Leveraging Machine Learning to Predict Geothermal Gradients Across Colombia

Core Concepts

A machine learning model accurately predicts the spatial distribution of geothermal gradients in Colombia using global-scale geological and geophysical datasets.

Abstract

The study presents a Gradient Boosted Regression Tree (GBRT) algorithm to predict the spatial distribution of geothermal gradients across Colombia. The model leverages a diverse array of geological and geophysical datasets, including topography, subsurface characteristics, geophysical anomalies, fault data, and proximity to basement rocks, to estimate the geothermal gradient.

The model was extensively validated, achieving a normalized root mean square error (nRMSE) of 0.12 and an R-squared value of 0.52 on the test set. This performance is comparable to or better than previous studies in other regions, demonstrating the model's accuracy and robustness.

The feature importance analysis revealed that elevation, Moho depth, and basement proximity are the most influential predictors of the geothermal gradient. The model's predictions align well with known regional trends and provide valuable insights into unexplored areas, such as the Amazon basin, where the model predicts high geothermal gradients.

The resulting geothermal gradient map highlights regions with significant geothermal potential, guiding future exploration efforts and supporting the development of Colombia's renewable energy resources. The study showcases the potential of machine learning techniques to enhance geothermal exploration and contribute to the sustainable energy transition.

Stats

The depth to the Mohorovičić discontinuity (Moho) is inversely correlated with the geothermal gradient.
Regions with higher free air and Bouguer gravity anomalies tend to exhibit higher geothermal gradients.
The proximity to active faults is associated with increased geothermal gradients due to potential fluid flow and permeability.

Quotes

"The model's output indicates a distinct sensitivity to geothermal indicators, revealing high-gradient values that are suggestive of underlying geothermal resources."
"The regions with the highest predicted gradients warrant immediate attention for exploration and data acquisition. These areas, flagged by the model, are likely candidates for sustainable geothermal development."

Key Insights Distilled From

Predicting the Geothermal Gradient in Colombia

by Juan... at arxiv.org 04-09-2024

https://arxiv.org/pdf/2404.05184.pdf

Deeper Inquiries

How can the model's predictive accuracy be further improved by incorporating more detailed regional geological and geophysical data

To enhance the model's predictive accuracy, incorporating more detailed regional geological and geophysical data is crucial. By integrating data such as lithospheric thickness, rock type variations, and specific heat flow measurements, the model can capture more nuanced relationships between these factors and geothermal gradients. Detailed information on fault distribution, subsurface structures, and local geothermal anomalies can provide valuable insights into the thermal behavior of the region. Additionally, incorporating data on hydrothermal systems, seismic activity, and volcanic features can further refine the model's predictions. By expanding the dataset to include these detailed regional data points, the model can better capture the complex interplay of geological and geophysical factors influencing geothermal gradients.

What are the potential limitations or biases in the training data that may have influenced the model's performance, and how can these be addressed

The training data may have limitations or biases that could have influenced the model's performance. One potential limitation is the uneven distribution of geothermal gradient measurements, with certain regions being overrepresented while others are underrepresented. This imbalance can lead to a skewed model that may struggle to accurately predict geothermal gradients in less sampled areas. Addressing this bias involves implementing weighting strategies to account for the underrepresented regions and ensuring a more balanced dataset. Additionally, the training data may lack information on specific geological features or subsurface properties that could significantly impact geothermal gradients. To mitigate this limitation, incorporating more diverse and detailed data on lithospheric structures, fault systems, and rock compositions can help capture a more comprehensive picture of the subsurface conditions. Conducting thorough data validation and verification processes can also help identify and correct any inconsistencies or errors in the training data, improving the model's overall performance.

Given the model's ability to identify high-potential geothermal regions, how can these insights be leveraged to support the development of Colombia's renewable energy infrastructure and contribute to the country's decarbonization efforts

The insights provided by the model on high-potential geothermal regions can be leveraged to support the development of Colombia's renewable energy infrastructure and decarbonization efforts in several ways. Firstly, these identified regions can serve as strategic locations for geothermal exploration and development projects, enabling the country to harness its geothermal resources efficiently. By focusing resources on these high-potential areas, Colombia can accelerate the deployment of geothermal energy technologies, reducing its reliance on fossil fuels and advancing towards a more sustainable energy mix. Furthermore, the model's predictions can guide policymakers and energy stakeholders in making informed decisions on investment priorities, regulatory frameworks, and infrastructure planning for geothermal energy projects. By aligning development efforts with the model's insights, Colombia can optimize its renewable energy transition, reduce greenhouse gas emissions, and contribute to global efforts to combat climate change.