Derivation and computation of full Hessian kernels in acoustic VTI media
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Hessian kernels have been utilized in acoustic and elastic FWI to potentially mitigate parameter trade-offs, but their application in vertically transversely isotropic (VTI) media remains underexplored. A significant practical challenge in constructing these kernels lies in the substantial disk space and I/O overhead required for storing and transforming wavefields during inversion. Here, we derive the full Hessian kernel expressions for VTI media using the velocity-stress wave equation and the second-order adjoint method. We then develop a wave-equation solver, Devito-VTI-Hdm (DVH), an extension of the Devito framework, to compute the gradient and Hessian kernels simultaneously on-the-fly with checkpointing. To absorb boundary wavefields, we implement the CPML in Devito and DVH. A synthetic test shows that the Hessian kernel construction using DVH is approximately twice as fast as the classical wavefield storage method over 500 time steps. The advantage of using DVH becomes more pronounced as the number of time steps increases. The results indicate that the Hessian kernels are more sensitive to small-scale structures than gradients, probably offering some potential for imaging fine-scale heterogeneities without requiring high-frequency stages. Furthermore, the Hessian kernel exhibits varying sensitivity to different model parameters, which may help explain inter-parameter trade-offs for a given data measurement and may offer some valuable insights for subsequent inversion.
Fields of Study & Subjects
Seismic Imaging and Inversion Techniques
Seismic Waves and Analysis
Geophysical and Geoelectrical Methods
Publication details
- Journal / source
- Second International Meeting for Applied Geoscience & Energy
- Volume
- 1
- Pages
- 1976–1980
- Publication date
- 10 February 2026
- Language
- EN