We investigate the behavior of a spherical blob of buoyant fluid as gravity forces it toward either a rigid horizontal boundary or a free surface. The diapir fluid is assumed much less viscous than the ambient fluid. This fundamental problem is the simplest unsatedy model relevant to the ascent of hot plumes of buoyant material toward Earth's surface or the base of the lithosphere and closely models the heads of starting plumes. As the diapir approaches the boundary, it collapses in the vertical and spreads horizontally while a layer of the surrounding mantle is slowly squeezed out from between the diapir and the surface. Experimental results for the thinning and lateral spreading of the buoyant fluid, and for the thinning of the squeeze layer, are given for both the case of a rigid, nonslip boundary and that of a free surface. These are compared with similarity scaling laws based on a balance between the buoyancy of the diapir and viscous stresses in the diapir's surroundings. We also observe that the squeeze layer can become gravitationally unstable, leading to a bifurcation from convection on the scale of the original plume to convection on scales much smaller than the diapir. The vertical exchange on smaller horizontal scales enables the plume to more rapidly approach the boundary. At the time instability occurs the diapir has spread to roughly twice its initial diameter.

Application of these results, and previous results from surface uplift, to the plumes responsible for continental flood basalts is subject to knowledge of the local value of upper mantle viscosity. If this is taken to be 3¿1020 Pa s, most uplift takes place over approximately 5 m.y. Eruption of voluminous basalts will not take place until at least 5--10 m.y. after the surface has reached its maximum elevation. If small-scale instabilities do develop within mantle plume heads, they may be an essential mechanism allowing the top of the plume to ascend to the shallow depths required for extensive melting. It may also explain the observation of Hooper (1990) that volcanism in the Deccan and Columbian Plateau begins before the onset of crustal extension. ¿American Geophysical Union 1991