Estimations for the Thermal State of the Planetary Mantle from Mean Field Deep Learning
Abstract
The heat transfer in the interior of planets depends on highly nonlinear processes originating from the nonlinearity in mantle properties, phase, spin, and transformational transitions. The thermal state of the planetary interior can be parameterized by the results from numerical convection models for Earth-like planets and planets with and without significant plate tectonics by traditional inversion methods. In this study we employ machine learning algorithms to train predictor models that estimate the thermal state of planets based on their curvature (f = rcmb ⁄ rsurf ), Rayleigh number, and internal heating for two end member planets: rigid and free-slip surface planets. The training samples are obtained from 3D-spherical control volume model results. We employ regression learning algorithms and show that supervised machine learning (SML) techniques can successfully predict the thermal state of the planets with active and non-active plate tectonics. The deep learned models provide higher prediction accuracies than those obtained from simple machine leaning models with polynomialized features for the presented highly nonlinear problems reaching to 99% for both mean mantle temperature and mean surface heat flux.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMDI33C0045S
- Keywords:
-
- 0545 Modeling;
- COMPUTATIONAL GEOPHYSICS;
- 8120 Dynamics of lithosphere and mantle: general;
- TECTONOPHYSICS;
- 8124 Earth's interior: composition and state;
- TECTONOPHYSICS;
- 8180 Tomography;
- TECTONOPHYSICS