Targeted nullspace shuttles as enablers of time-lapse seismic full waveform inversion with sparse monitoring acquisitions

Traditional time-lapse full waveform inversion (FWI) is sensitive to the repeatability of acquisition geometries between baseline and monitor surveys. This implies a need for extensive, costly acquisition over the lifetime of a time-lapse program, which is a barrier to the economical use of FWI for measurement, monitoring, and validation of CO2 geostorage. However, it is not clear that this sensitivity is a fundamental feature of time-lapse FWI. Qualitatively, we should expect the acquisition intensity required to characterize spatially-localized change to be significantly less than that needed to fully characterize the baseline subsurface. This has motivated an investigation into whether acquisition requirements may be relaxed, given appropriate strategies for differencing. Targeted nullspace shuttles are a convenient and relatively low-cost tool for assessing uncertainty in FWI, which have recently been adapted for use in forming time-lapse differences of FWI-generated models. These shuttles update our FWI results to obtain the smallest-difference models allowable given fixed misfits for each of the baseline and monitoring inverse problems. If it is true that fewer monitoring data are required to constrain the time-lapse change, we hypothesize much of the unwanted variability in the difference model arising from sparse acquisitions should not be necessary to maintain misfit. It follows that much of this noise should be eliminated by time-lapse shuttles. To find evidence for or against this hypothesis, we conducted a set of synthetic feasibility tests, which are suggestive that it does hold. Both surface and borehole experiments are simulated with an Earth model and CO2 plume constructed from well-log data and CO2 modeling results from the Carbon Management Canada Newell County Field Research Station. Baseline and monitoring misfit levels can lead to “shuttling to zero”, which is either a sign of the insufficiency of a given time-lapse dataset, or an indication that additional refining iterations are needed.