Mantle convection plays a fundamental role in driving evolution of oceanic and continental lithosphere. In turn it impacts a broad suite of processes operating at or close to Earth’s surface including landscape evolution, glacio-eustasy, magmatism and climate. A variety of theoretical approaches now exist to simulate mantle convection. Outputs from such simulations are being used to parameterise models of landscape evolution and basin formation. However, the substantial body of existing simulations has generated a variety of conflicting views on the history of dynamic topography, its evolution and key parameters for modelling mantle flow. The focus of this study is on developing strategies to use large-scale quantitative stratigraphic observations to asses model predictions and identify simulation parameters that generate realistic predictions of Earth surface evolution. Spot measurements of uplift or subsidence provide useful target observations but are often controlled by tectonic processes, yet avoiding areas where tectonics have influenced vertical motions is challenging. To address this issue, we use large inventories of stratigraphic data from across North America with contextual geophysical and geodetic data to constrain the regional uplift and subsidence history. We demonstrate that a suite of fairly typical simulations struggle to match the amplitude, polarity and timing of observed vertical motions. Building on recent seismological advances, we then explore strategies for understanding patterns of continental uplift and subsidence that incorporate (and test) predicted evolution of the lithosphere, asthenosphere and deep mantle. Our results demonstrate the importance of contributions from the uppermost mantle in driving vertical motions of continental interiors.