Efficient Forecasting of Production Statistics Combining Single-phase and Multiphase Flow Models

To accurately represent reservoir heterogeneity, geological models of subsurface formations are typically built at fine scales, incorporating a high density of gridblocks. However, uncertainty quantification, often necessitating simulations across multiple realizations or scenarios, becomes computationally intensive for such fine-scale models, particularly when applied to multiphase flow systems. In this study, we present a hybrid framework combining sparse high-fidelity multiphase simulations with cost-effective single-phase models to accelerate uncertainty quantification. By upscaling single-phase flow solutions, clustering representative realizations, and interpolating multiphase responses, our method achieves computational speedups exceeding 100x while preserving ensemble statistics (mean, standard deviation and cumulative distribution function). Inspired by multilevel Monte Carlo principles, we replace hierarchical resolution sampling with physics-aware fidelity transitions, validated across permeability fields with varying heterogeneity. The approach accurately predicts well performance (bottom-hole pressure, production rates) using fewer than 10 realizations versus conventional 1000+ Monte Carlo trials, offering a practical pathway for uncertainty-aware reservoir optimization.