Uncertainty quantification in 3D static modeling: Improving reserve estimation accuracy with adaptive physics-informed Monte Carlo simulation in Y-field, Niger Delta Basin, Nigeria

Abstract

Accurate hydrocarbon reserve estimation is critical yet challenging in the complex Niger Delta Basin. This study employs a novel adaptive Physics-Informed Neural Network (PINN) to improve reservoir characterization and Hydrocarbon Initially In Place (HIIP) predictions. Core data from Well J-2 (average porosity 0.25, permeability 516.01 mD) served as the calibration baseline. A linear model (gradient 1.057, intercept −0.0077) calibrated well log porosity for uncored wells, identifying three rock types with RT2 being dominant. The reservoir exhibits significant heterogeneity, with average shale volumes ranging from 0.18 to 0.41. Zonal analysis reveals high hydrocarbon potential in Wells J-1, J-3, and J-4, characterized by low water saturation (0.16–0.19) and high Net-to-Gross (NTG) values, notably in J-1 (0.85) and J-3 (0.72). Conversely, J-2 and J-5 show higher water saturation (0.57 and 0.98, respectively). Complexity and uncertainty scores were used to guide adaptive sampling. The complexity score averaged 16, peaking at 87.7 in the western sector, while model uncertainty remained negligible (<0.0005). A 100,000-iteration Monte Carlo simulation, grounded in these physics-guided distributions, yielded a realistic HIIP estimate of 65.96 MMbbl. Results indicate that the western reservoir sector, defined by higher permeability and NTG values, represents the optimal target for future development.