Considering the inherent uncertainty of structural geological models, the non-uniqueness of geophysical inverse problems, and the growing availability of data, there is a need for methods that combine different types of data and allow for updating knowledge in a consistent way. By making use of the development of efficient, gradient-based MCMC algorithms, probabilistic inversion provides a tool for this. To test to what extent we can reduce the uncertainty of an initial geological model, we integrate geological modelling into a Bayesian inverse framework. Additional information can then be included in this inverse framework through likelihood functions. The proposed methodology is tested on a geological model of the structurally complex Kevitsa deposit in Finnish Lapland. By starting with an initial interpretation-based 3D geological model, we define the uncertainties in our geological model by means of probability density functions. Magnetic data and geological interpretations of borehole data are used to define geophysical and geological likelihoods respectively. To use the magnetic data in the inference, the mathematical description of the magnetic forward calculation is implemented for a 3D voxelised space, linking the geophysical data through magnetic rock properties to the uncertain structural parameters. The result of the inverse problem is presented in the form of probability distributions and ensembles of the realised models through visual analysis. The former is a statistical consideration of the results, whereas the latter is a visual representation for direct interpretation in a geological sense. The uncertainties in these visual representations are best presented by means of information entropy, which allows for a quantitative analysis. The results show that well-defined likelihood functions can reduce uncertainties in geological models and build on the complementary strength of different types of data. Where probabilistic inversion inherently provides uncertainty analysis, finding a single representative solution is less trivial. Therefore we conclude that the strength of the used methodology mainly lies in data integration and uncertainty quantification.

The recent discovery of the Zama field in offshore Sureste Basin in the southern Gulf of Mexico has piqued interest in the Mexican petroleum industry regarding the hydrocarbon potential within the region. In this study, a deposition-only thermal model was developed in an area of interest (AOI) in this region. Using structural input obtained from seismic interpretation conducted in Petrel©, the model was constructed and simulated in PetroMod© to acquire initial estimates of the maturity of the Tithonian J100 source rock in the AOI in terms of the maturity parameters Vitrinite Reflectance (%Ro) and Transformation Ratio (TR). A one-at-a-time (OAT) sensitivity analysis approach was also undertaken to assess the impact of uncertainties in some inputs in isolation on the output maturity in order to identify key uncertain input parameters. Based on the modeling results, at the C-1 well location of interest within the AOI, the source rock was simulated to be within the wet gas generation window based on simulated %Ro values in the present, having emerged into oil window between 41 - 27 Ma (late Eocene-Oligocene) and into the wet gas window between 6 - 0.5Ma (late Miocene-Pleistocene). Across the wider AOI, present-day %Ro values indicate the predominant presence of late oil to wet gas generation windows, with most of the source rock across the AOI having emerged into the oil generation window between 23 Ma and 11.6 Ma (early to middle Miocene). The highest uncertainties in simulated %Ro values were associated with uncertainties in the Bajocian (J60) autochtonous salt sequence thickness distribution and with the maturity models used for model simulation. At the C-1 well location, TR values of 95% to near 100% in the present indicate that most of the kerogen (source rock organic matter) has already been converted into hydrocarbons, with kerogen conversion initiated between 55 - 48 Ma at the location. Across the wider AOI, average present-day TR values in the range of 81 - 97% were simulated. The highest uncertainties in simulated TR values were also associated with uncertainties in the Bajocian (J60) autochtonous salt sequence thickness distribution, and with the source rock maturation kinetics input used for model simulation. The timing of maturation of the J100 source rock at the C-1 well location - assumed to have been constrained by the OAT sensitivity analyses results for simplicity - in relation to the timing of other petroleum system elements and processes indicated by literature, suggests that hydrocarbon accumulations charged from the Tithonian source rock in the reservoir intervals in the well vicinity are possible i.e. presence of hydrocarbon accumulations cannot be ruled out based only on maturation timing considerations.

The active layer thickness has become an important indicator in climate change research as permafrost degradation has long been documented. The thawing of permafrost causes the release of greenhouse gases accelerating Arctic warming. Monitoring and quantifying spatial and temporal changes of the active layer are challenging but crucial for reliable climate projections. Geophysical methods offer a non-invasive investigation of electrical properties and their distribution in permafrost areas, revealing phase transitions from water to ice. Subsurface electrical resistivity images can be obtained through inversion of electromagnetic data, yet are inherently ambiguous because of the ill-posed nature of the inverse problem. Since regularization methods offer the possibility to stabilize the inversion, lateral and spatial constraints are incorporated in the inversion algorithm to produce quasi-2D and quasi-3D subsurface models. The developed methodology is evaluated based on synthetic data sets to determine suitable inversion parameters, which are subsequently applied to a field example from the Seward Peninsula, Alaska. Laterally constrained inversion methods based on a few-layer starting model succeed in resolving sharp interfaces in quasi-layered environments. In more complex settings minimum-structure models can retrieve accurate subsurface representations leveraging on vertical and horizontal smoothness constraints. Enforcing lateral and spatial consistency between neighboring soundings thereby yields a similar degree of model smoothness. The inverted field data confirms the conclusions drawn from the synthetic study, as meaningful three-layered models with regard to electrical resistivities are recovered, indicating resistive snow overlying the conductive active layer and highly resistive permafrost. However, the inversion results imply that the snow layer has a significant effect on the predicted model. The implemented constraints help in reducing the ambiguity of the models, but uncertainties introduced by limited data availability cannot be overcome. The potential of adopting spatial and lateral constraints to the inversion is shown, although it becomes evident that additional a priori information needs to be integrated in the objective function in order to comprehensively image the active layer.

The inversion process of data sets acquired with the electrical resistivity tomography can be a computational expensive calculation, because it requires a lot of forward simulations. Since computational resources are limited, the inversion for large field surveys can require immense computing times. One approach for reducing the computing times of forward simulations is the reduced basis method. It splits the problem into a parameter-dependent and parameter-independent part, which allows a precomputation of the parameter-independent part to speed up the calculation. Since the method of reduced basis has not been applied for the problem of electrical resistivity tomography so far, the question arises if and how it is applicable for this problem. Therefore, I performed forward simulations of the electrical potential distribution using the finite elements method and simulations using the reduced basis method for comparing the results. The methods give similar results, but unfortunately deviations above 1*10^4 occurred, while the error tolerance of the reduced basis method is set to 1*10^5. Since the reduced basis method has been successfully applied on other applications, I assume that the problem is in my reduced basis simulation and can be solved by more fine tuning. By comparing the computing times of the simulations, a speed-up of 109.512 is achieved by using the reduced basis method. Furthermore, the pay-off is reached after 8.803 simulations. During the project, I noticed that only considered the resistivities of the subsurface as a varying parameter in the reduced basis is not sufficient. The source position needs to be a variable parameter in simulations of the electrical resistivity tomography, as well. This is necessary to tap the full potential of the reduced basis method in the application of the electrical resistivity tomography. The reduced basis method gives significant speed-ups and therefore, it has the potential to reduce the computing times of electrical resistivity tomography inversions.

Backward erosion piping is a dike failure mechanism. It is the internal erosion process by which sand is eroded away from underneath a dike or levee by seepage flow. This erosion process progresses in the direction opposite to the direction of seepage flow and forms a small pipe directly beneath the dike. As erosion continues, this process can lead to dike failure. During this erosion process, the groundwater flow pattern is subject to continuous change, due to the growth of the pipe. Self-potential (SP) monitoring is sensitive to changes in the groundwater flow pattern, because of the electrokinetic coupling between fluid flow and the streaming potential. The SP field due to flow underneath a test dike was modeled with a FreeCAD -> Gmsh -> pyGIMLi workflow. This workflow can also be used to effectively resolve a wide range of standard and customizable geophysical modeling and inversion tasks. After modeling, field experiments were conducted, on the same test dike, to further assess the possibilities and limitations of SP monitoring to track the progress of backward erosion piping. Given that it is essential to have an accurate resistivity model in order to find the location of the SP source, an integrated electrical resistivity tomography (ERT) and SP monitoring system was designed. The electrodes used in this monitoring system were polarizable stainless steel stakes. The reliability of polarizable electrodes was greatly overestimated, as they turned out to give unstable SP measurements. The reason for the inferior reliability of polarizable compared to non-polarizable electrodes, was found through extensive literature research. The reason being that the largest potential in any electrode originates from the contact between the metal and the electrolytes in solution. The metal of polarizable electrodes is in direct contact with the electrolytes in the soil, which have variable concentrations. Therefore, the potential measured fluctuates together with the concentration of soil electrolytes in contact with the metal. The metal of non-polarizable electrodes, on the other hand, is in contact with a solution of its own salt, which has a constant concentration. Finally, piping is not expected to be measurable with SP monitoring, before a large sand boil is visible in the field. Once a positive SP anomaly develops at the sand boil, changes in the SP field due to the growth of the pipe are expected to be too small relative to the SP anomaly associated with water flow through the sand boil. Even though an integrated ERT and SP monitoring system is known to provide useful information about the hydrology of a dike, such a system is not sensitive enough to be able to monitor the development of a backward erosion pipe.

Ultrosonic pulse-echo testing is gaining increasing attention for its feasibility, accuracy, and versatility in structural examinations of concrete constructions. A1040 MIRA 3D PRO is an emerging low-frequency pulse-echo ultrasonic tomography device. Developed in 2021, this device is claimed to be suitable for manual and automated concrete structure assessment with the help of its functions, such as 3D-TFM (Total Focusing Method)/FMC (Full Matrix Capture) imaging. However, both its capacity and limitations need to be investigated in more detail. In this thesis, experiments are conducted first on a 2D concrete structure to examine the data acquired when the transducers are working in linear configuration and matrix configuration (array modes). Then, measurements were carried out on a 3D concrete structure to check the 3D properties of the device with the help of the data processing software InterSAFT. The results show a good correspondence between different array modes and that 3D SAFT could visualize 3D structures in an adjacent position. The data recorded at a greater distance was overlaid by noise, which is attributed to the fact that the frequency of the ultrasonic waves was set too high. Finally, further experiments on frequency calibration, data comparison, and measurements on more complicated specimens were suggested for future research on MIRA 3D PRO.

CO2 and H2S are reinjected at Hellisheiði, Iceland, to reduce the emissions of greenhouse gases. An active-source seismic campaign was done in July, 2021, to image the reinjection reservoir with various seismic methods for consecutive monitoring. At the same time, passive-source seismic data was recorded for imaging. We process this passive data with seismic interferometry. For this, an illumination analysis is performed to filter noise panels that are dominated by surface-wave noise and to keep panels dominated by body-wave noise. Afterwards, panels with near-vertical events are autocorrelated to retrieve a zero-offset section. The full set of selected panels is crosscorrelated to retrieve virtual shot records. These are processed with a simple reflection seismological processing workflow to obtain a stacked section. The results show that the autocorrelated zero-offset sections appear more noisy, but are characterised by higher frequencies, while the crosscorrelated stacked sections are characterised by a lower-frequency content and contain more dipping reflectors. The major, horizontal reflectors correspond between the two types of sections. A rudimentary interpretation of reflectors is done, based on the two types of sections and compared with a local geological model. This shows that the major lithological differences and the base of the Hengill volcano can be distinguished in the interpreted section.

The automatic interpretation of well logs has been the focus of research, especially in oil and gas industry, for more than 50 years and, aside from that, benefits from the fast developments of machine learning algorithms during the recent decades. Moreover, Bayesian inference is increasingly utilized to model geological data, enabling the consideration of all available information and a quantification of uncertainties. In order to combine unsupervised segmentation of well data with 3D geological modeling, a fully automated approach to directly create threedimensional structural models from raw well data is intended and, further, tested on synthetic data with different standard deviations. For this purpose, unsupervised segmentation, which considers the statistical nature as well as the spatial correlation of the data, is combined with a zonation method that extracts interface information from clustered data by maximizing probabilities within continuous zones. This data is then screened to automatically obtain geological information and, furthermore, is inserted into the structural modeling algorithm, which is based on implicit potential-field interpolation while at the same time honoring the geological spatial continuity. It is shown that unsupervised segmentation is capable of segmenting raw well logs and that the zonation appropriately determines boundaries between stratigraphic units. Model reconstruction demonstrates that the fully automated process is proffcient at recovering several common subsurface structures. Moreover, the implementation of a three-dimensional model in the segmentation process, filling the empty space between boreholes, reduces uncertainties in the geological modeling routine. The combination of unsupervised segmentation and 3D geological modeling, resulting in a fully automated process, taking all available information into consideration, is found to be a suitable method in order to build structural geological modeling directly from raw well logs.

Marchenko inversion is a new way to invert seismic or electromagnetic data recorded during geophysical surveys. The inversion method uses Marchenko theory. This is a recent development which enables the retrieval of Green's functions at any place in the subsurface. A non-recursive Marchenko inversion method has already been introduced but in this thesis a recursive Marchenko inversion method is implemented and analysed. A recursive scheme lies at the center of this new method. In this thesis, the new method is implemented and tested on a 1D subsurface model. The recursive scheme is first validated. This is done by computing a reflection response with it and comparing it with a reflection response resulting from forward modeling. After this, the accuracy of retrieved local reflection coefficients from the recursive inversion method is determined. This is done by comparison with exact reflection coefficients of the subsurface model. After this, several different parameters of the used subsurface model, data computation and the recursive inversion method itself are investigated for their influence on the accuracy of the inversion method. In particular interest is the effect of interval time errors because these result in errors that can build up rapidly through the recursion. However, the method has a big advantage. It is shown that the recursive Marchenko inversion method has a way to retrieve the magnitude of made interval time errors and correct for these when interval times are overestimated. In this way the error build up is stopped. In the end, it is shown that the new method delivers high accuracy results and has an advantage in computational expense compared to the existing recursive Marchenko inversion method. It is concluded that the new method shows promising prospects and that it is worthwhile to investigate the method further.

Full-waveform inversion within a deterministic framework commonly uses gradient-based methods to minimize a least-squares error function. Due to the non-linearity of the problem, this function has several local minima. To avoid them, it is necessary to start the optimization procedure from a good initial model. Within a probabilistic framework, Markov Chain Monte Carlo methods are used to sample a probability density function that represents the possible solutions. In high-dimensional problems, traditional MCMC methods become inefficient as random transitions are unlikely to find the regions of high probability. Hamiltonian Monte Carlo appears as a method that can efficiently find these regions without spending computation time on regions of no interest. To evaluate the probability of a sample, the least-squares error is calculated. Synthetic data is obtained through an existing 2D frequency domain full-wave modelling code and compared to the observed data. Through the Hamiltonian Monte Carlo sampler, the high probability regions are sampled and used to build the probability density function that represents the possible solutions. The algorithm was tested with both synthetic and real data. For the synthetic case a 1D model was inverted. The normalized least-squares error was reduced by 98%. In the deepest section the true model lies outside the uncertainty range of the estimated model. This appears as the result of sampling narrow regions of the probability density function. Then, the variance of the samples is underestimated and therefore the error. For the real dataset, a 2D model was inverted. This model lacks some of the large-scale features compared to deterministic full-waveform inversion results. However, a good model was found without any a priori information and almost any manual intervention. This suggest that, at least for the studied example, the algorithm can provide a good starting model for deterministic full-waveform inversion. The results obtained through this approach are almost identical to those obtained with more labour intensive adjustments.

All over the world a so called oceanic 'hum' can be recorded, dominating the ambient noise field. The hum is generated by nonlinear interaction of ocean surface waves, radiating acoustic signals in the atmosphere and water (named microbaroms), and coupling to the solid earth as microseisms. The hum can be generated by local near-coastal sources and distant deep ocean sources. Deep oceanic generated 'hum' occurs within the range 0.1-0.3 Hz, dominantly around 0.2 Hz. The near-coastal generated 'hum' occurs at higher frequencies. Microseisms are mainly recorded as surface waves corresponding to near coastal activities. These surface waves overwhelm the weaker deep ocean signals, body and surface waves, because they attenuate less due to the shorter travel path. Previous studies are done with individual 3-component sensors, thus spectral data of all kind of microseisms, or for teleseismic distances only. This study of deep ocean microseisms is done by seismic arrays to analyze regional microseisms. Arrays provide more information about the microseisms by applying beamforming. This enables an improved localization of the source area, in specific near coastal activity or deep ocean generated. To validate the observed results a comparison with the numerical simulated source area is be made. Denote that the use of seismic arrays requires verification on the performance of arrays due to local conditions (sensitivity variances) to find possible deviations regarding the back azimuth. It is shown that, using classical beamforming, the recordings are dominated by near coastal surface waves. Applying adaptive beamforming using multiple signal classification, also other events become visible in the f-k spectrum. Due to this method an improved deep ocean source area is obtained. This observed source area agrees with the simulated source area. Improving the localization requires to define array correction factors.

Current Port of Rotterdam procedure is to define the nautical depth based on the density of a fluid-mud layer, that settles at the bottom of the port. Rheological parameters could be an improved indicator for nautical depth and decrease the frequency and cost of dredging. I investigate the use of reflected seismic waves to derive in situ rheological parameters. I model a simplified port profile to investigate the effects of o_set on the types of seismic reflections acquired. Four models of how the density of the fluid-mud layer varies with depth are proposed. I undertake simple laboratory experiments to measure the effects of consolidation time, density and frequency has on the seismic velocities of the fluid-mud layer. From these laboratory experiments I calculate the shear modulus, bulk modulus, Young's modulus and Poisson's ratio for the fluid-mud layer of varying densities. Reflected seismic-wave velocities prove to be promising in deriving in situ rheological parameters, elastic constants and, potentially, viscosity.