New Mexico Geological Society Annual Spring Meeting — Abstracts

Geothermal energy has the potential to play an essential role in the global transition to renewable energy sources. Geodynamic processes are fundamentally driven by Earth's internal heat, with surface heat flow and geothermal gradients serving as critical indicators of subsurface thermal regimes. A significant portion of this heat originates from the radioactive decay of unstable isotopes primarily U²³⁸, Th²³², and K⁴⁰ which release energy via alpha, beta, and gamma particles. The Natural Gamma Spectrometer (NGS) tool, which measures Uranium (URAN), Thorium (THOR), and Potassium (POTA) abundances, remains the benchmark for estimating heat production. This study proposes a novel integration of Machine learning (ML) data-driven modeling and Empirical Experimental Correlations to enhance prediction of heat generation. Machine Learning Algorithms: ExtraTrees Regressor, MLP Regressor, Gradient Boosting Regrssor and Random Forest Regressor are trained and tested with well log data as inputs and Natural Gamma Spectrometer heat generation as outputs. The Regression algorithms performance are evaluated and compared the experimental empirical correlation for analysis. Our approach addresses gaps in geothermal resource assessment by enhancing heat flow models through improved heat generation estimates.