If you have any problems related to the accessibility of any content (or if you want to request that a specific publication be accessible), please contact us at firstname.lastname@example.org.
Machine Learning Techniques Applied to the Nevada Geothermal Play Fairway Analysis
AuthorSmith, Connor M
AdvisorFaulds, James E
AltmetricsView Usage Statistics
This study introduces machine learning techniques to the Nevada geothermal play fairway analysis (PFA), which provided geothermal potential maps for 96,000 km2 of west-central to eastern Nevada. The motivation for this project is to support the evaluation of geothermal resource potential and the exploration for undiscovered blind geothermal systems in the Great Basin region of Nevada. The previous PFA study succeeded in utilizing the weighted combination of various geologic, geophysical, and geochemical features, indicative of permeability and heat, to both generate detailed geothermal favorability maps and identify several new blind geothermal systems. However, the project faced key limitations and challenges, including robust statistical analyses for the estimation of weights of influence for individual parameters, some incomplete datasets, and a limited number of training sites.To mitigate these challenges, this study incorporates new data developments and innovative machine learning techniques. Data developments include new training data and translating both the play fairway datasets (original and enhanced) and newly available datasets into a form compatible with machine learning techniques. Following the evaluation of various supervised and unsupervised machine learning techniques with the available data, two primary approaches were selected based on their performance and functionality. These techniques include 1) supervised probabilistic Bayesian artificial neural networks to produce detailed geothermal potential maps with confidence intervals, and 2) unsupervised principal component analysis paired with k-means clustering to both identify spatial patterns in geologic and geophysical feature sets and create new combined feature inputs.The comparative analyses of two principal sets of geological and geophysical input features highlight the potential that machine learning techniques offer to improve on the PFA. The analysis of feature set one, which comprises a set of regional permeability and heat data, illustrates a promising design for supervised Bayesian neural networks modeling to improve on the original regional permeability modeling in the PFA. Results from this feature set are selected to organize spatial patterns for the major structural-hydrologic domains within the study area, including the Walker Lane, western Great Basin, central Nevada seismic belt, and carbonate aquifer. The analysis of feature set two, which includes the same regional feature layers as in feature set one with the addition of local permeability features, illustrates how a model design may find a balance between disparate data types to produce predictive favorability maps that yield similar results to the original fairway map from the PFA. Information presented in this study related to the spatial patterns of elevated geothermal potential may have promising implications for future geothermal exploration efforts in the Great Basin region of Nevada and beyond.