S063-0002
Full waveform inversion with crosstalk-free source encoding
Full waveform inversion with crosstalk-free source encoding
Wednesday, 16 December 2020
Poster
Abstract:
We present a new source encoding method that efficiently computes the gradient of the objective function in full waveform inversion. The method is based on the measurement of the Fourier coefficients of the wavefields after they reach steady state. The forward simulation contains all the sources with each source assigned with a unique subset of the frequency band and the stationary wavefield of each source can be extracted through an integration over a certain period of time. In this way, the gradient can be computed with only one forward and one adjoint simulation. Compared with other encoding methods, this method has no cross-talk noise between different sources and doesn't require each source to be recorded by all the stations. An additional benefit is that it enables double-difference measurement for all station pairs which eliminates the error from source time function and provides a robust way of measuring the amplitude. The effectiveness of this method is demonstrated through 3D regional synthetic inversions using S362ANI as the starting model and GLAD-M25 as the target model. We are currently working on applying this method to realistic 3D regional and global inversions.