Theme: 2. Mitigation

Topic Area: 2c. Carbon Capture & Storage: Scaling

Background/Objectives:

Oil and gas reservoirs represent attractive containers to sequester carbon dioxide (CO₂) because they are accessible, reservoir properties are known, and they previously contained stored fluids. However, planners must quantify the relative magnitude of the CO₂ storage resource in these reservoirs to formulate a comprehensive strategy for CO₂ mitigation. Even reconnaissance-type estimates of CO₂ storage resources of known oil and gas reservoirs may require complicated calculations involving (1) estimates of recoverable oil and gas, (2) reservoir properties (depth, temperature, pressure, etc.), and (3) the physical properties of oil and gas. This presentation provides results of the application of machine learning (ML) algorithms to by-pass these computations to yield rapid estimates of CO₂ storage resources.

Approach/Activities:

ML algorithms are computationally effective because they do not impose strong assumptions on the data-generating process that standard statistical procedures require. ML algorithms can capture highly complex relationships among predictor variables. In Attanasi and Freeman (2023), reservoir characteristics were used to estimate the CO₂storage resources for onshore and offshore oil and gas reservoirs located in Europe by applying a series of complex engineering calculations. Here, we propose a different approach that is data-driven rather than engineering-driven.

We begin with 1732 gas reservoirs and 824 oil reservoirs located in Europe, each containing CO₂ storage potential of at least one million metric tonnes of CO₂. We first select a random subset of the data to represent a training set consisting of reservoir characteristics and corresponding estimates of CO₂ storage resources based on engineering calculations, and then train a series of ML algorithms, including Random Forest, Gradient Boosting Trees, XGBoost, and deep neural networks, on this data set. The resulting models are then applied to the remaining data to assess the performance of each respective algorithm.

Results/Lessons Learned:

We demonstrate that the ML models perform well when predictions are compared to the engineering estimates on both the training and test data sets, representing a more data-driven approach to estimating CO₂ storage resources.

Attanasi, E.D. and Freeman, P.A., 2023, Reconnaissance survey for potential energy storage and carbon dioxide storage resources of petroleum reservoirs in Western Europe. Natural Resource Research 32:1839-1858.

Menu

Philip Freeman