MACHINE LEARNING TO CHARACTERIZE REGIONAL GEOTHERMAL RESERVOIRS IN THE WESTERN USA

Geochemistry is a critical tool for exploring and characterizing geothermal resources. Geochemistry provides critical information about water temperature, fluid type, possible in/outflow, number of reservoirs, and interaction between water and rock of a geothermal reservoir. The associated data consists of hidden information that is a proxy for thermal anomalies. This includes major cation/anion, trace elements, temperature, etc. In this study, we employ recent advances in machine learning (ML) to characterize regional characteristics using geochemistry data of five geothermal (GT) reservoirs in the USA. The used ML method is non-negative matrix factorization with customized k-means clustering (NMFk). NMFk is a robust method to identify (1) latent signals, (2) the optimal number of signals, (3) dominant attributes to characterize the signals, and (4) spatial signatures. The signals are encapsulated in the data but cannot be observed using traditional data analytics. Also, these distinct signals allow us to make a generalized prediction of attributes that characterize new geothermal resources. NMFk is applied to five GT reservoirs, which are the Great Basin, Steamboat Springs, eastern Nevada, Pinto Hot Springs, and Darroughs Hot Springs. This study successfully characterizes each reservoir, finds dominant attributes to characterize the unique behavior of the reservoirs. Finally, unique spatial characteristics are also interpreted.