Answers are grounded in this paper and cite their sources.
Simultaneous imaging method demonstrates effectiveness for delineating geological features in deep subsurface.
Imaging basement relief is challenging, given the unclear distribution of physical properties in the subsurface and the decreasing sensitivity of geophysical data to the geometry with the increasing burial depth. Most of the previous studies using the traditional Tikhonov-regularized inversion approach generally follow the path of inverting only for the geometry model with a known and uniform physical property model, and recent studies started to invert for both the basement relief and varying physical property models to improve the accuracy of imaging the basement relief. In this work, we present a supervised deep-learning-based simultaneous inversion method to image both the basement relief and the magnetic susceptibility variation in the basement rock while utilizing the conceptual prior information, like the presence of geologic faults, to guide the recoveries. Particularly, the U-net architecture, followed by a fully-connected layer, is adopted to map the information of multi-scale features from the data and prior-information domains into the inverted-model domains. Since there is no coupling between the basement relief and susceptibility, the network is replicated and trained separately with different emphases to invert for two models separately. The synthetic tests show that, although the trained network inverting for the susceptibility model is hindered by the low data sensitivity to susceptibility changes, it delineates the trend of the true susceptibility model and sharp boundaries between different sections, whereas the network inverting for the basement depth shows that its inversion result has higher fidelity compared to those from traditional methods. The field-data test over the Illinois Basin shows promising results on the inverted basement depth, while the inverted susceptibility models successfully reveal the mafic intrusion that was recorded in previous studies. Our field-data test also finds that the assumption of a non-magnetic sediment rock might be inappropriate when using the deep neural network to recover basement depth.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: