The Gregory et al. parametrization of momentum transport by convection is s
hown to be more prone to numerical instability than other parts of the conv
ection scheme, because of an additional element of numerical diffusion. The
limiting Courant number for Linear stability is (1 + c)(-1), where c is th
e parameter which controls the dependence of the pressure gradient on envir
onmental shear. Some revised numerical treatments cure the instability prob
lem but reduce forecast skill, apparently because the diffusive element of
the original formulation is beneficial, even though it is artificial. A pos
sible solution to the problem is to use an implicit treatment of the origin
al formulation which is absolutely stable.