A unified Framework for Multiscale Pore-Scale-to-Field Reservoir Forecasting

Thesis Abstract

This study addresses the fragmented understanding of fluid flow across scales in hydrocarbon reservoirs by proposing a unified framework that integrates pore-scale physics with field-scale forecasting to reduce uncertainty in production planning and enhanced oil recovery (EOR) design. The aim is to develop a multiscale reservoir forecasting framework that coherently links pore-scale transport phenomena, meso-scale heterogeneity, and macro-scale reservoir performance through a theory-driven, data-informed model. Specific objectives are (i) to quantify pore-scale flow mechanisms, including capillary pressure-saturation relationships, wettability effects, and pore-network connectivity, (ii) to upscale these mechanisms through a multiscale upscaling scheme that preserves essential nonlinearities and hysteresis, (iii) to integrate upscaled properties with a dynamic reservoir simulator to generate predictive field-scale forecasts under varying operating conditions, (iv) to validate the framework against rigorous field data from three representative reservoirs and perform sensitivity analyses to identify dominant drivers of forecast uncertainty, and (v) to formulate decision-support guidelines for reservoir management and EOR planning. The methodology adopts a mixed-methods, theory-driven research design anchored in percolation theory, homogenization theory, and stochastic upscaling. The population comprises carbonate and sandstone reservoirs with diverse pore structures. A sample of 12 representative core plugs per rock type is selected for detailed pore-scale imaging and core-flood experiments, totaling 24 cores, complemented by digital rock physics simulations. Data collection instruments include micro-computed tomography (micro-CT) for pore-network characterization, enhanced oil recovery laboratory experiments (waterflood and gas-CO2 schemes), capillary pressure-saturation curves, contact angle measurements for wettability, and time-lapse seismic-aligned dynamic data from field operators. Additional field data from three mature reservoirs provide production histories, pressure, and saturation data sets, totaling 10,000+ data points. Analytical techniques include Bayesian hierarchical modeling to fuse pore-scale parameters with upscaled properties, regression analysis to quantify correlations between pore metrics and macroscopic permeability, and multiscale finite element/finite volume upscaling with homogenization to derive field-scale parameters. Reservoir simulation employs a dual-physics upscaling approach embedding pore-scale derived relationships within a standard black-oil or compositional model, calibrated via history matching using Markov Chain Monte Carlo (MCMC) methods. Validation comprises cross-pailure testing and out-of-sample forecasts over a 20-year planning horizon, with performance metrics including root-mean-square error (RMSE) and Kling-Gupta Efficiency (KGE). Theoretical grounding integrates percolation theory for connectivity thresholds, Ho-Lee-like capillary dynamics, and pore-network modeling to interpret multiscale flow paths. Conceptual integration is formalized in a multiscale forecasting algorithm that propagates uncertainty from microscopic parameters to field-scale forecasts. Expected findings indicate that a small set of pore-scale descriptors—pore-throat size distribution, fractal dimension of pore networks, and wettability state—explain a substantial portion of field-scale heterogeneity and forecast uncertainty when upscaled with a retention-aware homogenization technique. The framework is anticipated to reduce forecast error by 15–25% relative to conventional upscaling approaches across tested reservoirs, and to identify critical thresholds in connectivity and wettability that trigger non-linear recovery responses under varied injection strategies. The study contributes new knowledge by operationalizing a repeatable, theory-consistent multiscale forecasting workflow that preserves nonlinear capillary and hysteretic effects, integrates uncertainty propagation, and provides a transparent bridge from micro- to macro-scale reservoir behavior. Practical implications include improved well-placement optimization, more robust EOR design, and enhanced risk assessment for production forecasting. Conclusions will emphasize the value of linking pore-scale physics with field-scale forecasting through a unified, probabilistic upscaling framework, demonstrating improved predictive capability and decision-support utility. Recommendations will address extending the framework to include chemical EOR effects, incorporating real-time downhole sensing data for continuous model updating, and developing standardized datasets for cross-site validation to generalize the framework across diverse reservoir types.