Abstract
A numerical forecast method is proposed in which cubic polynomial splines are used to fit the spatial variations of the dependent variable fields on a variable area grid. Thirty-six hour numerical forecasts with real weather data showed that the forecast phase lag and root-mean-square error are reduced by using splines compared to finite differences. Computation time is decreased by a factor of ½ due to the variable nature of the grid, and the forecast accuracy in the fine-grid region is only marginally affected by the surrounding coarse grid.
