This meeting will bring together leading researchers in Geosciences, Digital Twins, Optimization, data, uncertainty quantification and Modeling, with the goal of identifying key scientific challenges and future directions in the field of practical Digital Twins for Predictive Geosciences, with a special focus on applications in geosciences (e.g., environmental risks, hydrogeology, geothermal).

The meeting will consist of several small discussion sessions, where participants will have the opportunity to share their expertise, exchange ideas, form new partnerships, and contribute to the establishment of digital twin guidelines and new research directions on the topic.

Climate change has accelerated ambitious programs on transitional energy, like the Green Deal (from European commission) or the Sustainable Energy For All initiative (from United Nations), which aim to develop an efficient, low-carbon and environmentally friendly resource. Within this context, energy transition has a highly significant role to play. The deployment of new energy sources requires numerical tools to help the management of resources, risk assessment, and supporting operational decision making processes.

Whether we want to anticipate tomorrow’s weather, estimate the water supply, wind production, geothermal resources, monitor a CO2 site sequestration, or assess the seismicity risks, it is necessary to obtain an accurate understanding of existing conditions such as underground resources, fluid flows, geomechanics, geochemistry, geophysics, or geology. To achieve this understanding, we need a tool that captures the state of the system at a given moment: namely, a Digital Twin (DT).

Digital Twins are computational models that represent a physical entity. This concept involves creating a virtual replica of a physical object, system, or set of processes. Through the integration of sensors, real-time data, and advanced simulation models, Digital Twins enable the automatic monitoring and continuous analysis of system states. They provide powerful opportunities for optimizing operations, analyzing risks, forecasting failures and promoting sustainable resource management.

The development of Digital Twins is a complex undertaking that spans diverse fields from academy to industry. It requires a high level of domain expertise, as well as tools such as Artificial Intelligence, Optimization, and High-Performance Computing. This Shonan Workshop will convene leading experts across diverse disciplines to examine the advancements and applications of Digital Twin (DT) technology in geosciences, with applications to the Energy Transition and environmental risk mitigation. Currently, there is a critical gap in unified guidelines and standards for DT design, impeding the technology's deployment on a societal scale. Our goal is to bring together the research community to forge a common vision and identify priority research directions, laying the groundwork for DT innovations that address pressing environmental and energy challenges.

The specific scientific topics to be discussed during the Shonan Workshop include, but are not limited to:

• Metamodeling: Creating and using Digital Twins is a costly process. It is necessary to find ways to make it shorter without losing too much precision. To perform this trade-off, a low-fidelity model, which simplifies some key physical processes, is created from the high-fidelity DT. Several techniques have been researched to create such low-fidelity models (also called proxy, surrogate, meta models or emulators), such as Artificial Intelligence, Regression Models, Statistical proxies, Proper Orthogonal Decomposition, and Reduced Order Models.

• Large Scale Computing: Even though metamodelling can reduce the cost of using DTs, their design and fine tuning still requires a large amount of computational resources. In this sense, the development of more efficient computational techniques, from parallel computing to dedicated hardware for simulations, as well as energy efficient computing, is a key challenge for the future development and deployment of DTs.

• Model Optimization: Digital Twins correspond to complex models of physical systems, in specific instances in our world (such as the model of a specific ocean or subsurface area). As such, the deployment of a DT requires matching from dozens to thousands of parameters that specify how a set of physical rules correspond to a specific instance of a physical system. To find the parameters that match the output of a DT to the real data of a physical instance is a complex inverse optimization task, including problems such as: Non-linear optimization, Constrained Optimization, Multi-Task Optimization, etc. Handling this task requires better tools from fields such as Bayesian Optimization, Meta-Heuristic Optimization, and Numerical Optimization.

• Uncertainty Quantification: Even if the highest effort is spent in the choice of physical models and parameter optimization, Digital Twins still have to handle the challenge of Uncertainty. The challenge of uncertainty lies in the need to provide accurate and reliable predictions despite often fragmented, imperfect, or hard-to-interpret data. This is a particularly critical issue in subsurface applications, such as Geothermal and Seismic, where direct access to the data is limited. Uncertainty can come from incomplete, noisy or low resolution data sources, Interpretation biases from human operators, simplifications of the physical models, numerical errors and or computational uncertainty. The challenge of Uncertainty Quantification is to detect and estimate the effect of these different sources of uncertainty and, critically, describe the effect of the uncertainty in the output and interpretations of the Digital Twin.

• Data Integration: The data necessary for the development of Digital Twins come from various sources, such as geophysical surveys, geochemistry, geological models, remote sensing and sensor networks, among others. Each source has different formats, resolutions and uncertainties. Data Integration is the challenge of harmonizing these data types and integrating them with the computational model of the DT, which amounts to a complex and difficult inverse problem. Moreover, identifying the gap between what the data and the models represent and how to measure it is not a trivial task. Solving this challenge opens all sorts of questions in the data integration workflow, such as the metrics used, the objective function for the optimization, how to interpret and use the optimization results, and final decision-making tools. Achieving seamless data integration is essential to ensure that digital twins can reliably represent and predict subsurface behaviors.
• FAIR data is essential for developing a digital twin because it ensures that the data feeding the model is Findable, Accessible, Interoperable, and Reusable. Digital twins rely on continuous, reliable data flows to accurately represent and simulate the real-world system they mirror. When data follows FAIR principles, it becomes easier to integrate from multiple sources, share across teams or platforms, and reuse in different modelling contexts. This improves the consistency, transparency, and scalability of digital twins, while reducing errors, data loss, and duplicated effort. In short, FAIR data provides the solid, structured foundation that a trustworthy and efficient digital twin requires.

The objective of this Shonan meeting is to bring together practitioners of Digital Twins with different expertises and points of views: Domain Specialists, Optimization Specialists, and Industry researchers, to generate fresh ideas and create new collaborations in the field, in order to structure and draft some clear guidelines for the establishment of DT technologies in Geosciences.

The meeting itself will be organized as follows, with room for flexibility following the result of discussions:

• First day - Problem statement and introductions: On the first day, the organizers will present the main topics and the related sub-topics of the workshop for all participants, and the participants will briefly present their specialities and interests. A discussion session about the key challenges regarding energy transition will close the day.
• Second day - Challenges identification - Participants will be divided into smaller groups, each focusing on a specific topic, such as Model Optimization, Uncertainty Quantification, Data, Case Studies. Associated challenges will be discussed. Throughout the day, participants will rotate and contribute to the discussions in other groups to foster collaboration and cross-pollination of ideas.
• Third day - Narrowing and defining key challenges- Discussion and Panel Rounds: Each group will summarize their discussions from the previous day and present their findings to all participants. The organizers will create small, focused discussion groups based on the challenges and questions of the second day. In these discussions the participants will clarify some main challenges which should be addressed and research. A global Digital Twin coherence to connect the different topics will be carefully watched by the organisers.
• Fourth Day – Identification of Actions and Guidelines: Research Roadmap and Joint Work Proposals Representatives from each discussion group will present a summary of their key findings to all participants, highlighting the priority research areas (research questions) identified during the discussions. This input will serve as the foundation for developing a comprehensive research roadmap for the future of Digital Twins in Geosciences. The roadmap will guide future research efforts, including the design of collaborative research projects and special invited issues, ensuring a coordinated and impactful approach moving forward.