Time-lapse imaging using regularized FWI: a robustness study

SUMMARYFull Waveform Inversion (FWI) is an appealing technique fortime-lapse imaging, especially when the prior model informa-tion is included into the workflow. After baseline reconstruc-tion, several strategies such as: differential, parallel difference,and sequential difference can be used to assess the physical pa-rameter changes. Using the synthetic Marmousi data-sets, westudy which strategy could be more robust and give more accu-rate time-lapse velocity changes in the presence of noise. Weillustrate that the sequential difference method, starting from areconstructed baseline model and inverting the monitor data-set,can give a better result in the case of random ambient noise.However, the differential approach could also be interesting ifthe time-lapse response can be preserved from the noise level.INTRODUCTIONFWI is an alternative technique for velocity model building thatallows the reconstruction of high-resolution velocity models ofthesubsurfacethroughtheextractionofthefullinformationcon-tentofseismicdata(Tarantola,1984;VirieuxandOperto,2009).SincetheFWIapproachdelivershighresolutionquantitativeim-ages of macro-scale physical parameter, it could be a good can-didate for monitoring applications to reconstruct the parametervariation through a time evolution.Thetime-lapsereconstructionprocedurecanbedividedintotwosteps: (1) the baseline and (2) the monitor model reconstruc-tions. In order to obtain a robust high resolution time-lapseimage, it is necessary to reconstruct both baseline and monitormodels in a robust way. Recently, Asnaashari et al. (2012) haveproposedaregularizedFWIschemebasedontwomodelpenaltyterms in the misfit definition in addition to the data term: theTikhonov term (Tikhonov and Arsenin, 1977) to ensure smooth-ness, and a prior model term to attract the inversion towards agiven direction. The prior model misfit term is a way to in-troduce prior information into FWI workflow. For monitoringpurposes, where many different data types have been collectedin the target zone, such as sonic logs or stratigraphic recordings,such prior information should be used to increase the baselinereliability and accuracy, and also to recover time-lapse changesaccurately.Inthesecondstepofmonitoring,severalapproachescanbeusedfor the monitor reconstruction. The