Multi-task Unscented Kalman Inversion for joint inversion of receiver function and surface wave dispersion

TL;DR

This paper introduces a Multi-task Unscented Kalman Inversion framework for joint inversion of receiver function and surface wave dispersion, improving robustness, accuracy, and efficiency in geophysical modeling.

Contribution

It presents a novel Bayesian joint inversion method that shares information between different observations without derivatives, offering an efficient Gaussian approximation.

Findings

Demonstrates superior robustness and accuracy

Achieves high computational efficiency

Provides effective Gaussian posterior approximation

Abstract

Based on the recently developed theory of Unscented Kalman Inversion in computational mathematics, we proposed a Bayesian joint inversion framework, i.e., Multi-task Unscented Kalman Inversion (MTUKI), and apply it to the joint inversion of receiver function (RF) and surface wave dispersion (SWD). This method can share information between different observations in a derivative-free way and provide an efficient Gaussian approximation to the posterior distribution of model parameters (thickness and S-wave velocity in each layer of media). The theory and experiments show that our proposed framework demonstrates superior performance in terms of robustness, accuracy, and high efficiency.