The effect of increased horizontal resolution on the NCEP global ensemble mean forecasts
Academic Article
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
The primary goal of this paper is to explore why the use of increased horizontal resolution enhances the performance of the National Centers for Environmental Prediction (NCEP) global ensemble mean forecasts. Numerical experiments were carried out with a 10-member (five-pair) 0000 UTC subset of the NCEP global ensemble forecasts for a 30-day period during January-February 1999. Four sets of ensembles and corresponding control forecasts were generated. One ensemble was identical to the then-operational T62 horizontal resolution NCEP ensemble, while in the other three ensembles the horizontal resolution was increased to T126 out to day-1, day-3, and day-15 forecast lead times. Anomaly correlation and root-mean-square error, also decomposed into bias and variance terms, were used to evaluate the control and ensemble mean forecasts. As expected, the use of a higher-resolution model improves both scores. A newly developed condition for optimal smoothing indicates that the root-mean-square error for the high-resolution 10-member ensemble is nearly as low as it can be given its anomaly correlation. Therefore, further significant improvements in the ensemble mean forecasts can be achieved only through improved anomaly forecast patterns, and not through additional smoothing. The two main meteorological aspects of the higher-resolution-induced error reduction for both the control and the ensemble mean forecasts are 1) the maintenance of a more realistic time-mean flow, and 2) the better prediction of high-frequency transients along the midlatitude storm tracks. The effect of increased horizontal resolution, however, is markedly more positive on the ensemble mean than on the control forecasts. This is because the ensemble mean 1) efficiently filters out unpredictable small-scale features at high resolution, and 2) accentuates the relatively large systematic errors present in the low-resolution integrations.
