In statistical inverse problems, the objective is a complete statistical description of unknown parameters from noisy observations to quantify uncertainties in unknown parameters. We consider inverse problems with partial-differential-equation (PDE) constraints, which are applicable to many seismic problems. Bayesian inference is one of the most widely used approaches to precisely quantify statistics through a posterior distribution, incorporating uncertainties in observed data, modeling kernel, and prior knowledge of parameters. Typically when formulating the posterior distribution, the PDE constraints are required to be exactly satisfied, resulting in a highly nonlinear forward map and a posterior distribution with many local maxima. These drawbacks make it difficult to find an appropriate approximation for the posterior distribution. Another complicating factor is that traditional Markov chain Monte Carlo (MCMC) methods are known to converge slowly for realistically sized problems. To overcome these drawbacks, we relax the PDE constraints by introducing an auxiliary variable, which allows for Gaussian errors in the PDE and yields a bilinear posterior distribution with weak PDE constraints that is more amenable to uncertainty quantification because of its special structure. We determine that for a particular range of variance choices for the PDE misfit term, the new posterior distribution has fewer modes and can be well-approximated by a Gaussian distribution, which can then be sampled in a straightforward manner. Because it is prohibitively expensive to explicitly construct the dense covariance matrix of the Gaussian approximation for problems with more than [Formula: see text] unknowns, we have developed a method to implicitly construct it, which enables efficient sampling. We apply this framework to 2D seismic inverse problems with 1800 and 92,455 unknown parameters. The results illustrate that our framework can produce comparable statistical quantities with those produced by conventional MCMC-type methods while requiring far fewer PDE solves, which are the main computational bottlenecks in these problems.

Source estimation is essential for all wave-equation-based seismic inversions, including full-waveform inversion (FWI) and the recently proposed wavefield-reconstruction inversion (WRI). When the source estimation is inaccurate, errors will propagate into the predicted data and introduce additional data misfit. As a consequence, inversion results that minimize this data misfit may become erroneous. To mitigate the errors introduced by the incorrect and preestimated sources, an embedded procedure that updates sources along with medium parameters is necessary for the inversion. So far, such a procedure is still missing in the context of WRI, a method that is, in many situations, less prone to local minima related to so-called cycle skipping, compared with FWI through exact data fitting. Although WRI indeed helps to mitigate issues related to cycle skipping by extending the search space with wavefields as auxiliary variables, it relies on having access to the correct source functions. To remove the requirement of having the accurate source functions, we have developed a source-estimation technique specifically designed for WRI. To achieve this task, we consider the source functions as unknown variables and arrive at an objective function that depends on the medium parameters, wavefields, and source functions. During each iteration, we apply the so-called variable projection method to simultaneously project out the source functions and wavefields. After the projection, we obtain a reduced objective function that only depends on the medium parameters and invert for the unknown medium parameters by minimizing this reduced objective. Numerical experiments illustrate that this approach can produce accurate estimates of the unknown medium parameters without any prior information of the source functions.

Reservoir parameter inversion is an essential geophysical problem to quantify the volume of an oil and gas reservoir. This problem can be identified as a nonlinear multiparameter optimization problem, in which the rock-physics model links seismic measurements including the velocity of the P- and S-waves to reservoir properties including porosity, water saturation, and permeability. Due to the complexity of the rock-physics and the nonlinear relationship, local search methods fail if the starting model is too far from the true model. Stochastic methods such as the genetic algorithm have global search ability, but they suffer from weak local search ability and premature convergence. We have developed a novel hybrid genetic algorithm to overcome these drawbacks of conventional genetic algorithm. We have used concepts of the self-adaptive method and evaluation criteria from the simulated annealing method to develop a new self-adaptive selection operator and a new self-adaptive hybrid crossover operator. With the two new operators, the hybrid genetic algorithm has better local search ability and is able to prevent premature convergence. We applied the hybrid genetic algorithm to the reservoir parameter inversion problem, and we used the Biot-squirt model as a rock-physics model. Numerical experiments demonstrated the improved search performance of the hybrid genetic algorithm against conventional methods in terms of solution quality, convergence rate, and convergence speed. We also tested our method on a series of laboratory and field data, which also confirmed the effectiveness of our method.

Least-squares reverse time migration is well-known for its capability to generate artifact-free true-amplitude subsurface images through fitting observed data in the least-squares sense.However, when applied to realistic imaging problems, this approach is faced with issues related to overfitting and excessive computational costs induced by many wave-equation solves.The fact that the source function is unknown complicates this situation even further. Motivated by recent results in stochastic optimization and transform-domain sparsity-promotion, we demonstrate that the computational costs of inversion can be reduced significantly while avoiding imaging artifacts and restoring amplitudes. While powerfull, these new approaches do require accurate information on the source-time function, which is often lacking. With-out this information, the imaging quality deteriorates rapidly. We address this issue by presenting an approach where the source-time function is estimated on the fly through a technique known as variable projection. Aside from introducing negligible computational overhead, the proposed method is shown to perform well on imaging problems with noisy data and problems that involve complex settings such as salt. In either case, the presented method produces high resolution high-amplitude fidelity images including an estimates for the source-time function. In addition, due to its use of stochastic optimization, we arrive at these images at roughly one to two times the cost of conventional reverse time migration involving all data.

No abstract