Title: Seismic imaging with massive datasets using sparsifying transformation methods
Researchers: Dr Francois Lavoue, Dr Sergei Lebedev
The tremendous recent growth in the volumes of seismic datasets now offers an unprecedentedly dense sampling of the Earth’s interior. Combined with advances in imaging techniques, notably waveform tomography, this allows for an increasingly detailed imaging of the subsurface at various scales: from lithospheric scale, using dense arrays of broadband seismometers deployed all over the globe, down to the basin scale, using massive multichannel seismic datasets acquired for exploration purposes. It also presents a major challenge: the enormous size of the computational problems required for the processing, modelling, and inversion of the data. One issue is that conventional approaches for parametrizing high-resolution 3D models require prohibitively large computations. Another issue is the non-unicity of inverse-problem solutions, which is classically tackled by introducing ad hoc smoothing constraints. This project aims to address both of these issues by using new concepts from the compressive sensing theory, which exploit the sparsity of the models in wavelet basis (e.g. Candès and Wakin, 2008). Just as sparsity of digital images enables their compression (e.g. JPEG-2000), sparsity of subsurface images allows us to dramatically reduce the number of parameters involved in the inversions with a minimum loss of information (Simons et al., 2011).
Hypothesis Wavelet-based sparsifying transformations are expected to reduce the computational costs and increase the robustness of waveform inversions, while enabling a multiscale resolution analysis of the solutions.
Objectives of the research project (as described in the approved proposal) Adopt the very promising new methods (wavelet-based sparsifying transformations, developed in applied mathematics) for solving demanding modern geophysical problems that involve massive datasets. Implement and test the new methods with applications to very large geophysical datasets (including at whole-Ireland scale). Create readily usable tools for joint applications with industry partners.