Paper Summary
We classify full-waveform inversion (FWI) projects according to the types of events used to invert marine streamer data. In general, an adequate model update requires the seamless introduction of wavenumbers that are missing in the starting model. It has been our experience that this may require extraction of the lowest possible wavenumbers from the data, as well as the application of preprocessing procedures and scattering kernels that are optimal for each type of arrival. In a shallow-water environment, and with sufficient offsets, refracted arrivals and diving waves can provide an adequate inversion for shallow velocity structures. We demonstrate this principle using the linearized (Born) scattering kernel for field data from the North Sea. For depths below the deepest turning point of recorded refractions and diving waves, we show that reflections can also be used to advantage under Born scattering assumptions, if the signal-to-noise ratio of the data allows the use of exceptionally low frequencies, say down to 2 Hz. It is more common, however, to work with data that have good signal-to-noise ratio only for frequencies above 3-4 Hz. Under these circumstances, we show that deep reflections can still yield a useful inversion, provided that the velocity update is performed using a depth-integrated form of the reflectivity. For each iteration, this strategy adds nonlinearity to the velocity recovery, and it thus tolerates a starting model that is less accurate than that required using the Born scattering kernel. We demonstrate the application of this strategy for data from the Gulf of Mexico.