Induced seismicity (i.e., earthquakes triggered by human activities) is one of the most prominent and important risks associated with use of the underground for exploitation (natural gas, oil, geothermal energy, and mining) or for storage (waste water, CO2 sequestration). Within Switzerland, deep geothermal energy is one of the potential pillars of the Energy Strategy 2050 to provide 24/7 baseload power after phasing out of nuclear. Induced seismicity is one of the major challenges of this attractive green energy source, as evidenced by the St. Gallen and Basel geothermal pilot.
As natural permeability in the deep underground is too low to allow for economic geothermal water circulation, enhancing permeability by so‐called hydraulic stimulation remains the method of choice. However, hydraulic stimulation is, by definition, a process that induces seismicity as it injects water under high pressure into the reservoir volume to induce failure on pre‐existing, rough fractures in order to promote self‐propping permeability enhancement. Accordingly, fundamental and applied research with the ultimate goal to develop strategies and tools for managing and limiting induced seismicity is a major international focus of current research in geophysics and reservoir engineering. Predictive assessment of seismic risk in near‐real time is considered the most crucial aspect for proactive operational management of stimulation, exploitation or storage to minimize risk, avoid unacceptable seismic hazard and enhance the societal acceptance.
In this project, we propose to improve, to extend, and to adapt to current and upcoming generations of supercomputers the software framework that is currently used at ETH Zurich. Eventually, the work done in FASTER will enable or nearly enable the desired near‐real time assessment. A highly innovative feature of FASTER – from an application point of view – is that the software, for the first time in near‐real time application, will include simulating the mechanics of microseismicity rather than following a simple statistical approach, thereby greatly enhancing the quality of predictions. We will build on previous and current, non‐real time developments by the different partners of our team that already closely collaborate on this and related topics. The team is formed from core PIs of the Swiss Competence Center for Energy Research “Supply of Energy” (SCCER‐SoE). SCCER‐SoE has the mandate to perform application‐oriented research and development towards enabling deep geothermal energy in Switzerland. As a unique advantage of the proposed research, SCCER‐SoE will perform two hydraulic stimulation pilot and demonstration experiments on different scales (and will be involved in a third, industry‐lead one) during the duration of the project, i.e., large actual use cases for the software will be in place for testing and validation while the project is running.
Clearly, the huge amounts of incoming data demand for a software framework, which is tailored and optimized towards reservoir stimulation. The need for near‐real time analysis, the complex nature of the problem (coupled, non‐linear physical processes acting on complex fracture network geometries combined with sparse and uncertain data about these subsurface geometries), and the exponential scaling nature of seismicity require high performance computing in combination with scalable state‐of‐the art numerical methods and their efficient and flexible implementation as an essential ingredient.
From a computational point of view, the main challenges are the underlying inverse problem, i.e. the identification of the unknown fracture networks, and the fast solution of the corresponding forward problems. The currently used approach is based on the evaluation of many samples of reservoir stimulations with different fracture networks and boundary conditions. Each of the samples requires a forward problem with a fracture network to be solved. Thus, the main computational burden can be found in the evaluation of the samples. Moreover, the choice of the sampling strategy and the number of the computed samples obviously is important.
In FASTER, we will augment the current software framework for assessing microseismicity during reservoir stimulation developed by researchers at ETHZ/SED with fast and scalable solution methods for the arising inner forward problems (multigrid for fracture networks), which have been developed at ICS/USI in Lugano. We will furthermore improve on the side of the sampling strategy, which is used as an outer loop in the hazard analysis. On the side of the simulation software, this goal will be achieved by:
- Providing a library for the fast solution of reservoir stimulation with multigrid methods: Fracture‐Network Multi‐Grid (FNMG).
- Designing and implementing a control‐module for the sampling process, which incorporates different sampling strategies, which will allow for CPU as well as GPU based forward models.
- Testing and optimizing this software framework for the currently available supercomputers and cluster in Switzerland and preparing for upcoming generations of supercomputers.
For the FNMG library, we will ensure in particular excellent strong scalability on the level of one or several nodes, as this is highly relevant for smaller samples. We will, however, also make sure that FNMG scales well for larger problems, in order to allow for large scale reservoir simulations. This work will build on currently available software at ICS/USI.
For the control module, we will use python for calling different forward kernels for the reservoir stimulation. Efficient sampling strategies such as Multi‐Level Monte Carlo or multifidelity will be implemented. At ICS/USI, already such libraries for sequential sampling exist. On the software side, by using the library UTOPIA for an embedded domain specific language, also developed at ICS/USI, we will make sure that future changes in hardware will require only minimal changes on the software side.
We finally note that the work in FASTER will have a three‐fold outcome in very different facets of research and development:
- Near‐real time seismicity assessment based on incoming data in order to assist decision processes on the operator side of deep ego‐energy projects.
- Large‐scale prediction of the thermo‐hydraulic performance of the reservoir utilizing HPC systems.
- New libraries for the simulation of fracture networks (FNMG) and for efficient sampling methods in an HPC context.
Image copyright by Dr. Dimitrios Karvounis (ETH).
Prof. Dr. Thomas Driesner; PI; ETH
Prof. Dr. Stefan Wiemer; Co-PI; ETH
Prof. Dr. Domenico Giardini; Co-PI; ETH
Prof. Dr. Rolf Krause; Co-PI; USI-ICS
PhD Marco Favino; Researcher; USI-ICS
PhD Alessio Quaglino; Researcher; USI-ICS
Platform for Advanced Scientific Computing (PASC);