Scientific machine learning (SciML) and artificial intelligence (AI) are increasingly used to interpret heterogeneous subsurface data, accelerate multiphysics simulation, infer uncertain parameters, and support decisions in geotechnical, environmental, and geoenergy applications. Their practical value, however, depends on physical consistency, generalisation beyond training conditions, uncertainty characterisation, and validation against independent evidence.
This session invites contributions on recent methodological advances and innovative applications of data-driven, physics-informed, and hybrid physics-AI techniques for geotechnical, geothermal, and subsurface systems. Relevant applications include, but not limited to, site and subsurface characterisation; geothermal exploration, reservoir development, monitoring, and operation; groundwater flow, transport in porous media, multicomponent reactive transport, and geochemical processes; coupled thermo-hydro-mechanical-chemical processes; underground storage of heat, hydrogen, or carbon dioxide; and geotechnical infrastructure.
We welcome work on supervised and unsupervised learning, deep learning, physics-informed neural networks, neural operators, surrogate and reduced-order models, inverse modelling, data assimilation, Bayesian inference, digital twins, and other relevant studies. Contributions addressing heterogeneous or sparse data, multiscale modelling, uncertainty quantification, sensitivity analysis, model discrepancy, transferability, interpretability, and computational efficiency are particularly encouraged.
A central objective is to connect methodological development with physical and observational evidence. We therefore welcome studies integrating laboratory experiments, imaging and sensing methods, field monitoring, operational data, and numerical simulations for training, calibration, verification, and validation. Comparative studies that establish when machine learning or physics-informed methods improve upon conventional numerical, statistical, or inverse approaches and where they do not are especially valuable.