A Bayesian Gaussian Process Framework for the Discrete Inversion of 2D Gravity Data: Applications to the West Korea and Godavari Basins

Abstract

While Bayesian inference is widely applied to magnetotelluric, seismic, and magnetic datasets, its application to gravity anomalies remains relatively unexplored. To address this, we introduce InDIA (Inversion of Density Interface and its Application), a Python-based Bayesian inference framework utilizing a structural Gaussian process. InDIA is developed to estimate discretized subsurface layer depths and densities while explicitly quantifying uncertainty. Unlike previous techniques restricted to continuous depth profiles and localized issues, InDIA integrates diverse prior information to effectively resolve both local and regional gravity anomalies. This approach successfully mitigates the multi-parameter challenges and convergence issues common in local gradient-based optimization techniques, such as Adam. The superiority of this algorithm has been validated through various synthetic models (involving multi-prism configurations, two-layer models with lateral density variations, heterogenous subsurface model with lateral and vertical density variation along with faulted dipping models) featuring both constant and variable density distributions—whether lateral, vertical (prior-constrained), or both—incorporating Gaussian noise, Random-walk noise, Systematic noise and Salt and Pepper noise to replicate real conditions. Furthermore, the framework’s field applicability is demonstrated using two real gravity datasets. First, it tackles multi-layered subsurface profiling in the West Korea Basin. Second, it simultaneously resolves depth and prior-constrained vertical density variations in the Godavari Basin, India. In both field applications, the inverted parameters are geologically viable and closely align with previously established models, confirming InDIA as a reliable tool for gravity data inversion.