Numerical models of plate-scale convection confined to the upper mantle predict large deviations from observed ocean bathymetry, gravity, and geoid, while whole mantle models yield first-order agreement with these observations. The upper mantle models fail because there is insufficient radioactivity in the upper mantle to explain the surface heat flux, the upper mantle must therefore be heated mainly from below, the resulting hot boundary layer generates buoyant material, and when this buoyant material rises to the base of the lithosphere, it generates large positive anomalies in topograpy (2 km), the geoid (30 m), and gravity (50 mGal). The concept of the magnitude of a hotspot swell is introduced: it is the rate at which new topography is created, expressed as a volume per unit time, and is a measure of the buoyancy and heat fluxes in the underlying plume. By this measure the major Pacific hotspots dominate the Earth's total plume flux, but plumes account for less than 10% of the Earth's total heat flux. This is comparable to the amount of heat likely to be coming from the core and supports the idea that plumes originate at the core and supports the idea that plumes originate at the core mantle boundary and not at the 670--km transition. These results indicate that mantle convection is dominated by a plate-scale flow which penetrates throughout the mantle, with a secondary mode involving plumes rising from a weak thermal boundary layer at the bottom of the mantle.