| Title | Using Deep Learning to Map Quaternary Faults in Western USA |
|---|---|
| Authors | Bastien HERMANT, Lauriane KIERSNOWSKI, Mathieu BELLANGER |
| Year | 2025 |
| Conference | Stanford Geothermal Workshop |
| Keywords | Deep learning, fault mapping, quaternary faults |
| Abstract | Faults are an essential component of geothermal systems, particularly in the Great Basin region where most hydrothermal systems are fault-controlled. Accurate and complete fault mapping is therefore of prime importance for assessing the geothermal favorability of an area or a prospect, as it allows for defining the structural network, its orientation with respect to the stress field and the potential favorable structural settings (step-over, accommodation zone, etc). Quaternary faults databases exist across the western USA. However, their accuracy and the homogeneity of mapping between the sub-regions/states is sometimes insufficient for an objective analysis of favorability at a regional scale. This could create a bias between regions where fault mapping is robust and those where it is incomplete. To overcome this bias, TLS Geothermics has developed a deep learning approach using convolution neural networks for geological fault detection based on remote sensing images. By selecting the appropriate data, the neural network is able to produce fault prediction maps at regional scale. It can therefore be used to better assess geothermal potential at a regional scale. Using an adapted version of the neural network on a local scale and with higher resolution remote sensing data can also provide more robust structural models on a prospect scale. |