Predicting the geothermal gradient in Colombia: A machine learning approach

Abstract

Accurately determining the geothermal gradient is crucial for assessing geothermal energy potential. In Colombia, despite an abundance of theoretical geothermal resources, large regions of the country lack gradient measurements. This study introduces a machine learning approach to estimate the geothermal gradient in regions where only global-scale geophysical datasets and course geological knowledge are available. We find that a Gradient-Boosted Regression Tree algorithm yields optimal predictions and extensively validates the trained model, obtaining predictions of our model within 12% accuracy. Finally, we present a geothermal gradient map of Colombia that serve as an indicator of potential regions for further exploration and data collection. This map displays gradient values ranging from 16.75 to 41.20 °C/km and shows significant agreement with geological indicators of geothermal activity, such as faults and thermal manifestations. Additionally, our results are consistent with independent findings from other researchers in specific regions, which supports the reliability of our approach. • Machine learning model predicts geothermal gradient using global geophysical datasets. • Predictions match independent measurements with high accuracy. • New geothermal gradient map highlights Colombia’s unexplored regions. • Predictions follow regional trends from prior geological knowledge. • High gradients confirmed in Amazon, previously unexplored, providing a study baseline.

Topics & Concepts