Accurate estimation of reservoir porosity is a critical factor in evaluating geological formations for carbon dioxide (CO₂) storage in carbon capture and storage (CCS) projects. Porosity directly influences the storage capacity and injectivity of subsurface reservoirs, making its accurate prediction essential for effective CO₂ sequestration planning. Traditional porosity estimation methods based on core analysis are reliable but often expensive, time-consuming, and limited in spatial coverage. With the increasing availability of well-log data, machine learning techniques provide an efficient data-driven alternative for predicting reservoir properties. This study proposes a machine learning–based framework for porosity prediction using boosting ensemble algorithms to support CO₂ storage assessment. Well-log data collected from the Mena Murtee-1 well in the Darling Basin, Australia, are used as input features, while laboratory-corrected porosity values serve as the target variable. Data preprocessing techniques are applied to remove noise, handle missing values, and eliminate multicollinearity among input parameters. Ensemble boosting algorithms including AdaBoost Regression, Gradient Boost Regression, and Extreme Gradient Boost Regression (XGBoost) are implemented and evaluated using standard statistical performance metrics. Experimental results demonstrate that boosting ensemble algorithms effectively capture complex non-linear relationships between well-log parameters and porosity values. Among the evaluated models, Extreme Gradient Boost Regression achieves the highest prediction accuracy and provides reliable porosity estimates for subsurface formations. The proposed framework enhances reservoir characterization accuracy and supports efficient evaluation of geological formations for carbon dioxide storage.
Simla et al. (Thu,) studied this question.