Integration of Data, Numerical Inversion, and Unsupervised Machine Learning to Identify Hidden Geothermal Resources in Southwest New Mexico

Abstract:

Southwest New Mexico (SWNM) has several low- and medium-temperature geothermal resources in operation. Specifically, the Basin and Range province of SWNM has more potential geothermal resources (e.g., near Lordsburg area). Further exploration of these resources requires good data. Till now, SWNM has a total of 210 data points sampled at various spatial locations. These data points include three geological, 11 geochemical, two geophysical, and seven temperature-related attributes with substantial uncertainties. Geological attributes are porosity, fault density, and distance of fault from data points; 11 geochemical attributes include seven major cations and anions, Li, B, pH, and SiO₂; two geophysical attributes contain gravity and magneto-telluric data; and seven temperature-related data consist of basal-heat flow, aquifer temperature, temperature gradient, and four geothermometer data. To estimate the geothermal potential in the region, we first perform simulations related to coupled fluid flow and heat transport. That is, 1D lateral heat flow and tracer transport equations were solved and the primary five attributes (basal-heat flow, aquifer temperature, temperature gradient, and concentration of Li and B) were calibrated. Later, post-calibration uncertainties were quantified of these five attributes. Markov Chain Monte Carlo (MCMC) was used to generate distributions of these parameters. Using mean values of five MCMC distributions and other original values, a data matrix was created to utilize in an unsupervised machine learning technique called non-negative matrix factorization with customized k-means clustering (NMFk). NMFk is a robust technique for finding hidden patterns in data that are not visible in open eyes or other exploratory data analytics. Post-calibration uncertainty quantification revealed uncertainties in the primary five attributes. NMFk analysis will reveal hidden geothermal resources and indicate the best parameters to characterize the geothermal system in SWNM.