Seismic data acquired for exploration, development or production surveys are typically densely sampled, often driven by high-resolution imaging objectives. It is common to have tens or hundreds of thousands of shots in a marine survey. This poses a significant computational challenge for full-wavefield inversion (FWI), because each shot needs to be simulated individually. Even with the fastest computing systems, the cost increases tremendously while incorporating complex physics and inverting higher frequencies in the data. For example, the cost of 3D finite-difference elastic simulation is roughly 50 times the cost of acoustic simulation. And it increases by a factor of 16 if the frequency of the simulation is doubled. For 3D acoustic wavefield simulation, the cost is 256 times higher at 40 Hz than at 10 Hz. Therefore, large-scale 3D application of FWI is a challenging problem. The method involves several forward modeling and gradient computations of each field gather by numerically solving the wave equation. Even in a large computer environment, for an average seismic survey, the turnaround time for FWI can be significant. Furthermore, most FWI applications require several trial runs to choose suitable parameters to tune the inversion for a stable solution. The key to making FWI practical is to reduce the number of individual shots, thereby providing computational savings for both forward modeling and gradient computations.