João Paulo Guilherme Rodrigues Alves, Claudio Riccomini

This study applies machine learning techniques for forecasting petrophysical properties (density, porosity, and permeability) in the basalts of the Serra Geral Group, located in the Paraná Basin, Brazil. These properties are crucial for the successful implementation of carbon capture and storage (CCS), an important technology to combat climate change. Employing machine learning models—XGBoost, Gradient Boosting, and Random Forest—the research aims to overcome the limitations of traditional empirical methods that often fail to capture the complex variabilities in basalt formations. The models were applied to 28 wells within the study area. The interpolation of the well data indicated that the northern region of the Serra Geral Group in the State of Santa Catarina exhibits optimal conditions for geological storage. From 600 to 900 m, the basalts present suitable intervals ranging from 7 to 22 m thick, with density lows of almost 2.1 g/cm³, high peaks of 17.8 % apparent porosity, and permeability of 55 μD. The results demonstrated significant improvements in accuracy of property predictions compared to empirical methods from the literature, highlighting the potential of machine learning to enhance the feasibility and reliability of CCS in basaltic formations. This study contributes to the ongoing efforts to optimize CCS technology by providing a more accurate geological assessment of suitable storage sites.