Engineered Geosystems

Humankind is increasingly using the earth’s subsurface as a resource and a repository. Clear water reservoirs require the highest protection level as critical resources; nuclear waste and greenhouse gases must be stored safely for centuries; and the subsurface storage of energy in the form of heat, methane gas, and/or compressed air becomes a key technology for the energy revolution.

Since the industrial revolution, the waste products first pollute water, the number and complexity of major conflicts between these different uses steadily rise and continue to do so today. Contemporary problems include the long-range effects of mechanical, chemical, and thermal energy storage on groundwater, and the complex aftereffects of hydraulic stimulation on deep geothermal energy or shale gas extraction. Furthermore, research into the underlying processes does not solely concentrate on subsurface domains but also examines the interaction with the atmosphere. For instance, atmospheric pressure influences the efficiency of compressed-air energy storage; and the land surface water budget strongly depends on the interaction between the porous structure of the soil and the highly turbulent flow distribution in the atmospheric boundary layer.

Predicting and controlling the reciprocal influence of subsurface projects, their impact on (ground) water, and their interaction with the atmosphere are key challenges that must be solved for a sustainable, safe society. Classical simulation approaches fail to deliver useful predictions for decision support in this respect. Despite the increasing amount of in-situ and remote sensing data, relevant subsurface properties cannot be adequately characterized due to their immense scale complexity. Thus, microtomography rock images may offer a wealth of information but they only work for very small samples. At very large scales, geophysical exploration also generates rich data, but the intermediate scales relevant for decision-supporting predictions are data-poor. Cutting-edge porous media research results in ever-new model concepts, yet with a limited range and scale of validity, and with insufficient data to choose among them. Multiscale simulations and scale-bridging techniques suffer from data scarcity on relevant scales, which precludes unique calibration.

Appropriate machine learning methods extend numerical and physics-based homogenization techniques by learning smaller-scale models in agreement with basic physical principles. Dynamic recalibration and data assimilation algorithms use incoming data from subsurface sensors to continuously update model parameters and improve forecasts. The residuals observed during data assimilation support a context-sensitive, adaptive choice between scientifically competing model concepts, and they are used to learn intelligent bias corrections. Active learning algorithms suggest optimal extensions of subsurface sensor networks to enrich the incoming data stream. Novel control algorithms improve subsurface operations while reliably restricting environmental impact to acceptable threshold levels.

Runtime metadata collected from the entire simulation workflow is used for the dynamic adaptation of models, numerical schemes, and computing resources. Novel context-aware visualization techniques and traceability concepts make it possible to interpret and manage the wealth of data sources and simulation results in an intuitive way. Eventually, users interact with this distributed data-integrated simulation environment from anywhere using a variety of immersive devices, thus facilitating informed decisions about complex engineered geosystems on demand. Today, the methods and techniques do not yet exist for realizing such far-reaching scenarios, but the research outlined in this proposal prepares the ground for these and other innovations.