Seismic data reconstruction addresses the challenge of accurately restoring incomplete or damaged seismic datasets, which is crucial for subsurface imaging and exploration. The missing data is often due to equipment failure, signal loss, environmental noise, or poor geological conditions. It may also result from limitations in data acquisition, such as sparse sampling and noise interference. This thesis proposes a Clean Convergent alternating Projections onto Convex Sets (CCP) method for data reconstruction. By incorporating an initial value tweaking step into the alternating projections process in the Convergent alternating Projections onto Convex Sets (CP) method, this method aims to reduce ringing noise in the reconstruction results by the CP method. Initially, the traditional CP method and its application in seismic data reconstruction are introduced, highlighting its shortcomings in addressing ringing noise. To overcome this issue, a CCP method based on a non-local means algorithm for initial value tweaking is proposed and applied within the outer loop of the CP method. The theoretical foundation and implementation steps of the CCP method are discussed in detail, and its effectiveness is validated through a series of experiments. The experimental results demonstrate that, compared to the traditional CP method, the CCP method significantly reduces ringing noise and improves the quality of the reconstructed data. Various datasets, including images, 2D seismic section data from the SEAM II Arid model, and 3D seismic model cubes, are used to showcase the reconstruction capabilities of the CCP method in different scenarios. Finally, the thesis provides an in-depth discussion of the CCP method, including intermediate results in the outer loop, the necessity of data preconditioning, and parameter testing, offering a practical set of control parameter settings. The study shows that the CCP method has broad application prospects in data reconstruction, providing valuable references for future research.

In 3D seismic data acquisition, sufficiently dense spatial sampling is most often not possible, because of physical, financial, and temporal constraints. The resulting aliased data is a major obstacle for accurate subsurface images. Therefore, methods are needed to reconstruct these sparse data to an adequate sampling. We develop a novel kinematic wavefront-based seismic data reconstruction method that uses the existing nonlinear beamforming (NLBF) framework, which allows for building a more detailed sampling from a sparse input. We present the theory and methodology of our NLBF reconstruction algorithm and test it on a synthetic Society of Exploration Geophysicists Advanced Modeling (SEAM) Arid dataset and a field dataset. We attempt to answer the following question: can the NLBF framework be used for seismic data reconstruction, and how do these results compare to some conventional reconstruction methods? Control parameter tests are performed to find the optimal NLBF reconstruction, and results are compared to those from several control methods including the convergent alternating projection onto convex sets (POCS) and bootstrap POCS methods, which were also developed for this study. Our NLBF reconstruction method successfully reconstructs high-quality data on both datasets. We find that the optimal NLBF control parameters are ultimately dataset-dependent as the concrete data acquisition geometry plays a central role. Specifically for our datasets, optimal parameters include a time window of 12∆ts, an operator aperture of 600 m x 600 m, and a parameter trace interval of ∆x = ∆y = 60 m. The synthetic SEAM Arid NLBF reconstruction results show high trace fidelity to the ground truth. The field data NLBF results show better-reconstructed gaps, and wavefield variations closer to those of physical propagating waves in comparison to our control methods. These results show the potential of the NLBF reconstruction method to become a common data reconstruction tool used in the seismic industry.

The analysis of microseismic measurements acquired during borehole acquisition surveys is essential for a thorough understanding of the source mechanisms in hydraulic fracturing operations. Due to the injection of high-pressure fluids, induced fractures can produce seismic events of low magnitude that can be recorded and subsequently analyzed to infer the stress field at the location of the event. The seismic moment tensor has been widely used to describe general seismic sources as it can provide information about the type of motion and the distribution of forces. Estimating such quantities from the recorded data can significantly improve the real-time microseismic monitoring operations and help to make assumptions about the structure of the reservoir. However, several challenges have to be faced when working with microseismic borehole measurements. They are characterized by a low signal-to-noise ratio and a limited angle coverage, which may ultimately affect the predictions about the location and the fracturing behavior inside the reservoir. In this thesis the inversion of microseismic measurements for retrieving the moment tensor has been tackled by using a deep feedforward neural network. The seismograms used to train the network were generated through the discrete-wavenumber method. The neural network was used to predict the six independent moment-tensor components and the fault angles from seismic sources at different positions than those used during the training. The predictive capabilities of the network were tested for realistic borehole acquisition geometries and wave propagation models. The moment-tensor components were retrieved with good accuracy when using noisy seismograms and, non-double-couple source mechanisms. As the generation of synthetic data can be expensive in terms of memory consumption, the training and prediction have been also implemented in the frequency domain using only a limited portion of the Fourier transform of the seismograms. The inversion results indicated that using narrow bands can still yield satisfactory results when predicting the moment-tensor components.

A key practice in the development of unconventional hydrocarbon resources is to monitor hydraulic fracturing operations and analyze in detail the induced microseismicity, for understanding the extent of the volume affected by the fractures. Even though microseismic data are one of the few geophysical measurements that can be used for this purpose they are not always used at their full potential: the common analyses are limited to the interpretation of the locations of the induced seismic events in terms of fractured volume, while the propagation and the dynamic evolution of the microseismic events that are directly related to the source process are not studied. The source moment tensor can be used to describe the deformation mechanism occurring at the seismic source and the corresponding geomechanical parameters, which are relevant aspects for the monitoring of hydraulic fracturing operations. Its value can be estimated through an inversion process using the P-wave and S-wave waveforms from the observations at the seismic receivers. However, the limited aperture of the receiver arrays in borehole acquisition is inadequate to solve the inverse problem. The goal of this thesis project is study the feasibility of estimating the source moment tensor with borehole microseismic data making use of machine learning algorithms. The task is done by implementing a forward model to compute waveforms generated by known values of the source moment tensor in a borehole acquisition geometry (forward problem). Forward modeling was carried out using a modied version of the discrete wavenumber method. Afterwards, this data is used to train a neural network to estimate the moment tensor at the source given the microseismic data (inverse problem). The neural network was designed, adapting an existing architecture used in previous studies to solve similar inverse problems. The implementation was done entirely in the Python programming language, using the Tensorflow library for the machine learning parts. The network was trained with the dataset generated in the forward model until reaching an optimal mean squared error minimum. Finally, the capability of the network to invert microseismic measurements and retrieve acceptable values of the source moment-tensor components was tested using synthetic data, resulting in an average prediction accuracy of 0.997. Our results indicate that for this case, a neural network is able to satisfactory invert for the full moment-tensor components. Additionally, the neural network was also trained to handle the presence of Gaussian noise, achieving an average prediction accuracy of 0.990 for tested synthetic data with 10% Gaussian noise.

Forensic investigations focused on determining clandestine buried weapons, narcotics or even homicide evidence, can be expensive, inefficient and depend greatly on prior information and the tools available. Geophysical tools have potential to improve these investigations, on the ground that they can detect shallow buried objects in a non-invasive way. This report studies the use of electrical resistivity tomography (ERT) in combination with ground penetrating radar (GPR) to detect buried objects at two different sites, both with conditions found here in The Netherlands. The first is a test site at the Technical University of Delft, The Netherlands. Here, two plastic barrels are buried, one is empty and the other is filled with metal rods. The second site is at the ARISTA Facility in Amsterdam, The Netherlands. Here, human cadavers are buried for forensic research. The ERT was used in two ways, the first was to produce 2 dimensional (2D) surveys, which were combined to create a 3 dimensional (3D) model. The second method made measurements that created a 3D model directly. For each the dipole-dipole array was used. At the TU Delft site the results for the 2D method showed clear resistivity anomalies at the locations of the barrels. These anomalies corresponded to clear reflections in the GPR radargrams. The results at the ARISTA facility are inconclusive due to damage in the instrument used. The grid designs made for both ERT methods could however be used in a continuation of this study, and future research should be done within this topic to improve forensic investigations.

Since its inception in 1975, Genetics Algorithms (GAs) have been successfully used as a tool for global optimization on non-convex problems in a wide range of real world applications. Its creation was inspired by natural adaptation and selection mechanisms that evolve from one population of chromosomes to a fitter population by means of an artificial natural selection dictated by the operators of elitism, crossover and mutation. An advanced Genetic Algorithm (aGA) was proposed by Sun et al. [2017], and this algorithm seeks the global maximum of n-th dimensional non-convex functions. However, convergence speed is a key factor for the success of a global optimization algorithm when it comes to scalability; in production even a slight efficiency improvement matters as it can easily takes weeks or even months to process a huge data set. Therefore, the goal of this project is to improve the convergence speed of the currently available aGA by simultaneously enhancing both its global and its local search capabilities. To this end, two solutions were proposed. The first is a modified version of the well known Island model GAs and the second was named Self Adaptive Differential Evolution (SADE) fine tuning scheme.
After a successful demonstration of its improved performance on several multi-modal test functions, the enhanced Genetic Algorithm (eGA) is used to tackle two common non-linear geophysical problems: static correction and Common Reflection Surface (CRS) stacking. In the former, the near-surface related time-shifts are estimated without resorting to an explicit velocity-depth model; instead, the events of interest are aligned in a data-driven fashion by maximizing the stacking power. The latter is a novel alternative to the traditional Common Midpoint (CMP) stacking that has proven to yield higher quality images, specially when applied to low Signal to Noise Ratio (SNR) data or data with challenging structures like the anomalies encountered in the subsurface. This improvement in the quality of the stacked image is attributed to a non-local mean mentality, which enables using traces from different CMP gathers to the CMP gather being resolved. Owing to the high non-linearity of both problems, they are ideal test beds for global optimization algorithms. The effectiveness of the eGA is demonstrated on synthetic data sets with promising results in these problems.

The Jakobshavn glacier was responsible for approximately 1 mm eustatic sea level rise in the period of 2000 to 2010 [Howat et al. 2011]. As such, the Jakobshavn glacier became one of the largest outlet glaciers in Greenland [Joughin et al. 2004]. Ice flow velocities within the same period reached over 10 km/yr with strong seasonal variation [Howat et al. 2011, Joughin et al. 2012]. More recently from 2011 until 2013, even higher ice flow velocities of at least 15 km/yr were observed [Lemos et al. 2018]. Due to the relatively high ice flow velocities, the ice discharge plays the largest role in the mass balance (MB) of the Jakobshavn glacier. Quantification of the ice discharge from ice flow velocities is however, not a common procedure. Yet the evolution of the ice discharge of single glaciers not only improves understanding of the climate-cryosphere system, but also aids quantification of sea level contribution on a drainage basin scale. To that end, this study embodies an indirect ice discharge estimation of the Jakobshavn glacier over the period of November 2010 until March 2016 using altimetry (CryoSat-2 Level 1b (L1b)) and gravimetry (GRACE Level 2 (L2)) data in combination with a regional climate model (RACMO 2.3p2). By subtraction of the altimetric and gravimetric mass balance estimates from the atmospheric component (i.e. the surface mass balance (SMB)), two ice discharge estimates are obtained. This approach does not suffer from the drawbacks involved when estimating ice discharge from velocity fields directly, which are based on offset tracking. Data gaps for long polar nights and clouds in the visible spectrum and decorrelation in general, when the duration between subsequent images over the same location is long, are thus avoided. This is because offset tracking algorithms require recognisable characteristics in subsequent satellite recordings to determine the velocity, i.e. satellite recordings need to be sufficiently correlated. To derive the mass balance from altimetry data, adequate spatial sampling is desired. To that end, this study applies swath processing to CryoSat-2 L1b data with an adapted waveform sample selection criterion to obtain an unprecedented spatial sampling with about 2 order of magnitude more height observations compared to conventional retracking techniques. As a consequence, elevation changes can be derived at a relatively high spatial resolution (250 m).
The elevations are converted to elevation changes, volume change and mass change using weighted least squares estimations (WLSE), hypsometric averaging and density models, respectively. The GRACE-based mass change estimate is acquired using a point-mass assumption at the location of the Jakobshavn glacier. The known, simulated point mass is then scaled to the observed mass by GRACE. In addition, data weighting of GRACE Stokes' coefficients is attempted using the full noise covariance matrix. Subsequently, a LSE is used to infer the mass balance from the two time series (with and without weighting of the Stokes' coefficients).