Uncertainty quantification of well log predictions

Abstract Uncertainty quantification (UQ) in well log prediction remains a significant challenge within geoscientific research because conventional methodologies yield single point estimates without addressing the associated predictive uncertainty. The absence of uncertainty assessment may result in suboptimal decisions during exploration phases and reservoir characterization. This study addressed this limitation by applying the conformalized quantile regression (CQR) technique to predict missing AC (P-Sonic) acoustic log curves while simultaneously quantifying the uncertainty of these predictions. The analysis used nine Log ASCII Standard (LAS) files from the Volve Field, with exploratory data analysis revealing substantial missingness, particularly in AC (P-Sonic) logs. Pearson correlation analysis identified strong relationships between AC (P-Sonic), density (DEN), and neutron (NEU) logs. The conformal prediction framework partitions the dataset into training, validation, and test subsets, enabling the derivation of statistically valid prediction intervals. The CQR approach achieved AC (P-Sonic) log predictions with 95% coverage. This work demonstrated the value of UQ in well log prediction by providing accurate predictions accompanied by reliable prediction intervals, thereby supporting robust and informed decision-making in exploration and reservoir characterization. The methodology developed herein offers a comprehensive tool for risk management, enabling geoscientists to proactively identify and mitigate potential adverse outcomes associated with well-log data utilization.