Topography on land and bathymetry, its underwater depth equivalent, belong to the most fundamental attributes of the solid earth's surface. Over two thirds of the earth is covered by water, with about 90% of this area lying more than 1000m below the sea surface. In contrast to the land area, most of the deep sea remains largely unexplored and to date the topography of the Moon or Mars is much better known than the bathymetry of large parts of our own planet.Deep sea ocean mapping can directly be carried out with ship-bound echo sounders or indirectly through a remote sensing method known as satellite altimetry. Modern echo sounding technology allows for high resolution mapping with unsurpassed accuracy. Due to the vastness of the oceans, however, even after decades of mapping activity, the oceans are far from completely surveyed, and the echo soundings accumulated over the time with different, meanwhile evolving technologies are of highly varying quality. Satellite altimetry, on the contrary, provides virtually complete coverage of the entire globe, although the achieved resolution and accuracy is limited. For the compilation of consistent, ocean spanning Digital Bathymetric Models (DBMs) from raw depth measurements, an appropriate data basis is therefore a heterogeneous mixture of historical and contemporary echo soundings, complemented by satellite altimetry as needed. The North Atlantic is by far the best mapped of all oceans and as such it provides an ideal area to study scientific problems related to ocean mapping and DBM compilation. The heterogeneity and size of the global bathymetric data basis require powerful solutions to handle and process both data and metadata effectively. In this work, a spatial relational database in combination with a geographical information system (GIS) form a flexible tool kit for a DBM compilation, and a data model for the storage and retrieval of both data and metadata is developed. In a case study I show the potential of the available sounding data in the North Atlantic to derive a DBM with significant improvements over the models commonly used today. Many geoscientific applications require that data sets are sampled on a regularly spaced grid, notwithstanding the fact that data acquisition often provides measurements at irregular positions and with incomplete coverage.Several methods exist for interpolating and gridding raw data to obtain gapless grids. In ocean mapping, minimum curvature bicubic splines in tension are a commonly used approach. This work presents a refined technique, multiple resolution splines in tension. The method takes the local data density into consideration during the gridding process, in order to reduce gridding artifacts mainly caused by very inhomogeneous data coverage. It is shown that multiple resolution splines in tension allow for a high maximum grid resolution, without introducing artifacts that appear with regular splines in tension interpolation at the same resolution.