Joint reflection and diving FWI using graph-space optimal transport and structure-guided smoothing on benchmark data

Introduction Joint full waveform inversion (JFWI, Zhou et al., 2015) builds a P-wave velocity (Vp) macromodel exploiting simultaneously the information carried by diving waves and reflections (e.g. in RWI, Brossier et al., 2015), thus obtaining deep Vp-updates while enforcing the constraint on the shallow subsurface. Here we devise an acoustic JFWI+Impedance-WI (IpWI) strategy on the Chevron-2014 benchmark limited-offset reflection elastic dataset. JFWI is performed using a graph-space optimal transport objective function (GSOT, Métivier et al., 2019) and takes advantage from along-structure smoothing based on the impedance reflective image. We compare GSOT and L2 objective functions, and show the benefits of structure-oriented smoothing (Trinh et al., 2017). Finally, the JFWI solution is used as starting model of a multi-scale Vp-FWI, attaining an excellent match with the virtual log, a satisfactory focusing of the common image gathers (CIGs), and an improved stationarity of the source wavelet estimation.