Bauxite plays a crucial role in metallic and non-metallic industry. The surface-exposed salento-type bauxite deposits have been largely exploited and developed. With the increasing demand of these resources, it is important but very challenging to explore the potential bauxite deposits in the deep earth. In this paper, based on new developments in transient electromagnetic (TEM) technologies, we conducted explorations of the sedimentary bauxite in Dengli-Tianyanggumei and Dajia mining areas in western Guangxi province. To achieve a fast and high-resolution inversion, we adopt an array-based observation strategy for large-scale 3-D TEM and collect EM data inside and outside the transmitting loop. Compared to traditional TEM surveys, the observation strategy can quickly acquire the data for large-scale surveys and improve the data acquisition efficiency by more than 25 times. We then use a 3-D inversion algorithm to estimate the underground conductivity structure and analyze the distribution of the sedimentary bauxite. To do that, we discretize the undulating surface and transmitter-receiver locations with unstructured grids and employ the finite-element and quasi-Newton methods to achieve high-resolution imaging of subsurface electrical structures. Since the observation strategy greatly reduces the number of transmitters, the efficiency of 3-D EM inversions can be significantly improved. Experiments over two mining areas show that our inversions can clearly recover the underground resistivities. The inferred burial depth and spatial distribution of the sedimentary bauxite are in agreement with the drilling data. By combining ERT results and geological data, we illustrate the impact of faults on the spatial distribution of potential sedimentary bauxite deposits.

As the structure of the underground space becomes increasingly complex, traditional two-dimensional seismoelectric methods are no longer adequate for the comprehensive exploration. To achieve precise imaging of the underground space, it is in urgent need to develop three-dimensional full-waveform modeling techniques. In this paper, we propose a three-dimensional time-domain finite-element method to solve the seismoelectric wave field in saturated porous media. Since the electroosmotic feedback is very small, we can ignore the mechanical disturbance caused by the electromagnetic fields induced by seismic waves, and thereby can decouple the electrokinetic coupling equations and separately solve the seismic and electromagnetic waves. For the simulation of seismic wavefield, we employ the explicit finite-element method, and utilize a lumped mass matrix instead of a consistent mass matrix to facilitate explicit recursion. Additionally, we apply the complex frequency-shifted unsplit perfectly matched layer technique to effectively handle seismic boundary conditions. Then, the velocity fields obtained by solving the poroelastic equations serve as the source term of the electromagnetic equations, and the finite-element method is used to solve the electromagnetic wavefield. Considering that the huge velocity difference exists between the electromagnetic and seismic waves, we adopt an unconditionally stable implicit method for the solution of the electromagnetic wavefield. By combining explicit and implicit recursion, the computational efficiency can be improved significantly. The accuracy of our time-domain finite-element algorithm is validated by checking our results against the analytical solutions for a half-space model. Furthermore, we conduct numerical simulations and analyses on a typical block model and a modified SEG/EAEG salt dome model.

The conventional geo-electromagnetic data inversions are mostly based on the gradient optimization methods. However, this type of methods can only provide single “optimal” inverse model under specific prior conditions, which cannot effectively evaluate the reliability and uncertainty of the inversion results. The widely used uncertainty quantification methods are based on the theory of Bayesian inference. Although they have achieved success in many applications, yet they suffer from the curse of dimensionality and low efficiency. To overcome these problems, we propose a novel UQ strategy for geo-electromagnetic inversions based on Bayesian processes and surrogate modeling with adaptive deep neural network. In this method, an embedded DNN is used for the forward modeling in the Bayesian inference to improve computational efficiency. The training of the DNN is divided into two stages. First, a pre-designed small training set is used and the resulting DNN only gives low-accuracy result. Second, this DNN is fine-tuned dynamically during the Metropolis-Hastings sampling process, in which the training set is adaptively supplemented according to the modeling errors. Comparing to the conventional data-driven approach, this dynamically adaptive constructing method of the training set can greatly reduce the training set and constantly maintain high accuracy in the forward modeling. We demonstrate the effectiveness and practicality of our surrogate modeling Bayesian and analyze the effects of different sampling numbers, noise levels, prior distributions, and sampling radius. Comparing with the Occam’s inversion and conventional Bayesian inversions, our method shows good robustness and high accuracy, making it an effective Bayesian inversion technique.

Marine controlled-source electromagnetic (MCSEM) inversion plays a crucial role in hydrocarbon exploration and pre-drill reservoir evaluation. Deep learning techniques have been widely used in geophysical inversions. Although they work on theoretical data well, their performance on survey data needs to be improved. Since no constraint of physical laws is applied in the training phase, the trained neural network often exhibits large errors when extended to new datasets with different distributions from the train set. To solve this problem, we add a differentiable marine EM forward operator at the end of the neural network that maps the network-predicted results back to the response data. We incorporate a data error term to the loss function and the gradient of data error with respect to model parameters in the gradient back-propagation process so that we can successfully introduce the physical law constraints into the network training process. Experiments on synthetic data validate the effectiveness of our Physics-driven Deep Neural Network (PhyDNN) inversions. It performs significantly better than the conventional DNN as it can recover the model accurately while maintaining data fitting. Tests on theoretical data with different noise levels further demonstrate the superiority of our PhyDNN, which can achieve stable inversions under high noise levels. Moreover, we use the t-distributed stochastic neighbor embedding (t-SNE) algorithm to analyze the similarity between the train sets and real data. The results show that the real data falls within the data distribution of the train sets, ensuring the credibility of the inversion results. Finally, we use PhyDNN to invert an EM survey dataset acquired over a deep-sea sedimentary basin. The inversion results match well Occam’s inversions, indicating that our physics-driven network has enhanced the data adaptability and overcome the limitation of conventional DNN in handling new data.

Airborne electromagnetic (AEM) surveys usually covers a large area and create a large amount of data. This has limited the application of three-dimensional (3D) AEM inversions. To make 3D AEM data inversion at a large scale possible, the local mesh method has been proposed to avoid solving large matrix equations in 3D AEM modeling. However, the local mesh only saves the computational cost and memory during forward modeling and Jacobian calculations. When the survey area is very large, the cost for storing and solving the inversion equations can be very high. This brings big challenges to practical 3D AEM inversions. To solve this problem, we develop a 3D scheme based on the block coordinate descent (BCD) method for inversions of large-scale AEM data. The BCD method divides the inversion for large models into series of small-local inversions, so that we can avoid solving the large matrix equations. Numerical experiments demonstrate that the BCD method can get very similar results to those from the existing inversion methods but saves huge amounts of memory.