At the Earth's surface, continents and oceans impose different thermal boundary conditions at the top of the mantle. Laboratory experiments are used to investigate the consequences of this for mantle convection. The upper boundary of the experimental tank was made of copper plates enforcing a fixed temperature and had a conductive lid of finite width in the middle. Beneath this lid, the thermal boundary condition was of the ''mixed'' type, with a Biot number depending on the dimensions and thermal conductivity of the lid. Experimental values of the Biot number were scaled to Earth values. Experiments were run for a large range of Rayleigh numbers, from 104 to 107, and for several lid widths.

The effects of temperature-dependent viscosity and of the shape of the lid were investigated. At steady state, in all cases, there is an upwelling beneath the conductive lid, which feeds two symmetrical and elongated convective cells. Three different dynamic regimes were identified as a function of Rayleigh number, independently of the lid width. At Rayleigh numbers lower than 1.2 105, the upwelling is steady both in geometry and temperature structure. At Rayleigh numbers between 1.2 105 and 2 106, this central upwelling is intermittent. At larger values of the Rayleigh number, there is no longer a simple upwelling structure, but a set of small plumes rising together and distorted by a cellular circulation of large horizontal extent. Thus the conductive lid always imposes a large-scale flow pattern. The length of these convective cells is a function of lid width. It is equal to the lid width at large values and decreases to the Rayleigh-B¿nard value as the lid width decreases to zero. A fluid loop model explains the most important features of this form of convection. The cell length is such that the upwelling temperature is minimized for a given Rayleigh number and lid width and is an increasing function of lid width and a decreasing function of Rayleigh number. Transient experiments demonstrate that the large-scale flow structure develops rapidly with even small horizontal temperature differences. Implications for the Earth are that large-scale convection cells exist in conditions which, in the absence of continents, would probably lead to a chaotic convection pattern dominated by plumes. At high Rayleigh number, continental breakup is effected by a large-scale line upwelling structure which includes a number of individual plumes. ¿ American Geophysical Union 1995