%0 Thesis
%A Hoven, Dominik
%T Multi-dimensional GPR full-waveform inversion for small-scale hydrogeophysical soil characterization
%V 643
%I RWTH Aachen University
%V Dissertation
%C Jülich
%M FZJ-2024-06115
%@ 978-3-95806-781-3
%B Reihe Energie & Umwelt / Energy & Environment
%P IX, 163
%D 2024
%Z Dissertation, RWTH Aachen University, 2024
%X A detailed understanding of the processes within the critical zone, which covers the area from the earth’s surface down to the aquifer, is essential for sustainable resource management and environmental protection. This zone exhibits complex flow and transport processes and supports critical ecosystem services such as water supply, agriculture, and climate regulation. However, imaging the complex critical zone accurately, especially at high resolutions required for a detailed analysis, presents significant challenges because of the variability of soil water content and complex subsurface structures. This thesis introduces a novel 2.5D ground penetrating radar (GPR) full-waveform inversion (FWI) method that enhances subsurface imaging by accurately incorporating 3D geometries, such as air and water filled boreholes, finite length antenna models, and lysimeter geometries, in the forward modeling of the GPR FWI. Furthermore, the 3D-to-2D data transformation with its assumptions, e.g. for the far-field, necessary for 2D GPR FWI, is not required with this method. We show in synthetic studies with different inversion methods (2D FWI, 2.5D FWI, 2.5D FWI with borehole, and 2.5D FWI with borehole and antenna) an improved source wavelet reconstruction with the inclusion of realistic borehole and antenna geometries for the data. The inclusion of these geometries in the forward model of FWI approaches can significantly improve the accuracy of conductivity reconstructions, with a reduction in the mean relative absolute error of conductivity of more than 20% compared to simple 2D FWI and 2.5D FWI. The improvement is particularly noticeable in high-contrast zones. Although including antenna geometries significantly increases computational requirements by a factor of í10, the quality of reconstruction remains similar to the case with only borehole inclusion. In contrast to ray-based inversion (RBI), where artifacts arise when using high-angle data (72.35°), FWI still provides reliable results. In a following analysis, we tested if a model that includes boreholes and finite length antenna models for experimental data measured with transmitter and receiver positioned in air and water filled boreholes can improve the effective source wavelet estimation. A synthetic test shows that using this approach, only one wavelet can be used for the reconstruction of both the unsaturated and saturated zone. However, we still observed challenges with the current antenna model to account for the different coupling in air filled boreholes for measured data. Using the new 2.5D FWI with borehole and antenna models and a single source wavelet, the results of the saturated zone reconstruction were similar to those observed in previous studies where four effective source wavelets were considered. To obtain reliable results in the unsaturated zone, it is necessary to adapt the antenna model to resolve existing discrepancies. Next to an improved reconstruction of small-scale structures in aquifers, small-scale processes in the soil-plant-atmosphere continuum are also of interest. In order to achieve a higher reconstruction resolution with the FWI for these processes, higher frequencies are necessary. In a first part, we indicate the constraints imposed by high-frequency GPR data, which require more precise starting models to fulfill the half-wavelength criterion of the GPR FWI. This cannot be met by the regular starting model approach of using RBI models. We show that a frequency-hopping approach can be used to generate starting models that meet these requirements. Furthermore, we investigated the influence of first-arriva and amplitude changes in the source wavelets on high-frequency GPR FWI. Utilizing an adapted heterogeneous model, we were able to show a more detailed reconstruction with higher frequency data compared to lower frequency data. In a next step, we extended the model building process of the 2.5D GPR FWI and are now able to include more complex geometrical structures like lysimeters in the forward model. As we faced challenges to use the 2D GPR FWI on experimental high-frequency data acquired on lysimeters, we first investigated the different GPR waves in synthetic studies at lysimeters filled with homogeneous and heterogeneous soils. We show the complexity of the GPR data, that includes air, direct, and reflected waves. We created a synthetic 3D GPR lysimeter dataset with a center frequency of 450 MHz and applied the novel 2.5D GPR FWI to this dataset. It demonstrates an exceptional good reconstruction of the soil and fit of the dataset by the inversion results, effectively simulating air-, real soil-, and reflected waves as well as revealing intricate soil properties. The newly developed 2.5D GPR FWI presented in this thesis enables the modeling and reconstruction of small-scale structures with high resolution. The application ranges from aquifer characterization to the now possible inversion of GPR data measured at lysimeter, providing a foundational framework for future research in high resolution subsurface imaging.
%F PUB:(DE-HGF)3 ; PUB:(DE-HGF)11
%9 BookDissertation / PhD Thesis
%R 10.34734/FZJ-2024-06115
%U https://juser.fz-juelich.de/record/1032276