Cooperative deep learning inversion: Seismic-constrained CSEM inversion for salt delineation

Abstract

Salt structure imaging is one of the most important problems in the field of hydrocarbon exploration. To resolve this issue, the integration of diverse geophysical data has emerged. In this study, we proposed the cooperative inversion with seismic and controlled-source electromagnetic (CSEM) data based on the supervised deep learning (DL) technique for precise salt delineation. CSEM data, which are effective in distinguishing a salt body with high electrical resistivity from the surrounding media, were used as the data of the inversion, and a high-resolution information derived from seismic data was applied as the constraint. To combine the seismic constraint with CSEM data, the modified UNet was adopted as an inversion operator based on DL. For training the DL model based on the network, resistivity models, including a salt body with arbitrary shape and size, and the corresponding CSEM data calculated through numerical modeling were generated and used as the label and input data, respectively. In addition, the seismic constraints, which were supposed to be obtained from the seismic image, were provided to the DL model in the training phase. Finally, we applied the optimum model to the test data acquired using the modified SEAM model. Test results demonstrated that the integration of seismic constraint leads to enhanced delineation of the salt body by providing definite upper boundary. This study has presented the promising potential of DL inversion to integrate multiple geophysical data.