Civil Engineering and Environmental Systems

Advancing a Systems‑Level Perspective in Geotechnics: Uses of Artificial Intelligence and Machine Learning

The special issue “Advancing a Systems‑Level Perspective in Geotechnics: Uses of Artificial Intelligence and Machine Learning” invites the geotechnical community to rethink the role of data‑driven methods in engineering practice. While early applications of machine learning in geotechnics focused on outperforming empirical or regression‑based models, the field now faces a more fundamental challenge.

A generalised ML/AI model capable of performing well across diverse geological conditions and datasets reduces dependence on individual case‑specific models and enables ML/AI to function as a system‑level component rather than a project‑bound tool, leading to better decisions.

Therefore, this special issue aims to shift the conversation from prediction alone toward physics‑informed ML, hybrid modelling, and the integration of constitutive laws, and probabilistic structures into modern workflows. By embedding physical reasoning directly into ML architectures, the goal is to develop models that are not only accurate but also mechanically consistent, interpretable, and robust across diverse geotechnical conditions.

In addition, the issue emphasises the importance of combining non–data‑informed and data‑informed approaches to improve model generalization for decision-making. This includes integrating ML with well‑established empirical equations, and constitutive models, as well as embedding probabilistic and statistical structures directly into ML penalty algorithms. In geotechnics, where heterogeneity, stress dependency, and drainage conditions vary widely, such hybrid approaches offer opportunities to design adaptive workflows in which non‑data‑informed components adjust data‑informed ones. Feature engineering, feature importance, and feature selection play an important role in this shift from case dependency to system-level approach, enabling the creation of physically meaningful features, reducing dimensionality, and keeping geotechnical interpretation at the forefront rather than hidden within a black‑box model.

At the same time, the issue highlights emerging opportunities in multimodal learning, where geological logs, laboratory tests, satellite imagery, geophysical surveys, numerical simulations, and textual site reports can be fused into unified representations of subsurface behaviour. Complementing this, the integration of uncertainty quantification and probabilistic prediction enables models to express both aleatory and epistemic uncertainty, supporting risk‑aware engineering decisions. By bringing these elements together, this special issue aims to take a step toward a more holistic, system‑level view of geotechnics, where data, physics, uncertainty, interpretation, and engineering judgment are treated as interconnected components rather than isolated tasks. This system‑oriented perspective encourages new methodological based on ML/AI developments that reflect the complexity of geotechnical environments and the interdependence of their governing processes.

Submissions from soil and rock mechanics, foundations, offshore geotechnics, slopes, and tunnelling are welcome, including methods, applications, reviews, thought pieces, and benchmark studies. The editors particularly encourage contributions that explore hybrid modelling, multimodal frameworks, uncertainty quantification, and physics‑aligned ML approaches that contribute to more integrated and holistic decisions relevant to geotechnical problems.

Papers we consider in scope will have one or more of the following characteristics:

• Physics‑informed ML models integrating constitutive laws or governing equations
• Hybrid frameworks combining data‑driven and non‑data‑informed methods
• Studies evaluating model generalisation on unseen or cross‑site datasets
• Feature engineering, feature importance, or feature selection for better problem-solving.
• ML and AI approaches to evaluating the value of collecting additional data.
• Uncertainty quantification and probabilistic ML for risk‑aware decisions
• System‑level or interdisciplinary approaches linking data, physics, and engineering judgment
• Reproducible workflows, or standardized evaluation protocols
• Simplifying models using penalty/loss-function algorithms informed by physics, empirical knowledge and well-established basics of geotechnical engineering

Examples of topics within this scope are:

• Physics‑informed neural networks for constitutive modelling
• Hybrid ML–numerical workflows for settlement or stability prediction
• Cross‑site generalisation studies for liquefaction or rate of penetration models
• Feature‑engineered stress‑ or drainage‑dependent soil descriptors
• Probabilistic ML models with quantified aleatory and epistemic uncertainty
• Goal-oriented multimodal fusion (e.g., field logs, lab tests, remotely sensed data, qualitative judgment)
• Benchmark datasets for geotechnical ML model comparison
• Uncertainty‑aware ML for slope stability or foundation design
• System‑level modelling frameworks integrating physics, data, and uncertainty

The journal’s emphasis on systems and their analysis means that this Special Issue welcomes manuscripts that contribute to a broader, system‑level understanding of geotechnical engineering. Consequently, the application of AI/ML without advancing system‑level insight, methodology, or goals, falls outside the scope of this issue.