Semester of Graduation

Spring 2022


Master of Science in Petroleum Engineering (MSPE)


Craft and Hawkins Department of Petroleum Engineering

Document Type



Gravity survey has played an essential role in many geoscience fields ever since it was conducted, especially as an early screening tool for subsurface hydrocarbon exploration. With continued improvement in data processing techniques and gravity survey accuracy, in-depth gravity anomaly studies, such as characterization of Bouguer and isostatic residual anomalies, have the potential to delineate prolific regional structures and hydrocarbon basins. In this study, we focus on developing a cost-effective, quick, and computationally efficient screening tool for hydrocarbon exploration using gravity data employing machine learning techniques. Since land-based gravity surveys are often expensive and difficult to obtain in remote places, we explore the use of satellite-based gravity, which is available throughout the Earth and updated periodically. Since the accuracy and resolution of the satellite gravity data are lower than land-based gravity measurement, satellite data was enhanced through a deep-learning-based super-resolution technique. We compare the use of land-based, satellite-based, and enhanced-satellite Bouguer and isostatic gravity data for the classification of hydrocarbon regions using both supervised and unsupervised machine learning techniques. In addition, a comparison of geostatistical models and Random Forest regression is performed for geospatial interpolation. The use of different combinations of input features (Bouguer, isostatic gravity, latitude, longitude coordinates) and prediction classes (oil, gas, oil and gas, no hydrocarbons) are evaluated and compared. Results indicate the successful application of supervised machine learning workflow for hydrocarbon classification using Bouguer and isostatic gravity anomalies with good prediction accuracy for both land-based and satellite-based gravity data. The results from unsupervised machine learning were less robust in comparison.

Committee Chair

Sharma Jyotsna