This dissertation focuses on the developpement of data assimilation methods to reconstruct the circulation of the Earth's mantle and the evolution of its surface tectonics for the last 200~Myrs. We use numerical models of mantle convection in which the surface dynamics is similar to the Earth's. By combining these models with plate tectonics reconstructions, it is possible to estimate the structure and evolution of the temperature field of the mantle. So far, the assimilation of plate tectonics reconstructions was done by imposing specific boundary conditions in the model (force balance, imposed velocities...). These techniques, although insightful to test the likeliness of alternative tectonic scenarios, do not allow the full expression of...