Unsupervised Learning Analysis of a Multi-Parameter Geophysical Database for Abu Dhabi Hydrocarbon Reservoirs

Abstract

Abstract Unsupervised learning technique was implemented to delineate locations of potential hydrocarbon reservoirs at offshore Abu Dhabi, U.A.E. Integrating several geophysical observations that are sensitive to different physical parameters in a single scheme results in more constrained and high-resolution geophysical models. We created a database comprising gravity and magnetic field data as primary attributes. Then, by implementing nonlinear inversion modelling of gravity and magnetic field data, depth-to-basement and depth-to-salt structures were derived as complementary attributes. Applying k-means clustering technique on the preprocessed data around the Ghasha oil field, the areas with higher probability of hydrocarbon reservoirs were distinguished. The regions that encompass known oil fields cover shallow Infracambrian salt interface which is associated with lower gravity signals due to lower density. Nonetheless, these areas are located over higher gravity and magnetic values resulted from the shallower basement which obscures the negative gravity anomaly expected from the salt structure. The clustering results also indicates that a different process is needed for creation of smaller hydrocarbon reservoirs. These cases are mostly located over the edges of the clusters where we have a transition from positive to negative anomalies. This is justified as trapping mechanism over the domes is different than the slopes.