The integrated formulation of two related experiments promises the consistency of the simulation and inversion results. The combination of parameters estimates, such as maximum likelihood points or mean value of an distribution, derived from the inversion of two separate inversions, leads to biases in the solution. On the other hand, the complete probability distribution of the two solutions could be combined correctly. However, assigning and handling proper probability distributions is difficult. The integrated and simultaneous inversion of both experiments leads to unbiased estimates. Additionally, the inversion allows a scientist to integrate a wide variety of experiments, including seismic time-lapse surveys that differ in geometry and instrumentation.
Limitations of integration. The integration of various experiments into a single inversion is conceptually simple and optimal, but poses several severe practical problems. For example, where should integration stop? An ambitious manager may want to include economic considerations and optimize for profit rather than for subsurface parameters. While I believe that such a formulation is actually desirable, I doubt the problem can be implemented successfully today.
Only the data of closely related experiments should be combined to an integrated inversion, for example, experiments that resolve the same model parameters, or the model parameters are related by well known direct physical relationship. Problems that are not at all or loosely connected should probably be simulated and inverted in separate computations.
Additionally, computational size and the availability and cost of compute time will limit our capability to integrate. The inversions of a series of small problems are less expensive than the inversion of the integrated problem.
Finally, the experts for each experiment need to collaborate in the integration of the experiment. I believe that flexible a software framework Schwab and Schroeder (1997) can facilitate easy integration of the various experiments without unduly burdening the individual expert scientist.