Characterizing Steam-Filled Fracture Zones at the Soda Lake Geothermal Field Using Seismic Double-Beam Neural Network (DBNN)

Abstract

Previous geological studies in the Basin and Range of the Western U.S. have shown that conventional geothermal systems are usually located in geological settings of intense fracturing. The field studied here, Soda Lake geothermal field is one of the first geothermal fields developed in Nevada. The field contains steam vents and warm ground. However, most of the recent geothermal developments have been blind, with no surface hot springs or steam, and future developments are also likely to be blind. Therefore, characterizing subsurface fractures may provide an approach to unraveling blind geothermal systems. The sizes of small-scale fractures are much smaller than the seismic wavelength. Therefore, seismic migration cannot directly image small-scale fractures. We have demonstrated that our seismic double-beam (DB) method is effective in characterizing the fracture parameters in synthetic data: fracture orientation, density, and compliance. Augmented by a machine learning algorithm, our new double beam neural network (DBNN) algorithm can predict the locations and orientations of discrete fractures. We apply our DBNN method to the 3D Soda Lake seismic data to identify additional blind geothermal resources, particularly the shallow steam-charged fracture zones with large fracture compliance values. We identify four possible drilling targets showing high fracture compliance values, with one of them (Well 41-33) previously verified as a hot steam zone via drilling. Our seismic results on fractures and faults, in addition to known geology, well-logging information, and production data, can be used to identify new drilling targets.