A STUDY OF NUMERICAL FORECASTING ERRORS

Abstract

Studies of numerical forecasting errors have revealed that the principal difficulties encountered in models currently operational at the Joint Numerical Weather Prediction Unit arise from errors in the 500-mb. forecasts. These errors are common to both barotropic and baroclinic models. The errors in the 1000?500-mb. thickness forecasts are considerably smaller. Results show that systematic errors are introduced (1) by the use of the geostrophic approximation and (2) by the approximations used on the boundaries. The cases presented in this paper show that the first type of error is virtually nonexistent in non-geostrophic, barotropic forecasts. Boundary errors are particularly serious in cases where the boundaries are meteorologically active, suggesting the future use of hemispheric forecast grids. The elimination of these two types of errors from all forecasts appears to be essential before the smaller errors, such as truncation errors and errors due to baroclinic development, non-adiabatic effects, etc., can be isolated and studied satisfactorily.