Management of the monitoring data acquisition uncertainties within deep geological repositories

Digital twins and 3D models can be used to enhance the understanding of repository evolution, particularly in the context of monitoring data management, modelling, and visualisation. These tools can be used to identify and mitigate risks, ensuring the repository's long-term safety and minimizing potential hazards as well as enhance stakeholder trust and engagement by providing transparent and accessible information about the repository's status and safety measures.​

The Monitoring Equipment and Data Treatment for Safe Repository Operation and Staged Closure (MODATS) work package of the European Joint Programme on Radioactive Waste Management (EURAD) focused on the monitoring during the operational phase of repository programmes to gain understanding on the processes affecting the barrier and host rock performance and evolution, and to build further confidence in the long-term safety case. Work was based on the six deep geological disposal test cases from Europe. The project has proposed a workflow for data handling from acquisition to decision support. Some of these experiments have been running for decades and within this time frame data science has evolved widely providing many new methods and tools.

Modelling, simulations and various monitoring campaigns provide increasing amount of data which can be utilised throughout the entire lifecycle of the deep geological disposal facility including the siting, pre-operational, operational and post-closure periods. Monitoring data can be used to visualise the operations and improve the reliability of the models, simulations and digital twins.

Within the MODATS project a transparent and flexible data curation pipeline has been defined to support the overall data management process. This process is an important tool to manage and reduce monitoring data uncertainties. This paper will present the concept.

Reliable data treatment and analysis require extensive additional information beyond the acquired sensor values from one or a few sensors. Datasets from multiple sensors need to be treated and analysed together, incorporating cross-checking and adding information about known events. Applying domain knowledge of the entire environment and the used sensors and data acquisition setup is crucial to ensure correct interpretations. Over a long timeframe, the amount of data accumulates, and data management concepts evolve, making it possible to revisit the old datasets and find new interdependencies between unforeseen phenomena.

Use of Artificial Intelligence (AI) is rapidly increasing and bringing new tools to manage acquired data and build and improve models. AI requires extensive data cleaning to avoid outcomes reflecting sensor problems. Each AI result is subject to data pre-treatment. Defining Digital Twins (DT) involves determining its scope, purpose, and whether it is used for "as is representation" or predicting future behaviour of an entity, as well as identifying the target group of users. There is no one-size-fits-all solution, but there need to be definitions for specific purpose DTs.

Connecting and opening complex data and model systems and platforms to the DT reliably and securely is essential. Interoperability of various actors and systems, as well as the creation of common definitions and standards, must be addressed. Confidentiality and information security versus openness need also to be considered.