h1

TY  - THES
AU  - Mozaffari, Amirpasha
TI  - Towards 3D crosshole GPR full-waveform inversion
PB  - Rheinisch-Westfälische Technische Hochschule Aachen
VL  - Dissertation
CY  - Aachen
M1  - RWTH-2022-03244
SP  - 1 Online-Ressource : Illustrationen
PY  - 2022
N1  - Englische und deutsche Zusammenfassung. - Veröffentlicht auf dem Publikationsserver der RWTH Aachen University
N1  - Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2022
AB  - High-resolution imaging of the subsurface improves our understanding of the subsurface flow and solute transportation that can directly help us protect groundwater resources and remediate contaminated sites. The ground penetrating radar (GPR) is a useful non/minimal invasive method that consists of a transmitter (Tx) unit that emits electromagnetic (EM) waves and a receiver (Rx) that measures the arriving electromagnetic waves and can provide high-resolution tomograms of the subsurface properties. In specific, the crosshole GPR setup in which two-neighbouring boreholes are placed in the earth can provide much more in-depth access to the target area. However, the interpretation of the GPR data remains challenging. The simpler ray-based inversion (RBI) is computationally attractive while fail to provide high-resolution tomograms as the results always smoothed over the target area. The full-waveform inversion (FWI) can provide detailed subsurface tomograms that can carry up to more than an order of the magnitude resolution compared to RBI from the same data set. A sophisticated method such as FWI requires detailed modelling tools and powerful inversion algorithm that needs significant computational resources. In last decades, by exponential increase in computing power and the memory, alongside to wider usage of high performance computing resources; FWI application in GPR data gain popularity. All these computational advances such as FWI method. could be very demanding to be modelled in 3D domain. Thus, some fundamentals assumptions are made to reduce the computational requirements, especially computational time and required memory by using 2D modeling domain. Despite the usefulness of these simplifications, these assumptions led to introducing inaccuracy that compromises the performance of the FWI in complex structures. We investigated the effect of the assumption that enables us to use a 2D model instead of a computationally expensive 3D modelling to simulate the EM propagation. These assumptions are made for specific state that not necessary is always valid, and therefore it can introduce inaccuracies in transferred data. Study of several synthetic cases revealed that the performance of the 3D to 2D transformation in complex structures such as high contrast layer is much lower than what is anticipated. Therefore, in the complex subsurface system; 2D transferred data inherently carry inaccuracy that jeopardises the accuracy of any further analysis such as FWI. Thus, we introduced a FWI that utilise a native 3D forward model to use the original measured 3D data. The novel method is called 2.5D FWI, and it showed improvements compared to 2D FWI for synthetic and measured data.A better modelling tool such as the 3D forward model provides a useful platform for simulating the subsurface and measuring devices involved to a higher degree of accuracy. We used previously introduced 3D forward model to build a realistic model of the GPR Tx and Rx antenna that called finite-length antennas and the boreholes that these antenna are placed to carry out the measurements. Our studies showed that realistic antenna and borehole-fluid representation provides more realistic travel-time and wave-form shape for GPR data. These more accurate simulated data increases the accuracy of the FWI results as reducing the uncertainty in the inversion system.It is a known issue for GPR community that EM waves that travelled with a high-angle between the Tx and Rx shows inconsistency in their travel-time and therefore could jeopardise travel-time inversion results. Even though this effect is almost consistent, there was no concerts reason for this issue except systematic erroneous measurements. Thus, it is common pre-processing standard to discard these high-angle data (usually above 50°). Our findings regarding the contribution of the borehole-fluid to changes in travel-time of the EM waves showed the high-angle travel-time is not an error rather than consistent effect the borehole-fluid in travel times. We laid the mathematical explanation of this phenomena and introduced a correction method that could predict this issue and compensate for it. Lastly, we applied this correction method on the realistic synthetic data and showed that RBI results improved when the correction method is used.
LB  - PUB:(DE-HGF)11
DO  - DOI:10.18154/RWTH-2022-03244
UR  - https://publications.rwth-aachen.de/record/843585
ER  -