Accurate prediction of permeability in geothermal reservoirs remains a significant challenge for energy extraction and reservoir management, as conventional laboratory correlations often fail to capture the complex, nonlinear interactions among porosity, stress, and fluid properties under high-temperature, high-salinity conditions. To address these limitations, researchers developed an enhanced data-driven framework that integrates Nuclear Magnetic Resonance (NMR) laboratory measurements with machine-learning algorithms. By utilizing an expanded dataset of 204 core samples, the study aimed to improve the precision of permeability modeling in environments where traditional methods frequently fall short.
The research team trained and validated six distinct machine-learning models: Decision Tree, AdaBoost, Random Forest, K-Nearest Neighbors, Multilayer Perceptron, and Linear Regression. The input variables selected for the modeling process included effective porosity, total porosity, effective stress, salinity, and NMR relaxation parameters, specifically Tâ‚‚lm and Tâ‚‚mode. While the team employed the Leverage technique to conduct outlier detection, they ultimately retained all 204 samples, including those identified as high-leverage, to ensure the framework preserved the natural geological variability inherent in the dataset.
The performance of each model was rigorously evaluated using K-fold cross-validation and standard statistical metrics, including R², root mean square error (RMSE), and absolute average relative error (AARE). The results indicated that the Random Forest model achieved the strongest generalization on unseen data, recording an R² of 0.968, an RMSE of 52.414, and an AARE of 62.046%.
A SHAP-based sensitivity analysis was conducted to determine the relative influence of each feature on the predictive output. The findings confirmed that effective porosity was the most influential predictor of permeability, followed by total porosity and effective stress. While salinity and NMR relaxation parameters also contributed to the model, their impact was moderate compared to porosity metrics. By bridging experimental NMR characterization with advanced predictive modeling, this proposed workflow offers a physically consistent and computationally efficient alternative to traditional laboratory methods, providing a scalable solution for evaluating geothermal reservoir potential in challenging subsurface environments.
Subject of Research: Data-driven predictive modeling for geothermal reservoir permeability using NMR-derived features.
Article Title: Data‑driven characterization of geothermal reservoir permeability using NMR‑derived features
Article References: Yadav, K., Alshudukhi, J., Quadri, N. S., Dutta, D. K., & Mottaghi, M. (2026). Data‑driven characterization of geothermal reservoir permeability using NMR‑derived features. Geothermal Energy. https://doi.org/10.1186/s40517-026-00420-6
Image Credits: AI Generated
DOI: 10.1186/s40517-026-00420-6
Keywords: Geothermal Energy, Reservoir Management, Nuclear Magnetic Resonance, NMR, Permeability Prediction, Machine Learning, Random Forest, Data-Driven Modeling, Porosity, Effective Stress, Salinity, Geothermal Reservoirs, Core Sample Analysis, Predictive Modeling, Energy Extraction, Statistical Analysis, SHAP Analysis, Subsurface Characterization, Fluid Properties, Petrophysics
News Source: Violet Maxwell. (November 17, 2026). Random Forest Model Improves Geothermal Reservoir Permeability Predictions Using NMR Data. Scienmag.



