In the rapidly evolving domain of geological sciences and environmental engineering, the fusion of artificial intelligence with traditional fieldwork is redefining how researchers analyze complex geological formations. A recent groundbreaking study from Chinese scientists presents a sophisticated synergy between human expertise and machine learning algorithms to revolutionize the interpretation of rock discontinuities—a critical factor for the safe storage of high-level radioactive waste (HLW). Published in the prestigious journal Environmental Earth Sciences, this research demonstrates how photogrammetric techniques, combined with human-machine interactive processes, can produce refined, high-fidelity identification of intricate discontinuity networks within rock masses at a candidate HLW repository site in China.
Accurate identification and mapping of rock discontinuities, such as fractures, joints, and faults, are crucial for assessing the mechanical stability and hydrogeological behavior of underground repositories intended for HLW. These discontinuities influence not only the structural integrity of the surrounding rock but also pathways for potential radionuclide migration, thereby directly affecting the long-term safety evaluations required by regulatory bodies worldwide. Traditionally, geologists rely on manual field surveys to classify and measure these features—a process both time-consuming and prone to observational bias. This new innovative approach leverages photogrammetry, which uses overlapping photographic images to construct precise 3D models of rock faces, providing rich spatial data that can be meticulously analyzed by both human experts and AI systems.
The integration of human-machine interactive methods distinguishes this research from previous attempts that relied heavily on either manual mapping or automated detection alone. Human intuition and geological expertise remain intrinsic to the interpretation process, guiding machine learning models to improve accuracy in complex scenarios where purely automated systems might falter. In this hybrid workflow, initial photogrammetric data acquisition is followed by AI-assisted processing tools trained to recognize textural and geometric patterns typical of various discontinuity types. Operators then validate and refine these outputs through iterative feedback loops, correcting misclassifications and enhancing the model’s learning, which systematically elevates overall detection performance.
Employing a multi-scale photogrammetric approach, the researchers were able to capture discontinuities at both macro and micro levels, affording unprecedented resolution and contextual understanding of spatial relationships. This fine-grained data enabled not only the identification of visible fracture networks but also characterization of subtle features such as surface roughness, aperture variability, and connectivity—parameters integral to hydrogeological modeling and geomechanical simulations. The methodology detailed in the study can serve as a blueprint for future investigations in other geologically complex settings beyond HLW sites, including mining operations, earthquake fault zones, and tunnel construction projects.
One of the most compelling aspects of the research lies in its application to a real-world HLW repository candidate site in China—a country embarking on a robust nuclear energy expansion and, consequently, confronting the challenges of safe radioactive waste disposal. The repository site presents an environment with intricate geological heterogeneity, where existing approaches were insufficiently precise for regulatory demands. Through the combined photogrammetric and human-machine technique, the team achieved a high-confidence mapping of rock discontinuities that can better inform risk assessments for containment breaches and groundwater contamination over the repository’s projected lifespan.
The research also underscores the role of technological democratization in geological sciences. By developing user-friendly interfaces that facilitate seamless communication between human operators and algorithmic systems, the study lowers barriers to adopting advanced analytical tools in field geology. Such interfaces empower even less-experienced personnel to contribute effectively by capitalizing on machine intelligence while maintaining expert oversight, thereby accelerating data processing timelines and improving data quality. This democratization effect holds great promise for scaling similar methodologies globally, especially in regions where expert geological resources are scarce.
From a technical standpoint, the photogrammetric techniques employed involved the use of high-resolution digital cameras mounted on drones and terrestrial platforms, enabling comprehensive coverage of otherwise inaccessible rock surfaces. The imaging datasets were processed using state-of-the-art Structure-from-Motion (SfM) algorithms to generate dense point clouds and textured 3D meshes. These digital twins of the geological outcrops formed the basis for subsequent AI-driven feature extraction. Convolutional neural networks (CNNs) were adapted to classify discontinuities by learning from manually annotated training datasets. The human-machine interactive system incorporated active learning strategies, wherein the AI system solicits input selectively from human experts on ambiguous regions, continually refining its classification accuracy.
Notably, the study highlights the significance of contextual geological knowledge embedded within the interactive process. Unlike fully automated systems that treat each image or segment in isolation, the human-machine loop allows for the incorporation of regional geological history, tectonic stress orientation, and lithological variations. This comprehensive approach reduces false positives and negatives in discontinuity detection, yielding a dataset that is not only geometrically sound but also geologically meaningful. This depth of insight is essential for subsequent engineering analyses that dictate repository design, material selection, and operational protocols aimed at ensuring long-term isolation of radioactive waste.
The environmental implications of this research are profound. Robust characterization of rock discontinuities enables improved predictive modeling of fluid flow and contaminant transport through fractured media, a long-standing challenge in hydrogeology. By refining these models, decision-makers can better evaluate scenarios of radionuclide leakage and consequent ecosystem impacts under various geological and climate-driven perturbations. This enhanced predictive capability supports both the safeguarding of human populations and the preservation of natural resources over millennia—aligning with the stringent sustainability goals intrinsic to nuclear waste management frameworks.
Furthermore, this innovative approach sets a precedent for interdisciplinary collaboration among geologists, computer scientists, environmental engineers, and policymakers. The convergence of domain-specific knowledge and AI technology in the study exemplifies how complex societal challenges like radioactive waste disposal can benefit from integrated problem-solving frameworks. The researchers advocate for continued development of human-machine systems that evolve dynamically through field applications, user feedback, and advances in machine learning algorithms, thereby remaining adaptable to diverse geological conditions and evolving societal standards.
The prospective scalability of this human-machine interactive photogrammetric method also opens avenues for its application in other geotechnical projects. For instance, mining companies could deploy similar techniques to rapidly evaluate rock stability and orebody continuity, while civil engineering projects might use them to assess landslide risks or optimize tunnel alignments. By extending this pipeline to real-time monitoring scenarios, stakeholders could gain timely insights into structural deformations or emergent fracture formation, facilitating proactive mitigation measures.
In conclusion, this seminal study epitomizes how cutting-edge technology, when thoughtfully integrated with indispensable human expertise, can surmount traditional limitations in geological investigations. The human-machine interactive refined identification of complex rock discontinuities via photogrammetric imaging represents a monumental leap forward in ensuring the safety and environmental stewardship of critical infrastructure such as HLW repositories. As nuclear energy remains a pivotal component of global energy strategies, research such as this equips societies with the tools necessary to responsibly manage associated risks while advancing scientific understanding of the Earth’s complex subsurface architecture.
Subject of Research: Rock discontinuity identification and characterization using human-machine interactive photogrammetric techniques at a high-level radioactive waste repository site.
Article Title: Human-machine interactive refined identification of complex rock discontinuities using photogrammetric techniques: Case studies from a candidate HLW repository site in China.
Article References:
Yang, Y., Xu, W., Li, X. et al. Human-machine interactive refined identification of complex rock discontinuities using photogrammetric techniques: Case studies from a candidate HLW repository site in China. Environ Earth Sci 84, 277 (2025). https://doi.org/10.1007/s12665-025-12287-0
Image Credits: AI Generated
