Smoothness‐constrained time‐lapse inversion of data from 3D resistivity surveys. Issue 1 (1st March 2013)

• Record Type: Journal Article
• Authors: Loke, M.H.; Dahlin, T.; Rucker, D. F.
• Is Part Of: Near surface geophysics. Volume 12:Issue 1(2014)
• Page Start: 5
• Page End: 24
• Publication Date: 2013-03-01
• Subjects: Earth (Planet) -- Surface -- Periodicals; Geophysics -- Technique -- Periodicals; Engineering geology -- Periodicals; Geophysics -- Periodicals; Planets -- Surfaces; Engineering geology; Geophysics -- Technique; Geophysics; Earth (Planet); Periodicals; 550
• DOI: 10.3997/1873-0604.2013025
• Languages: English
• ISSNs: 1569-4445

Abstract

ABSTRACT: Three‐dimensional resistivity surveys and their associated inversion models are required to accurately resolve structures exhibiting very complex geology. In the same light, 3D resistivity surveys collected at multiple times are required to resolve temporally varying conditions. In this work we present 3D data sets, both synthetic and real, collected at different times. The large spatio‐temporal data sets are then inverted simultaneously using a least‐squares methodology that incorporates roughness filters in both the space and time domains. The spatial roughness filter constrains the model resistivity to vary smoothly in the x‐, y‐ and z‐directions. A temporal roughness filter is also applied that minimizes changes in the resistivity between successive temporal inversion models and the L‐curve method is used to determine the optimum weights for both spatial and temporal roughness filters. We show that the use of the temporal roughness filter can accurately resolve changes in the resistivity even in the presence of noise. The L1‐ and L2‐norm constraints for the temporal roughness filter are first examined using a synthetic model. The synthetic data test shows that the L1‐norm temporal constraint produces significantly more accurate results when the resistivity changes abruptly with time. The model obtained with the L1‐norm temporal constraint is also less sensitive to random noise compared with independent inversions (i.e., without any temporal constraint) and the L2‐norm temporal constraint. Anomalies that are common in models using independent inversions and the L2‐norm and L1‐norm temporal constraints are likely to be real. In contrast, anomalies present in a model using independent inversions but that are significantly reduced with the L2‐norm and L1‐norm constraints are likely artefacts. For field data sets, the method successfully recovered temporal changes in the subsurface resistivity from a landfill monitoring survey due to rainwater infiltration, as well as from an experiment to map the migration of sodium cyanide solution from an injection well using surface and borehole electrodes in an area with significant topography.