Potential field geophysics enhancement using contemporary deep learning

This thesis reports on three bodies of work across two topics in geophysics; enhancing the high-frequency content of gridded potential field data using super-resolution; and learning a representative function of a potential field extent using implicit neural representation. The first topic enhances the value of geophysical surveys by predicting high-frequency components in sparsely sampled potential-field grids. The second presents a straightforward neural network framework for high-quality grid regularisation, processing, and data integration. Each body of work includes a real-world geophysical case to demonstrate the effectiveness and challenges of the method with open-access data.