An efficient sequential strategy for realizing cross-gradient joint inversion: method and its application to 2-D cross borehole seismic traveltime and DC resistivity tomography. Issue 2 (29th January 2018)

• Record Type: Journal Article
• Authors: Gao, Ji; Zhang, Haijiang
• Is Part Of: Geophysical journal international. Volume 213:Issue 2(2018:May)
• Page Start: 1044
• Page End: 1055
• Publication Date: 2018-01-29
• Subjects: Electrical resistivity tomography (ERT) -- Joint inversion -- Seismic tomography; Geophysics -- Periodicals; 550
• DOI: 10.1093/gji/ggy026
• Languages: English
• ISSNs: 0956-540X

Abstract

Summary: Cross-gradient joint inversion that enforces structural similarity between different models has been widely utilized in jointly inverting different geophysical data types. However, it is a challenge to combine different geophysical inversion systems with the cross-gradient structural constraint into one joint inversion system because they may differ greatly in the model representation, forward modelling and inversion algorithm. Here we propose a new joint inversion strategy that can avoid this issue. Different models are separately inverted using the existing inversion packages and model structure similarity is only enforced through cross-gradient minimization between two models after each iteration. Although the data fitting and structural similarity enforcing processes are decoupled, our proposed strategy is still able to choose appropriate models to balance the trade-off between geophysical data fitting and structural similarity. This is realized by using model perturbations from separate data inversions to constrain the cross-gradient minimization process. We have tested this new strategy on 2-D cross borehole synthetic seismic traveltime and DC resistivity data sets. Compared to separate geophysical inversions, our proposed joint inversion strategy fits the separate data sets at comparable levels while at the same time resulting in a higher structural similarity between the velocity and resistivity models.