A Novel 2.5D Deep Network Inversion of Gravity Anomalies to Estimate Basement Topography

Abstract

A novel 2.5D intelligent gravity inversion technique has been developed to estimate basement topography. A deep neural network (DNN) is used to address the fundamental nonuniqueness and nonlinearity flaws of geophysical inversions. The training data set is simulated by adopting a new technique. Using parallel computing algorithms, thousands of forward models of the subsurface with their corresponding gravity anomalies are simulated in a few minutes. Each forward model randomly selects the values of its parameter from a set of predefined ranges based on the geological and structural characteristics of the target area. A DNN model is trained based on the simulated data set to conduct the nonlinear inverse mapping of gravity anomalies to basement topography in offshore Abu Dhabi, United Arab Emirates. The performance of the trained model is assessed by making predictions on noise-free and noise-contaminated gravity data. Eventually, the DNN inversion model is used to estimate the basement topography using pseudogravity anomalies. The results show the depth of the basement is between 7.4 km and 9.3 km over the Ghasha hydrocarbon reservoir. This paper is the 2.5D and improved version of the research (SPE-211800-MS) recently presented and published in the Abu Dhabi International Petroleum Exhibition & Conference (31 October–3 November 2022) proceedings.