SCAIM Seminar: Tristan van Leeuwen, EOS, UBC

Images obtained from (active) seismic data are used by the oil and gas industry for geophysical exploration. Waveform inversion tries to obtain a detailed image of the subsurface by solving a PDE-constrained optimization problem. The scale of the problem –- typically 10^9 unknowns –- and the size of the data – typically 10^{15} data points – severely limits us in the approaches that we can use to solve the problem. On top of that, the problem is highly non-linear and the least-squares objective exhibits local minima.

In this talk, I will give an overview of typical issues in waveform inversion and present recent research that addresses some of these issues. This includes optimization techniques that work with only small portions of the data at a time, robust data-fitting schemes and formulations that incorporate other unknown parameters, such as the source signature. Finally, I discuss the object-oriented framework that we have developed in Matlab to test these algorithms.