Machine Learning for Integrating Geophysical and Geotechnical Data in Subsurface Characterization

Subsurface characterization is fundamental to geotechnical engineering and infrastructure development, but direct investigations are often costly, time-consuming, and spatially limited. Geophysical methods, such as seismic reflection, seismic refraction tomography, multichannel analysis of surface waves, and electrical resistivity tomography, provide efficient and spatially extensive measurements but do not directly measure the needed geotechnical parameters or lithological units. Machine learning offers a means of linking geophysical responses with geotechnical data and generating spatially continuous estimates of subsurface properties, such as Cone Penetration Test profiles, Standard Penetration Test values, seismic velocities, strength, stiffness, and density-related parameters. Recent studies have demonstrated the feasibility of predicting geotechnical parameters from seismic reflection data, including applications in offshore wind farm site characterization. However, evidence remains concentrated in selected investigation environments and application domains, while applications to onshore infrastructure projects are often constrained by sparse borehole and in-situ testing data. Key challenges include improving prediction accuracy, improving the reliability and sharpness of predictive uncertainty estimates, evaluating prediction suitability for specific engineering purposes, and demonstrating the practical applications of the predicted parameters. Further investigation is therefore needed to establish robust, transparent, and engineering-relevant workflows for integrating geophysical and geotechnical data and assessing the reliability of the resulting predictions.This Research Topic aims to advance machine learning methods for integrating geophysical and geotechnical data to improve the characterization, interpretation, and prediction of subsurface conditions. It will examine how different learning strategies and approaches, data sources, and validation frameworks can support reliable predictions across diverse applications in both onshore and offshore. Contributions may address the development and benchmarking of new methods, the evaluation of existing approaches, or their implementation in practical site investigation, design, and decision-making workflows. Particular emphasis is placed on determining when predictions are sufficiently reliable and fit for purpose, quantifying and communicating uncertainty, and assessing the implications of model outputs for engineering practice.To gather further insights into machine learning-based integration of geophysical and geotechnical data for subsurface characterization, including its application to site investigation, infrastructure planning, engineering design, and risk-informed decision-making, we welcome articles addressing, but not limited to, the following themes: • Prediction of CPT profiles, including cone resistance, sleeve friction, and pore-water pressure, from geophysical data using machine learning• Prediction of SPT N-values, soil behaviour types, lithological units, soft layers, and weak zones using geophysical data and machine learning• Estimation of P-wave velocity (Vp), S-wave velocity (Vs), porosity, small-strain shear modulus, strength, stiffness, and density-related properties using geophysical data and machine learning• Integration of seismic reflection, seismic refraction tomography, MASW, ERT, or other geophysical datasets with geotechnical data for machine learning-based predictions• End-to-end workflows for data integration, preprocessing, feature engineering, and model development• Comparative assessment of machine learning algorithms and implementation strategies, including sensitivity to input features, preprocessing methods, hyperparameters, and training data• Interpretability and explainability of machine learning predictions• Validation and reliability assessment of model predictions, including blind testing using independent held-out data, and assessment of geological plausibility and spatial consistency in predicted two-dimensional subsurface sections• Quantification, propagation, and communication of prediction uncertainty• Fit-for-purpose evaluation in relation to engineering tolerances, consequences of error, and decision contexts• Comparison of machine learning-based and conventional site-characterization approaches• Applications of machine learning-based predictions to site selection, infrastructure planning, preliminary site assessment, foundation siting, and engineering design• Risk assessment and evaluation of the implications of predicted parameters for engineering decisions• Lessons learned, limitations, best practices, review studies, and perspective papersAppendix: Relevant contribution types include exploratory studies, case studies, methodological developments, practical implementation studies, comparative studies, sensitivity analyses, studies comparing machine learning with conventional characterization methods, review papers, and perspective papers. Topic editors: Dongfang Qu, Haibin Di, Yuting Zhang Source: https://www.frontiersin.org/research-topics/85736/machine-learning-for-integrating-geophysical-and-geotechnical-data-in-subsurface-characterization