Work Package II

Inversion and prediction by neural networks

SAR derived displacement fields constitute one of the major sources to explore the subsurface feature of the Earth by means of inversion. The latter involves modeling the deformation source and physical mechanism at depth that explain the displacement fields observed at the Earth’s surface. Solving such an inverse problem means finding a set of model parameters that best fit the observations. The state-of-the-art approaches, e.g. Markov Chain Monte Carlo (MCMC), require overwhelming amount of computing resources and prior knowledge about the model parameters to retrieve. The recent advent of neural networks based machine learning paradigms enables the development of new solutions to tackle the previous shortcomings of knowledge-driven inversions.

In this Work Package (WP), neural networks models have been proposed to resolve geophysical inverse problems.