Multi-objective waveform inversion of shallow seismic wavefields. Issue 3 (28th November 2019)

• Record Type: Journal Article
• Authors: Pan, Yudi; Gao, Lingli; Shigapov, Renat
• Is Part Of: Geophysical journal international. Volume 220:Issue 3(2020)
• Page Start: 1619
• Page End: 1631
• Publication Date: 2019-11-28
• Subjects: Inverse theory -- Waveform inversion -- Surface waves and free oscillations; Geophysics -- Periodicals; 550
• DOI: 10.1093/gji/ggz539
• Languages: English
• ISSNs: 0956-540X

Abstract

SUMMARY: It has been increasingly popular to use shallow-seismic full-waveform inversion (FWI) to reconstruct near-surface structures. Conventional FWI tries to resolve the earth model by minimizing the difference between observed and synthetic seismic data using a certain criterion (conventionally, l 2 -norm of waveform difference). In this paper, we propose a multi-objective waveform inversion (MOWI) in which the similarity of data is quantified and minimized using multiple criteria simultaneously. By doing so, we expand the dimensionality of objective space as well as the mapping from data space to objective space, which provides MOWI higher freedom in exploring the model space compared to single-objective FWI. We combine three different scalar-valued objective functions into a vector-valued multi-objective function which measures the similarity of the waveform, the waveform envelope, and the amplitude spectra of the data, respectively. This multi-objective function takes not only trace-based waveform and wave packet similarity but also the dispersion characteristics of surface waves into account. Furthermore, the uncertainty in the inversion result could be estimated and analysed quantitatively by the variance of the optimal models. We propose a modified ϵ-constraint algorithm to solve the multi-objective optimization problem. Two synthetic examples are used to show the advantages of using MOWI compared to single-objective FWI. We also test the efficiency of MOWI by using two synthetic shallow-seismic examples, which confirm that MOWI can converge to a better result compared to the conventional single-objective FWI.