Answers are grounded in this paper and cite their sources.
Accurate prediction of permeability in geothermal reservoirs is essential for optimizing energy extraction and reservoir management. Conventional laboratory correlations often fail to capture the nonlinear interactions among porosity, stress, and fluid properties under high-temperature, high-salinity conditions. This study presents an enhanced data-driven framework that integrates Nuclear Magnetic Resonance (NMR) laboratory measurements with machine-learning algorithms to model permeability using an expanded geothermal dataset of 204 core samples. Effective porosity, total porosity, effective stress, salinity, T₂lm, and T₂mode were used to train and validate six models: Decision Tree, AdaBoost, Random Forest, K-Nearest Neighbors (KNN), Multilayer Perceptron (MLP-ANN), and Linear Regression. Outlier detection was performed using the Leverage technique; however, all samples including high-leverage points, were retained to preserve the natural geological variability of the dataset. Model performance was evaluated using K-fold cross-validation and statistical metrics (R 2 , RMSE, and AARE). Random Forest achieved the strongest generalization (R 2 = 0.968, RMSE = 52.414, AARE = 62.046%), demonstrating superior stability on unseen data. SHAP-based sensitivity analysis confirmed effective porosity as the most influential predictor, followed by total porosity and effective stress, while salinity and NMR relaxation parameters contributed moderately. The proposed workflow provides a physically consistent and computationally efficient approach for permeability estimation, bridging experimental NMR characterization with predictive modeling and offering a scalable alternative to traditional laboratory methods for geothermal reservoir evaluation.
No takes yet. Share an insight, caveat, or question.
Yadav et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: