@phdthesis{ArboledaZapata2023,
  author    = {Arboleda Zapata, Mauricio},
  title     = {Adapted inversion strategies for electrical resistivity data to explore layered near-surface environments},
  doi       = {10.25932/publishup-58135},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-581357},
  school      = {Universit{\"a}t Potsdam},
  pages     = {115},
  year      = {2023},
  abstract  = {The electrical resistivity tomography (ERT) method is widely used to investigate geological, geotechnical, and hydrogeological problems in inland and aquatic environments (i.e., lakes, rivers, and seas). The objective of the ERT method is to obtain reliable resistivity models of the subsurface that can be interpreted in terms of the subsurface structure and petrophysical properties. The reliability of the resulting resistivity models depends not only on the quality of the acquired data, but also on the employed inversion strategy. Inversion of ERT data results in multiple solutions that explain the measured data equally well. Typical inversion approaches rely on different deterministic (local) strategies that consider different smoothing and damping strategies to stabilize the inversion. However, such strategies suffer from the trade-off of smearing possible sharp subsurface interfaces separating layers with resistivity contrasts of up to several orders of magnitude. When prior information (e.g., from outcrops, boreholes, or other geophysical surveys) suggests sharp resistivity variations, it might be advantageous to adapt the parameterization and inversion strategies to obtain more stable and geologically reliable model solutions. Adaptations to traditional local inversions, for example, by using different structural and/or geostatistical constraints, may help to retrieve sharper model solutions. In addition, layer-based model parameterization in combination with local or global inversion approaches can be used to obtain models with sharp boundaries. In this thesis, I study three typical layered near-surface environments in which prior information is used to adapt 2D inversion strategies to favor layered model solutions. In cooperation with the coauthors of Chapters 2-4, I consider two general strategies. Our first approach uses a layer-based model parameterization and a well-established global inversion strategy to generate ensembles of model solutions and assess uncertainties related to the non-uniqueness of the inverse problem. We apply this method to invert ERT data sets collected in an inland coastal area of northern France (Chapter~2) and offshore of two Arctic regions (Chapter~3). Our second approach consists of using geostatistical regularizations with different correlation lengths. We apply this strategy to a more complex subsurface scenario on a local intermountain alluvial fan in southwestern Germany (Chapter~4). Overall, our inversion approaches allow us to obtain resistivity models that agree with the general geological understanding of the studied field sites. These strategies are rather general and can be applied to various geological environments where a layered subsurface structure is expected. The flexibility of our strategies allows adaptations to invert other kinds of geophysical data sets such as seismic refraction or electromagnetic induction methods, and could be considered for joint inversion approaches.},
  language  = {en}
}
@article{ArboledaZapataGuillemoteauTronicke2022,
  author    = {Arboleda-Zapata, Mauricio and Guillemoteau, Julien and Tronicke, Jens},
  title     = {A comprehensive workflow to analyze ensembles of globally inverted 2D electrical resistivity models},
  series = {Journal of applied geophysics},
  volume    = {196},
  journal   = {Journal of applied geophysics},
  publisher = {Elsevier},
  address   = {Amsterdam},
  issn      = {0926-9851},
  doi       = {10.1016/j.jappgeo.2021.104512},
  pages     = {12},
  year      = {2022},
  abstract  = {Electrical resistivity tomography (ERT) aims at imaging the subsurface resistivity distribution and provides valuable information for different geological, engineering, and hydrological applications. To obtain a subsurface resistivity model from measured apparent resistivities, stochastic or deterministic inversion procedures may be employed. Typically, the inversion of ERT data results in non-unique solutions; i.e., an ensemble of different models explains the measured data equally well. In this study, we perform inference analysis of model ensembles generated using a well-established global inversion approach to assess uncertainties related to the nonuniqueness of the inverse problem. Our interpretation strategy starts by establishing model selection criteria based on different statistical descriptors calculated from the data residuals. Then, we perform cluster analysis considering the inverted resistivity models and the corresponding data residuals. Finally, we evaluate model uncertainties and residual distributions for each cluster. To illustrate the potential of our approach, we use a particle swarm optimization (PSO) algorithm to obtain an ensemble of 2D layer-based resistivity models from a synthetic data example and a field data set collected in Loon-Plage, France. Our strategy performs well for both synthetic and field data and allows us to extract different plausible model scenarios with their associated uncertainties and data residual distributions. Although we demonstrate our workflow using 2D ERT data and a PSObased inversion approach, the proposed strategy is general and can be adapted to analyze model ensembles generated from other kinds of geophysical data and using different global inversion approaches.},
  language  = {en}
}
@article{RumpfTronicke2014,
  author    = {Rumpf, Michael and Tronicke, Jens},
  title     = {Predicting 2D geotechnical parameter fields in near-surface sedimentary environments},
  series = {Journal of applied geophysics},
  volume    = {101},
  journal   = {Journal of applied geophysics},
  publisher = {Elsevier},
  address   = {Amsterdam},
  issn      = {0926-9851},
  doi       = {10.1016/j.jappgeo.2013.12.002},
  pages     = {95 -- 107},
  year      = {2014},
  abstract  = {For a detailed characterization of near-surface environments, geophysical techniques are increasingly used to support more conventional point-based techniques such as borehole and direct-push logging. Because the underlying parameter relations are often complex, site-specific, or even poorly understood, a remaining challenging task is to link the geophysical parameter models to the actual geotechnical target parameters measured only at selected points. We propose a workflow based on nonparametric regression to establish functional relationships between jointly inverted geophysical parameters and selected geotechnical parameters as measured, for example, by different borehole and direct-push tools. To illustrate our workflow, we present field data collected to characterize a near-surface sedimentary environment Our field data base includes crosshole ground penetrating radar (GPR), seismic P-, and S-wave data sets collected between 25 m deep boreholes penetrating sand- and gravel dominated sediments. Furthermore, different typical borehole and direct-push logs are available. We perform a global joint inversion of traveltimes extracted from the crosshole geophysical data using a recently proposed approach based on particle swarm optimization. Our inversion strategy allows for generating consistent models of GPR, P-wave, and S-wave velocities including an appraisal of uncertainties. We analyze the observed complex relationships between geophysical velocities and target parameter logs using the alternating conditional expectation (ACE) algorithm. This nonparametric statistical tool allows us to perform multivariate regression analysis without assuming a specific functional relation between the variables. We are able to explain selected target parameters such as characteristic grain size values or natural gamma activity by our inverted geophysical data and to extrapolate these parameters to the inter-borehole plane covered by our crosshole experiments. We conclude that the ACE algorithm is a powerful tool to analyze a multivariate petrophysical data base and to develop an understanding of how a multi-parameter geophysical model can be linked and translated to selected geotechnical parameters.},
  language  = {en}
}