Abstract: The global transition toward lower-carbon energy increases the need for dependable subsurface characterization, both to develop geothermal resources and to design and monitor safe carbon capture and storage. That makes seismic imaging methods that are reliable, efficient, and cost-effective more important than ever. This thesis develops multi-dimensional seismic processing and imaging methods that remain effective when acquisition is limited or unconventional, with a focus on sparse seabed surveys and seismic-while-drilling (SWD) data. The first contribution targets sparse seabed acquisition, where irregular and widely spaced receiver geometries challenge conventional imaging workflows. Redatuming relocates the recorded wavefield from the acquisition surface to a chosen subsurface datum (or vise versa), providing a foundation for imaging at depth. Traditionally, redatuming relies on single-scattering approximations. Marchenko methods extend redatuming to full-wavefield data, offering a more comprehensive approach to subsurface imaging. Despite their undoubted potential, currently available Marchenko methods are not suitable for data acquired with sparse seabed seismic acquisitions. To address this gap, this thesis introduces the Upside-down Rayleigh–Marchenko (UD-RM) method, a practical redatuming scheme that reconstructs full-wavefield responses, including free-surface effects, from irregular and sparse receiver geometries. A complete multi-dimensional pre-processing framework has also been developed and applied to North Sea field data, leading to images with reduced multiple-related artifacts and fewer pre-processing requirements than single-scattering-based approaches. Synthetic and field examples demonstrate that the approach maintains an accurate representation of subsurface structures under increasingly sparse receiver acquisitions, supporting applications such as $\mathrm{CO}_2$ storage characterization and monitoring. To reduce the dominant computational cost of retrieving focusing functions of Marchenko and UD-RM methods, the thesis further develops deep-learning-based acceleration strategies. Self-supervised learning and an optimization-based inverse solver are used to speed up the estimation of focusing functions while preserving imaging quality. Incorporating spatial position information, time masking, and balanced amplitude handling yields substantial efficiency gains in both synthetic and field tests without compromising the final images. The second contribution addresses SWD, where continuous drill-bit vibrations act as a seismic source and are recorded at the surface during drilling, offering major cost advantages and near-real-time potential. This thesis proposes an SWD workflow based on multi-dimensional deconvolution (MDD) to transform drill-bit illumination into equivalent virtual surface sources and to retrieve virtual reflection responses that can eliminate the effect of the unknown source signature and surface-related multiples. Two practical challenges are addressed: (i) identifying and attenuating the direct-arrival component using a data-driven global-optimization strategy, and (ii) making MDD feasible for long continuous recordings through segmentation and the summation of correlation estimates. The resulting method is validated on both synthetic and field datasets. Finally, robustness is enhanced through a plug-and-play regularized MDD framework that integrates a pre-trained deep-learning denoiser within the inversion process, producing cleaner retrieved responses and more reliable imaging. Overall, the methods developed in this thesis expand artifact-reduced imaging capabilities under acquisition-limited conditions and unconventional source scenarios. They provide practical routes to more reliable subsurface characterization and monitoring for emerging energy applications, strengthening the contribution of geophysics to the energy transition.