Global sensitivity analysis of multistage injection in geothermal reservoirs using surrogate models

A characterization regarding the mechanical response of subsurface reservoirs is of increasing interest for energy-related applications, including geothermal energy production and storage of georesources and waste. Modelling the dynamic response of geological formations to fluid injection often relies on fully coupled thermo-hydro-mechanical (THM) models, which provide a high-fidelity representation of the governing physical processes. These models support operational and design decisions under significant geological and parametric uncertainties. However, their high computational cost severely limits their applicability in large-scale statistical analysis and thus limiting the potential to account for these uncertainties.

Still, understanding how uncertainties in reservoir and operational parameters influence application-relevant outcomes is essential for stimulation design and risk mitigation. Global sensitivity analysis offers a quantitative framework to identify the controls on selected quantities of interest (QoIs). The choice of a QoI is inherently problem-dependent and reflects the specific operational objective or risk-related question being addressed, making it a central element in the interpretation of model results.

To overcome the computational demands of full-order THM simulations, we employ non-intrusive reduced-order modeling techniques to efficiently and accurately approximate the transient reservoir response. Projection-based model reduction methods target accurate, physics-based response characterization, resulting in interpretable, physics-consistent, and scalable surrogate models. We train surrogate models using solutions of the coupled THM equations. These surrogates are then used to perform global sensitivity analyses for different choices of QoIs. Finally, we demonstrate the proposed workflow through an application to the Groß Schönebeck geothermal field, featuring a multistage injection scenario, that provides a basis for future analyses targeting induced seismicity.