Harnessing Machine Learning for Caprock Capacity and Seal Integrity Assessment in CCS: A Multi-Attribute Seismic Inversion Study from the F3 Block, North Sea

Abstract

Accurate characterization of subsurface petrophysical properties is a critical prerequisite for evaluating the suitability of geological formations for carbon capture and storage (CCS), particularly in identifying high-capacity reservoirs and effective sealing intervals. This study explores the use of a machine learning approach—Random Forest (RF) regression—for multi-attribute seismic inversion to predict porosity and acoustic impedance in the F3 Block, offshore Netherlands. The integration of ten seismic attributes with two well log datasets enables the construction of predictive models capable of resolving complex lithological variations within deltaic settings. The RF algorithm’s robustness against geological noise and its ability to model nonlinear relationships offer significant advantages over conventional inversion workflows, especially in heterogeneous and interbedded formations. The results demonstrate that RF-based inversion produces petrophysical volumes with improved spatial continuity and alignment with depositional patterns, offering a promising avenue for CCS site screening and reservoir-seal evaluation. The method’s ability to capture subtle textural and facies changes also enhances understanding of potential CO₂ migration pathways and trap integrity. This research underscores the potential of data-driven inversion frameworks in supporting geoscientific decision-making for CCS development, particularly in data-limited or geologically complex offshore regions.