DNN Inversion of Gravity Anomalies for Basement Topography Mapping

Abstract A gravity inversion technique using Deep Neural Networks (DNN) was developed to construct the 2D basement topography in offshore Abu Dhabi, UAE. Forward model parameters are set based on the geological features in the study area. Hundreds of thousands of synthetic forward models of the basement and their corresponding gravity anomalies are generated in a relatively short time by applying parallel computing. The simulated data are input to our DNN model which conducts the nonlinear inverse mapping of gravity anomalies to basement topography. To assess the model's robustness against noises, DNN models are retrained using datasets with noise-contaminated gravity data whose performances are evaluated by making predictions on unseen synthetic anomalies. Finally, we employed the DNN inversion model to estimate the basement topography using pseudo gravity anomalies over a profile in offshore UAE.