On the determination of the predictability of a four-level geostrophic model using the effects of grid point numerical errors

Abstract

Numerical models of the atmospheric general circulation can be improved by increasing the resolution of the horizontal and vertical finite difference grid. This sensitivity is not confined solely to the resolution of the grid; the model can be shown to be highly sensitive to the precision of the arithmetic used in the computations. This is demonstrated here by writing the FORTRAN forecast code in double and single precision, with all real-valued constants used in the program suitably defined. Searle & Davies (1975b) described the divergenrs of the eddy kinetic energy, K’, computed from double and single precision forecasts respectively, after 1500 timesteps of ½ hour duration. In the present work, these experiments are extended and an attempt will be made to measure the difference between the two forecasts by the definition of a suitable statistic to quantify the analysis.