Times are displayed in (UTC-07:00) Mountain Time (US & Canada) Change

272-6 The Importance of Subsurface Structure for Predicting Hydrothermal Upflow When Using Supervised Machine Learning

Session: Hydrogeology of Geothermal Resources: From Shallow Systems to Deep Reservoirs

Presenting Author:

Stanley Mordensky

Authors:

Mordensky, Stanley Paul¹, DeAngelo, Jacob², Caraccioli, Pascal Domingo³, Burns, Erick Robert⁴, Lipor, John Jacob⁵, Pepin, Jeff Donald⁶ (1) US Geological Survey, Geology, Minerals, Energy, Geophysics Science Center, Bozeman, Montana, USA, (2) US Geological Survey, Geology, Energy, Minerals, and Geophysics Science Center, Chicago, Illinois, USA, (3) Department of Earth and Atmospheric Sciences, Cornell University, Ithica, New York, USA, (4) US Geological Survey, Geology, Minerals, Energy, and Geophysics Science Center, Moffett Field, California, USA, (5) Portland State University, Electrical & Computer Engineering, Portland, Oregon, USA, (6) US Geological Survey, Central Energy Resources Science Center, Denver, CO, USA,

Abstract:

As approaches for applying supervised machine learning to create geothermal resource maps and similar products (for example, play fairway analyses) mature, opportunities to improve model predictions become increasingly concentrated in the choice of more informative independent input datasets (that is, input features). Recognizing that topography in the Great Basin of the western United States reflects subsurface structures that contribute to permeability (for example, faults), recent work by DeAngelo and others isolated local topography from regional trends in elevation change to deconvolute signals that may help to locate hydrothermal systems. Herein, we evaluate the impact of replacing raw elevation with separated features for regional elevation trend and detrended elevation as input features on favorability maps by assessing model performance through the ability to predict known power‑producing hydrothermal systems not used in the training data and through visual inspection of the favorability maps. In doing so, we find that deconvoluting elevation signals enhances the predictive skill of the model for known power‑producing systems not used in the training data and transforms predicted high-favorability areas from cutting across basins and ranges to following range fronts within basins, providing a more accurate and geologically defensible favorability map.

Geological Society of America Abstracts with Programs. Vol. 58, No. 5, 2026

doi: 10.1130/abs/2026AM-15736

The Importance of Subsurface Structure for Predicting Hydrothermal Upflow When Using Supervised Machine Learning

Description

Session Format: Oral

Presentation Date: 10/13/2026

Presentation Start Time: 03:10 PM

Presentation Room: CCC, 105

Back to Session