Abstract
The effect of changing the horizontal interpolation for evaluating the quantities at departure points in the global spectral semi‐Lagrangian variable resolution NWP model ARPEGE/IFS is described. The two interpolations involved are conventional cubic Lagrangian and Hermite with estimates of derivatives by fifth‐order compact upwind differencing (CUD‐5). Their properties are compared with examples of different one‐dimensional and two‐dimensional linear‐advection tests. With each kind of horizontal interpolation, twelve 5‐day forecasts are performed, beginning on the fifth day of each month from February 1994 to January 1995, using the same vertical non‐interpolating scheme.
On average, the difference between forecasts with these interpolations in geopotential and mean sea‐level pressure fields is negligible within the first three days, but noticeable on the fifth day, especially in the mean sea‐level pressure field. However, these improvements are present beyond the range for which the variable resolution strategy is designed.
The root‐mean‐square errors for the relative‐humidity fields at 1000 hPa and 850 hPa demonstrate some improvement as a result of the change of interpolation, from the second day of the forecasts onwards. Hermite CUD‐5 interpolation also results in a reduction of the amount of negative moisture.