Error Climatology of the 80-Wave Medium-Range Forecast Model

Abstract

A climatology of the once-daily (0000 UTC) 1000-hPa error fields of the National Meteorological Center's 80-wave Medium-Range Forecast (MRF) model is studied. An analysis of the error field has been conducted over the contiguous United States and over the Northern Hemisphere from 20° to 80°N for three warm and four cool seasons (9 September 1987 to 6 March 1991). Temporal and spatial mean error fields over various integration lengths are presented. The skill, as measured by the anomaly correlation, has not significantly changed over the lifetime of the 80-wave MRF model. Anomaly correlation values at 1000 hPa and 500 hPa show that the model is retaining useful information about the anomalies in the height field out to about one week. A reduction in the model biases may reflect an improvement in model physics (longwave radiational calculations, etc). The cool and warm seasons have distinctly different spatial error patterns. The 1000-hPa warm season shows spurious height falls over the southwestern United States that grow with increasing integration length. The 1000-hPa cool season underestimates the intensity of low pressure systems over and east of Hudson Bay and overestimates their strength over the Pacific Northwest. Principal components analysis of the 429-variable error covariance matrices for the cool and warm seasons identifies 6 orthogonal variables that explain over 60% of the original error variance. MRF model problems appear to be related to problems the model has with simulating the atmosphere's interaction with orographic features (Alberta and Colorado Rockies), storm tracks and baroclinic zones (Gulf Stream region and United States-Canadian border), and persistent atmospheric features (Hudson Bay low, eastern Pacific subtropical high, and desert Southwest heat low).