Image-domain time-lapse inversion with extended images

Time-lapse seismic data is used to monitor subsurface changes occurring between consecutive surveys. One way to look at the differences is to create images from each survey and simply subtract them to obtain a difference. It is possible to model image differences directly from the slowness change between two surveys: this leads to a forward modeling operator that essentially performs differential migration. We further exploit the concept of image-domain time-lapse inversion by introducing extended-image time-lapse differences. Extended images are generalizations of the conventional imaging condition, and the resulting images are an extension of subsurface image gathers. We discuss objective functions for image-domain time-lapse inversion from both conventional and extended images. These depart from the more conventional approaches to time-lapse inversion whose objective functions are parameterized in the recorded data domain. By analyzing the forward operators for conventional and extended image differences, we show that the use of image extensions yields a larger data space thus offering additional constraints for time-lapse inversion. We illustrate these concepts numerically by comparing sensitivity kernels as well as inversion results from conventional versus extended time-lapse images.