Data-Driven Prediction of Caprock Breakthrough Pressure and Permeability for CO2 Storage Integrity: A Meta-Analysis

Carbon capture and storage (CCS) is a key strategy for achieving net-zero emissions, with long-term storage security depending on caprock integrity. Caprocks act as sealing barriers preventing upward CO2 migration; however, breakthrough pressure and permeability remain difficult to predict across variable geological settings. This study presents a data-driven meta-analysis of caprock sealing performance using published datasets covering confining pressure, temperature, depth, porosity, permeability, and breakthrough pressure. Missing values were treated using Multiple Imputation by Chained Equations (MICE), and predictive performance was evaluated using ordinary least squares regression, polynomial regression, neural networks, physics-informed residual/hybrid models, and least absolute shrinkage and selection operator (LASSO) regression with polynomial feature expansion. Severe multicollinearity was identified between temperature and depth, and conventional regression showed weak permeability prediction due to multicollinearity and geological heterogeneity, whereas the physics-informed residual model improved performance from R2 = 0.28 to R2 = 0.77 by incorporating depth- and temperature-dependent residual corrections. For breakthrough pressure, the physics-informed hybrid model showed strong fitting performance, while the LASSO-derived empirical equation provided a more generalizable prediction with a validated test R2 of 0.71 and a mean absolute error of 2.31 MPa. The proposed framework supports preliminary screening for safe CO2 injection pressure design and long-term storage integrity assessment.