We have studied the development of multiscalar mantle flow beneath mid
-ocean ridges using a three-dimensional numerical technique. Modeling
includes various modes of heating and the shear flow induced by plate
motion away from a spreading center with and without transform offsets
. The computational box consists of two layers differing in their visc
osities. It is shown than when flow is generated by heat sources locat
ed for the most part in the lower layer, convective cells have horizon
tal dimensions proportional to the entire thickness of the box. It is
only when flow is driven by buoyancy forces in the top layer, or by tr
ansient cooling along the top interface, that transverse rolls are fou
nd to be confined within the top low-viscosity layer. These rolls rema
in generally located far from the ridge plane, except in the case of a
viscosity contrast of 3 orders of magnitude between the bottom and to
p layers. We conclude that segmentation beneath mid-ocean ridges requi
res both strong viscosity contrasts and the lack of deep convective fl
ow. Marine geophysical observations suggest that mantle upwelling bene
ath mid-ocean ridges is two-dimensional (sheetlike) beneath intermedia
te. fast and hotspot-influenced spreading centers and three-dimensiona
l (plumelike) beneath slow-spreading centers and ridges overlying cold
regions of the mantle, such as the Australian-Antarctic Discordance.
We propose that these observations can be explained by variations in t
he viscosity structure (and temperature) of the mantle along the mid-o
cean ridge system. Models including a transform offset show that the a
ctive portion of the transform is a region of mantle downwelling. The
toroidal flow induced by the offset shifts the plumes away from the ri
dge crests. The effects of the toroidal field are strongly magnified w
hen the return flow cannot penetrate in the lower layer due to a large
viscosity contrast. Stresses generated by the flow partly explains th
e development of inside comer highs and the evolution of transform fau
lts.