Inferring Permeability in Geothermal Fields Using Seismic Imaging and Machine Learning

In many geothermal systems, fluid-flow permeability is controlled by faults and fractures. Large-scale faults may be visible in migrated seismic images, but small-scale fractures are commonly below conventional seismic resolution. Characterizing these structures before drilling is therefore important for identifying potential geothermal-fluid pathways and reducing exploration uncertainty. It is commonly observed that geothermal seismic data are often noisy, which can limit the effectiveness of conventional methods.

Because 3D seismic survey is costly, its use in geothermal exploration must provide information beyond conventional structural imaging to justify the investment. This talk presents novel seismic imaging and machine-learning methods developed to characterize faults and fracture networks and infer permeability-related properties. Large-scale faults are detected from seismic images using convolutional neural networks, while small-scale fractures are characterized using a physics-guided Double-Beam Neural Network.

The Double-Beam Neural Network analyzes interference patterns produced by seismic waves scattered from fractures to estimate fracture locations, orientations, spacing, connectivity, and relative compliance—properties that may be related to fluid content and permeability. The methods were tested using synthetic data and applied to a 3D field seismic dataset from the Soda Lake geothermal field in Nevada.

The analysis identified a shallow steam-filled fracture zone near a known productive well and revealed three additional prospective drilling targets. Machine-learning analysis also delineated deeper fault systems that may influence geothermal-fluid circulation. These results demonstrate how seismic-wave physics, imaging, and machine learning can add significant value during geothermal exploration and development by extracting permeability-related information from geothermal reservoirs. The approach may help identify blind geothermal systems, reduce exploration uncertainty, and improve the selection of drilling targets.