Simulated Forecast Error and Climate Drift Resulting from the Omission of the Upper Stratosphere in Numerical Models

Abstract

The error growth associated with the usual upper boundary formulation (and location) in numerical weather prediction (NWP) and general circulation models (GCMs) is studied. The experimental strategy is to look at the growth of differences between the equilibrated climate simulation of a control simulation and three ensembles of 30 day simulations branching off of the control. The control simulation is a seasonal integration of a medium horizontal-resolution GCM with 30 levels extending from the surface to the upper mesosphere. Each ensemble consists of nine cases with initial conditions taken at 10-day intervals from the control. The main experiment uses a model identical to that of the control except that only the bottom 15 levels (below 10 mb) are retained. The additional experiments either perturb the initial conditions or alter the physical parameterizations (horizontal diffusion) to obtain information on the significance of the results of the main experiment. It is found that random error growth rate in the troposphere for the case of the altered upper boundary is slightly faster than that for initial-condition uncertainty alone. However, this is not likely to make a significant impact in operational forecast models at present because the uncertainty in the initial conditions is so large. Systematic errors in the troposphere due to the upper boundary are relatively small prior to day 20. However, the ten day mean errors from days 20–30 are about the same magnitude as the anomalies that extended range forecasts are attempting to predict. The upper boundary treatment is likely to cause significant systematic errors in such forecasts. In the lower stratosphere, the errors are substantial within the first few days, particularly in the winter hemisphere.