting to its own climatology and temporal variability. At extended ranges, the model showed better skill over the North Pacific than North Atlantic when the season advanced to the colder period of the DERF90 (dynamical extended-range forecasts 1990) experiments. The model also displayed dependence on circulation regimes, although the skill fluctuated widely from day to day in general. Blocking flows in the forecast were found to systematically retrogress to the Baffin Island area from the North Atlantic. Therefore, improvements of the model's systematic errors, including its drift, appear to be essential in order to achieve a higher level of forecast performance. However, no generalization can be made due to the usage of a low-resolution model and the experiments being carried out over a rather short time span, from only 3 May to 6 December 1990.