Project lead: Aleksandr Y. Aravkin (website)
Collaborators: Dmitriy Drusvyatskiy (UW) and Tristan van Leeuwen (Utrecht)
Geophysical exploration, reservoir characterization, and subsurface imaging are all examples of seismic inverse problems. These problems are extremely large scale: model parameters of interest are typically gridded parameter values over a 3-dimensional mesh. Data for these problems are collected over several months, and literally carried by the truck-full for processing.
Using low-rank nonconvex optimization formulations we can interpolate 80% missing seismic data, obtaining full data volumes from severely under-sampled sources. These volumes are used for further processing and imaging at a fraction of the cost. Some of the results are included below.
For a preprint, please see: http://arxiv.org/abs/1607.02624