High‐Precision Delineation of Curie Point Depth Under Remanent Magnetization Using Physics‐Informed Neural Networks

Abstract The Curie point depth (CPD), a key indicator of the lithosphere's thermal structure, is typically estimated using spectral analysis or interface inversion methods. However, these approaches often neglect the effects of remanent magnetization, leading to substantial uncertainty. To address this limitation, we propose a Curie‐Physics Informed Neural Network for high precision CPD estimation under strong remanence conditions. Our approach integrates realistic geological modeling by synthesizing a double magnetic interface with spatially correlated remanent magnetization. Synthetic magnetic anomalies are then efficiently generated using frequency domain computations. Furthermore, to ensure geological rationality, we incorporate a spectral constraint into the network, leveraging it as physical information to guide and regularize the training process. The effectiveness of our method is validated on both synthetic magnetic anomalies and field data from the Ordos region, China, demonstrating an improvement in precision under remanence conditions compared to conventional approaches.