We study incorporating well log information into Full Waveform Inversion (FWI) and compare the result with a line search optimization scheme. We test this approach using a poor starting model i.e. a linear velocity v(z) model and also using smoothed version of the true velocity model. Our results show that incorporating well information into FWI saves a lot of computational time compared with a line search optimization scheme. Although the inverted model is far from convergence, we learn a few interesting things: conventional FWI cannot by itself provide good estimate for layer properties when the starting model is poor such as the case of a linear v(z) model, the information from well logs can be used to constrain the line search calculation and still save computational time, finally low frequencies are crucial for convergence.
View full article as PDF (2.12 Mb)