Dynamic Coarsening for Efficient Simulation of Geothermal Energy Applications

Geothermal production and storage typically involve heat transport that is strongly localized around wells, in contrast to hydrocarbon recovery where pressure and multiphase flow processes act over much larger spatial scales. This spatial separation implies that significant computational savings can be achieved by concentrating resolution in regions with high temperature gradients while coarsening the grid elsewhere. In this work, we develop and investigate a dynamic coarsening framework for geothermal reservoir simulation in which coarse cells are constructed by aggregating an underlying fine-scale geomodel into a set of nested partitions. At each timestep, a dynamically adapted grid is assembled based on indicators that measure local solution complexity, enabling refinement near wells and other active regions while retaining coarser resolution elsewhere. The framework is flexible-applicable to general unstructured grids comprising arbitrary polyhedral cells-and computationally efficient because geometric and transmissibility properties for all partitions can be precomputed. Key novel elements include a sequential splitting scheme derived directly from the discrete conservation equations, a rigorous energy-conservative mapping between scales for arbitrary polyhedral grids, and geothermal-specific coarsening indicators that can be applied either statically, based on time-of-flight, or dynamically, updated at each timestep. We evaluate the method on representative underground thermal energy storage scenarios and compare results against fully resolved fine-scale simulations using various refinement indicators and coarsening strategies. All numerical experiments are performed using the geothermal module of the open-source MATLAB Reservoir Simulation Toolbox (MRST).