The two main volcanic centers on Mars, Tharsis and Elysium, are often interpreted in terms of mantle plume hotspots, even though there are several problems with the plume hypothesis for Mars. We present results of 2D cylindrical shell numerical mantle convection experiments in which we try to ascertain whether flushing of the hot lower mantle could provide a mechanism for the generation of a small number of plume-like features, i.e., localized upwelling of hot material. In this scenario the formation of hot upwellings is driven from the top by cold downwellings rather than from a hot thermal boundary layer at the CMB. First we construct a range of Mars interior structure models consistent with observations in order to demonstrate that the presence of a thin lower mantle in the martian interior is a viable scenario. Then we use a series of numerical convection experiments to investigate the effects of solid-state phase transitions, different stratified and temperature-dependent viscosity models, and the presence of a thick southern hemisphere crust on the operation of such a mechanism. Our results show that it is possible to generate hot strong localized upwellings from top-down dynamics if the lithosphere is thin or actively involved in the convective pattern. The presence of a thick, immobile, insulating southern hemisphere crust reduces the number of upwellings, and the perovskite phase transition causes a focusing of the upwellings. Further experiments demonstrate that an initial 500 Myr phase of mobile lid is sufficient to start this process create an upwelling which is stable for billions of years. (c) 2006 Elsevier Inc. All rights reserved.