A series of 10-day forecast experiments has been carried out to invest
igate the sensitivity of a global semi-Lagrangian model to the value o
f the uncentering parameter epsilon, the magnitude of the time step fo
r the dynamics, and the numerical treatment of the physical parameteri
zations (semi-Lagrangian versus Eulerian). The model has been run at a
resolution of 2 degrees latitude by 2.5 degrees longitude with 20 ver
tical levels. Results from the experiments with Values of epsilon rang
ing from 0 to 0.4 show that epsilon = 0.2 gives the best overall forec
asts. The experiments with the time step for the dynamics varying from
15 to 60 min indicate that the forecasts are sensitive to the time st
ep for the dynamics, even when the time steps for the physical paramet
erizations are held constant. The forecasts with the 60-min time step
for the dynamics show the best overall objective skill scores. The two
versions of the semi-Lagrangian model, one with Eulerian physics and
the other with semi-Lagrangian physics, give similar forecast skill sc
ores. The semi-Lagrangian model is also compared with a corresponding
Eulerian model. It is found that the forecasts from the two models hav
e similar quality, even though the time step for the dynamics in the s
emi-Lagrangian model is 16 times as long as that in the Eulerian model
and the physical parameterizations have been developed and tuned for
the Eulerian model.