Scientific Machine Learning for Solving Multiphysics Inverse Problems in Earth Science (MULTIVERSE)

Introduction

Geophysics is important to study the earth below its surface without drilling or digging, for understanding of the geology, plate tectonics, volcanism, seismic activity, but also mineral resources, geoenergy, groundwater and contaminant detection. Through remote interactions, like electromagnetism or gravity, geophysical fields measured on the earth allow the inference of subsurface properties, e.g. electrical resistivity or density. However, when resolving large and complex environments the procedure for recovering the 3D earth properties, inversion, is computationally extremely intense and at the same time produces ambiguous results. In MULTIVERSE, physics-informed neural networks will be used to achieve significant speed-up of these computations by learning of the physical laws and then predicting fields much more efficiently than in rigorous computations, and add error estimates of the model parameters, which is so far hard to achieve. This will ultimately increase the imaging power of geophysics.

Contact person: Jochen Kamm

Return to projects