Full waveform inversion: Looking closely into the Earth with seismic waves and big computers
Estimating the state of our planet is one of the main objectives of geophysics. We want to learn about the properties of the ground beneath us, the air around us, and water that surrounds us to for example say something about what is going to happen (i.e. predict nature) or what has happened in the past (i.e. reconstruct nature). Inferring the physical properties of the Earth system is an integral objective of the geophysicist.
Full-waveform inversion (FWI) is one method to infer the properties of a medium, by means of recording waves that pass through it. Based on numerical wave propagation through potentially complex media, it is the natural extension of ray tomography and finite frequency tomography. In geophysics, FWI is used mainly by seismologist to extract as much information from seismic recordings as possible.
Although FWI was conceptualized already in the late 70's, it only recently became practical when computational resources caught up with its requirements. FWI is therefore a relatively recent addition to the seismological toolbox, with many researchers exploring its facets. Not only is simulating wave propagation through a complex medium a computationally demanding task, but the fact that FWI is a non-linear optimization problem means that uncertainty quantification is not straight-forward.
This talk will feature recent efforts to extend FWI to large-scale Earth-tomography problems (continental and global), as well as non-linear uncertainty quantification for FWI using advanced Monte Carlo algorithms. Additionally, some efforts to transfer FWI knowledge between seismology and other fields of research where tomography is applied will be discussed.