Abstract

Gravity anomaly inversion plays a significant role in geophysical exploration by enabling the interpretation of subsurface density distributions from surface gravity measurements. This project presents a comparative and integrative study of two approaches for gravity anomaly inversion: the Gauss-Newton method and Artificial Neural Networks (ANN). The Gauss-Newton method is implemented as an iterative, gradient-based optimization algorithm to minimize the misfit between observed and forward-modeled gravity data. In parallel, ANN is utilized as a data-driven approach to learn nonlinear relationships between anomaly patterns and corresponding subsurface parameters. By training the ANN on synthetic datasets generated through forward modeling, the model demonstrates the ability to generalize and predict subsurface features effectively. The results show that while the Gauss-Newton method provides precise, physics-based inversion, the ANN offers significant advantages in speed and generalization. The combined use of both methods enhances inversion reliability, offering a hybrid framework that is both computationally efficient and robust for complex geological settings.