What Does the Mantle Remember of Its Convection History?

Mantle convection is a non-linear system. Convection patterns are therefore not directly proportional to initialization parameters. We investigate which parameters affect convection evolution, and the timescales over which these effects are observable. To do this, we run thousands of mantle convection simulations, each model having different, randomly selected, thermal, rheological, and compositional starting parameters. Our convection program is StagYY, run in 2D spherical-annulus geometry with self-consistent mineral physics properties calculated in Perple_X. StagYY allows melting and basaltic crust production and includes decaying radiogenic heat production in both core and mantle.

At various timesteps after initiation, the temperature, density, and composition fields can be extracted. We train neural networks to find the most likely initial parameters responsible for any convection pattern. The networks produce probability density functions, so that the relative likelihood of any value of the inputs parameters can be calculated, allowing inversion of the non-linear convection systems. This means that the uncertainty of the result is also provided.

After two billion years, we can find initial mantle potential temperature, initial CMB temperature, approximated initial mantle composition, reference viscosity, depth dependent viscosity, yield stress and the existence and thickness of any initial layer of primordial material. Our results suggest that the signal from these initialisation parameters is retained in the pattern of convection for much longer than 2 Gy. Our ultimate goal is to use tomographic models of the real mantle to investigate the ‘initialisation’ parameters of the Earth. We can also apply the same methodology to back-step convection, allowing us to build up a history of Earth’s mantle convection.