Capturing scales of heterogeneity in models of fluvial geothermal reservoirs: Grid resolution, upscaling strategies, and hierarchies of sedimentary architecture

• A systematic assessment of geothermal meander-belt fluvial reservoir models is undertaken • The relative importance of finer- vs coarser-scale heterogeneities is not readily predictable • Well-doublet performance does not vary systematically with static model resolution • Flow-based permeability upscaling and arithmetic averaging return similar flow behavior • Injected coldwater plume shape varies in relation to well orientation and heterogeneity. This study investigates the influence of modeling choices related to the scale of reservoir heterogeneity on the predicted performance of geothermal doublets in fluvial low-enthalpy geothermal reservoirs. Fourteen geocellular grids were created to systematically analyze the impacts of numerical grid resolution, permeability upscaling methodology, and modeled scales of sedimentary architecture, using MODFLOW-2005 and MT3D-USGS to simulate groundwater flow and heat transport for well-doublet operation over a 35-year period. The results reveal complex relationships between these choices and simulated reservoir behavior: the considered factors have significant influence on injection pressures but only a modest effect on production temperatures (with variations within 2°C after 35 years across all models), likely due, at least in part, to a relative dominance by thermal diffusion over heat advection in the considered scenarios. Simplification of geological architecture through omission of fine-scale features may augment the hydraulic impact of larger flow barriers, such as abandoned-channel mud plugs. The highest injection pressures were simulated on grids that embody sedimentary architectural elements but lack internal facies heterogeneity. The permeability upscaling method also has an effect: simulations on grids upscaled using harmonic averaging consistently yield the highest near-injector pressures, followed by those based on geometric averaging and arithmetic averaging. The dynamic behavior of grids upscaled via flow-based upscaling closely approximates that of grids upscaled using arithmetic averaging, suggesting that bulk hydraulic behavior is dominated by the connectivity of high-permeability units. The performance gap between grids following different upscaling methods decreases significantly for higher grid resolution. Simulations of geological models that incorporate increasingly detailed geological features predict cold-water plumes with slightly more complex shapes and tortuous fronts, as documented by values of plume surface-to-volume ratio. The complexity of the cold-water plume shape, as measured by the surface-to-volume ratio, is slightly higher for well doublets oriented at a high angle to the channel-belt axis, but does not increase systematically with the resolution at which fine-scale features are represented.