Answers are grounded in this paper and cite their sources.
Downward continuation can enhance weak anomaly signals and enable the conversion of data from different altitudes, making it a crucial technique in potential field data processing. However, due to the strong ill-posedness of the problem, its practical application has long been hindered by noise amplification, divergence, and instability. To address this, this article proposes a convergent and stable downward continuation method based on a physics-informed neural network (NN), which combines data-driven learning with physical constraints to effectively constrain the ill-posed problem. Specifically, Dirichlet boundary conditions and the Laplace equation of the geomagnetic field are incorporated into the NN as physical regularization terms, thereby limiting the exponential amplification of noise during the downward continuation process. Compared with traditional downward continuation methods, the proposed approach exhibits significantly enhanced noise resistance and improved stability at larger continuation depths. Tests on synthetic models show that the method can reduce the root mean square error under different noise levels, while maintaining high-resolution recovery of anomaly features. Applications to real aeromagnetic data from different regions indicate that the results obtained by this method are closer to actual measurements, demonstrating that the proposed approach can achieve accurate and stable continuation for real data. It provides a stable and constrained new method for the downward continuation of the potential field data.
No takes yet. Share an insight, caveat, or question.
Chan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: