Mini-batch-based joint inversion of surface wave dispersion and gravity anomaly for high-resolution crustal velocity structure of China and its adjacent regions

Abstract

High-resolution crustal velocity models are essential for understanding the tectonic evolution of China and its adjacent regions, while inversion based solely on an individual geophysical dataset inherently suffers from uncertainty and limited resolution. To mitigate these limitations, we propose a mini-batch-based joint inversion method that integrates surface wave dispersion and gravity anomaly, and apply it to achieve high-resolution crustal shear-wave velocity structure beneath China and its adjacent regions. We improve the joint inversion framework by Du et al. (2021) through randomly partitioning the surface wave dispersion data into mini-batches, to overcome computation and memory issues with large dataset. Besides, to enhance the sensitivity and constraints of Bouguer gravity anomaly on crustal and uppermost mantle structures, we apply a spherical harmonic expansion to remove long-wavelength background signals. The proposed mini-batch strategy ensures stochastic utilization of the full dataset, significantly reduces computational costs, accelerates convergence, and improves resistance to noise interference. The joint inversion results exhibit apparent improvements in structural resolution, allowing for clearer delineation of segmented low-velocity zones across the Tibetan Plateau, relatively high-velocity anomalies beneath the Lhasa and southern Qiangtang Terrane, stable cratonic features beneath the Sichuan and Tarim Basin, and low-velocity anomalies beneath the Qilian, Changbai and Taiwan orogenic belts. The results also provide refined constraints on the geometry and depth of the Moho interface, demonstrating high sensitivity to complex tectonic regions. These findings provide important constraints on regional tectonic mechanisms. The proposed joint inversion method holds broad application potential in large-scale, massive-data geophysical imaging.