Research Roundup
Physics-guided unsupervised deep learning approach for the inversion of receiver functions.
Dalai B, Kumar P, Sen MK, Haldar C.
GEOPHYSICAL JOURNAL INTERNATIONAL
https://doi.org/10.1093/gji/ggaf347
Inversion of converted-wave data is essential for constructing realistic velocity models of Earth's internal structure, but is often complicated by the presence of dipping and anisotropic geological formations and the non-uniqueness of solutions. This study develops a physics-guided unsupervised deep-learning approach that integrates implicit neural representation with forward modeling to infer subsurface parameters directly from data without labelled inputs. By embedding physics within the learning process and employing transdimensional automatic layer detection, the approach achieves stable, high-resolution, and physically consistent inversion results in both synthetic and real field datasets.
Fig: Application of the proposed physics-guided unsupervised deep learning approach at station NP050 (Hi-CLIMB network). The inverted model captures crustal dip and anisotropy, with modeled receiver functions closely matching observations across back-azimuths.
