Integrating multi-modal data is central to the digital transformation in infrastructure health monitoring. Non-destructive testing (NDT) methods such as Ground Penetrating Radar (GPR) and Magnetometry are often combined to predict the presence, depth, and ferrous characteristics of subsurface structures. Despite advancements in NDT data acquisition, processing, and evaluation, interpretation still largely depends on expert judgment or rule-based approaches, which can introduce subjectivity and inconsistencies and limit repeatability.
Although visualisation techniques are improving, many systems still rely on simplified two-dimensional representations, such as radargrams, to portray complex subsurface features, making interpretation difficult. While three-dimensional visualisation methods and immersive extended reality (XR) technologies have been suggested to improve spatial understanding and lessen cognitive load, their adoption in industry remains limited due to challenges with integration and validation.
A recent study identified technical challenges in immersive visualisation, including frame-rate instability and high GPU usage in a VR prototype using GPR and magnetometry data. However, technical performance alone does not determine success; stakeholder perspectives are essential to ensure XR outputs align with the analytical needs and decision-making processes of geophysical professionals. Furthermore, stakeholder-centred development can enhance usability, practicality, and relevance, supporting wider adoption of such technologies.
This study extends the prototype to conduct preliminary user testing with six expert geophysics stakeholders. Thematically analysed feedback, gathered through semi-structured interviews, revealed the following: (1) Dynamic Data Interrogation: the need for interactable colour maps; (2) Interactive Spatial Querying: selecting sectors to access metadata; (3) User-centric Design: ensuring digital outputs match field-expert mental models; and (4) Integrated Data Cataloguing: enabling the selective management and visualisation of multi-sensor datasets.
Acknowledgements: The authors acknowledge the funding support from the University of West London and contributions from ITINERIS, GAIA iLAB, and the Faringdon Research Centre. We also appreciate a visiting scholarly exchange between UWL and CNR-IREA that supported the study.
