An interpretable and explicit machine learning technique for predicting CO₂ storage and oil production in residual oil zones

Reliable determination of CO2 storage capacity and cumulative oil production is essential for the successful integration of carbon capture and storage with enhanced oil recovery (CCS-EOR). In this study, we develop an explicit and interpretable artificial neural network (ANN) model to predict CO2 storage mass and cumulative oil output in residual oil zones of depleted oil reservoirs, encompassing a broad spectrum of key reservoir characteristics and operational settings. The model relies on nine influential input parameters, including depth, porosity, and CO2 injection rate, and was trained using a large dataset generated from reservoir simulations. The developed ANN models demonstrated strong predictive performance on the testing data, achieving R² values of 0.9982 and 0.9617 for CO2 storage and cumulative oil production, respectively. Unlike many machine learning models in the subsurface domain, the developed model is presented in an explicit form, enhancing its adaptability for integration into software platforms used in reservoir management. Furthermore, model interpretability was ensured through the application of the connection weights algorithm, which quantified the relative influence of each input variable and its effect direction on the predicted outputs. Results indicate that increasing values of parameters such as reservoir thickness, permeability, porosity, and injection rate positively influence both CO2 storage and oil recovery. Conversely, higher values of bottom-hole pressure, depth, and residual oil saturation (Sorw) were associated with a reduction in these outputs. Additionally, the computational efficiency of the model was evaluated, and a practical implementation pathway was proposed through its conversion into a graphical user interface (GUI). The resulting tool offers a fast, reliable, and user-friendly solution for optimizing CCS-EOR strategies in field applications. Overall, this work presents a significant advancement in data-driven reservoir forecasting, providing actionable insights for enhancing both CO2 storage and enhanced oil recovery in complex reservoir environments. Document Type: Original article Cited as: Longe, P., Iyiola, Z., Ejehu, O., Onu, J. An interpretable and explicit machine learning technique for predicting CO2 storage and oil production in residual oil zones. Sustainable Earth Resources Communications, 2026, 2(1): 13-36. https://doi.org/10.46690/serc.2026.01.02