The dominant parts of the recorded field in a typical marine CSEM survey consist of refracted and guided events. Reflected events are very weak or non-existing. Thus, classical seismic imaging techniques like stacking, post-stack migration or pre-stack migration are either ruled out or sub-optimal as methods for imaging the resistive subsurface. Only relatively low frequency components are above the ambient noise level at intermediate and large source-receiver offsets (above 2 km) due to the diffusive nature of the propagating fields. The default setup for marine CSEM data acquisition is that all receivers record data with offsets of 12.5 km or more. Low frequencies (0.1 Hz to 10 Hz) are transmitted to reach sufficiently deep into the subsurface. Phase velocities are typically in the range 1 km/s to 10 km/s for these frequencies. Thus, marine CSEM data are well suited for FWI since low frequency, refracted data are available at large source-receiver separations. The obvious price to pay for having only low frequencies available is limited resolution. This is similar for the inversion of low frequency seismic data.
Forward modeling is a key component for any inverse scheme. One option for modeling marine CSEM data is to solve the Maxwell equations in the wave domain. The desired diffusive solution can then be extracted from the wave solution by a transform along the time axis. The reason for following this procedure is numerical efficiency. It can be demonstrated that this transform along the time axis effectively extracts refractions from the wave solution. The same transform can be applied to seismic data. The benefit is that low-frequency/large-wavelength refracted fields can be extracted from relatively high frequency seismic data. The actual inversion must then be formulated in the diffusive domain much like the inversion of marine CSEM data.