Regional Prediction of Geothermal Systems in the Great Basin, USA using Weights of Evidence and Logistic Regression in a Geographic Information System (GIS)

1. Abstract Two modelling challenges were encountered while building a data-driven predictive map of geothermal potential for the Great Basin in the western United States. The first was that many of the input evidence layers that represent young faults or active tectonics were not conditionally independent with regards to geothermal training sites. The second was that a significant portion of the Great Basin is underlain by regional groundwater aquifers that are known to impede the formation of surface hot springs, thus potentially biasing the statistical weights of evidence layers relative to known geothermal systems. The conditional-dependency issue was resolved by combining some evidence layers together to form hybrid evidence maps based on quantitative or qualitative relationships among the data, and logistic regression was used to minimize some remaining conditional dependencies. The effect of groundwater aquifers was mitigated by defining the initial study area of the model as outside the known aquifers and then projecting posterior probability into the aquifer regions by matching unique conditions of the input evidence maps. The logistic-regression model predicts 23.3 +/0.2 geothermal systems in the aquifer regions while only 18 are known, suggesting that the aquifers regions are either under-explored or are concealing some geothermal systems.