On the predictability limit of convection models of the Earth's mantle

International audienceReconstructing convective flow in the Earth's mantle is a crucial issue for a diversity of disciplines, from seismology to sedimentology. The common and fundamental limitation of these reconstructions based on geodynamic modelling is the unknown initial conditions. Because of the chaotic nature of convection in the Earth's mantle, errors in initial conditions grow exponentially with time and limit forecasting and hindcasting abilities. In this work we estimate for the first time the limit of predictability of Earth's mantle convection. Following the twin experiment method, we compute the Lyapunov time (i.e. e-folding time) for state-of-the art 3D spherical convection models, varying rheology and Rayleigh number. Our most Earth-like and optimistic solution gives a Lyapunov time of 136±13 My. Rough estimates of the uncertainties in best guessed initial conditions are around 5%, leading to a limit of predictability for mantle convection of 95 My. Our results suggest that error growth could produce unrealistic convective structures over timescales shorter than that of Pangea dispersal