NEWS & EVENTS

Understanding Earth's subsurface structure is essential for resource exploration and hazard assessment. Traditional seismic inversion methods rely on adjoint-state methods for gradient computation. They face some major limitations: complex manual derivations and poor flexibility across different wave physics. These barriers significantly impact research efficiency and limit our ability to understand deep Earth structures. To overcome these barriers, we propose a novel seismic inversion framework using automatic differentiation (AD). AD is a modern computational technique that automatically computes exact derivatives through chain rule decomposition. To show the generality of our framework, we performed ten cross-scenario tests across domains (time or frequency), wave types (acoustic, SH, P-SV, visco-acoustic or visco-elastic), and losses (waveform, traveltime or amplitude). We also evaluated our method on the OpenFWI benchmark dataset to compare it with NN methods. Practicality is further demonstrated by a checkerboard test in the Nankai subduction zone, which is challenging for NN methods due to the lack of suitable training datasets. For a field application, we applied ambient noise differential AD tomography to data from the Southeastern Suture of the Appalachian Margin Experiment (SESAME) and obtained three 2D Love-wave shear velocity (Vsh) models. The imaged Paleozoic suture zone, Mesozoic rift basins, and Moho interface are consistent with previous studies. Our results highlight the unifying role of AD in geophysical inverse problems beyond gradient computation, showing promise for broader future applications across geoscience.

For additional information, please contact Mr. Xi WANG, u3009920@connect.hku.hk.