In many respects, regional gravity field modeling in polar areas is of special, and growing, interest. On the one hand, high-resolution and precise geoid models are an important input parameter to investigate and quantify manifold geophysical, oceanographical and glaciological phenomena, e.g., the determination of the mean dynamic ocean topography, or the application of the hydrostatic equilibrium condition in the areas of ice shelves, sea ice, or subglacial lakes. Moreover, geoid models are in general needed as a reference for height models. On the other hand, because of the unavoidable polar data gaps in satellite measurements due to the inclination (the so-called “polar gap”), terrestrial gravity data are indispensable also for global gravity field modeling. However, the available terrestrial (ground-based) gravity datasets, in particular of Antarctica, are very sparse and heterogeneous. For example, over the Antarctic continent the true resolution of even the most recent combined global geopotential models such as EGM2008 or EIGEN-6C only corresponds to that of the satellite-only models derived from GRACE and GOCE, respectively. Furthermore, standard techniques of regional geoid modeling cannot be readily used in this area. Apart from the heterogeneity of the data as a practical challenge the additional density contrast implied by the covering sheet needs to be accounted for from the theoretical point of view. This complex situation is the starting point for the present cumulative dissertation. Whereas the individual publications present the results of selected regional case studies, the intention of the following summary is to draw an integrated picture aiming at explaining the geophysical phenomena as both applications and influencing factors in the context of regional geoid modeling. Using the example of the Weddell Sea it is shown how sea-ice coverage affects the quality and reliability of marine gravity field models derived from radar satellite altimetry. At present, these models are the only input data to the high-resolution global geopotential models. At the same time, the refined regional model and geostrophic velocities derived thereof are employed to demonstrate how even sparse and heterogeneous terrestrial gravity data may provide a contribution to simultaneously calibrate and unify the available datasets. As a result, near the coast differences at the order of some decimeters could be observed in comparison with EGM2008, originating partly from systematic effects and noise in the global model. In the continental areas, its omission error even yields a standard deviation of 0.75 m and attains a maximum of more than 3 m. Another refined and, owing to appropriate input data, very precise and highly resolving geoid model is derived for the region around subglacial Lake Vostok. In combination with ice-surface heights and ice thickness data it is used to provide observational evidence that the lake is in a state of hydrostatic equilibrium. There, the additional geoid signal w.r.t. GOCE is a bit smaller (0.56 m standard deviation). However, considering the residual deviations of the apparent lake level (0.26 m standard deviation) the significant and necessary, as compared to the resolution of GOCE, contribution of a regional geoid model to this application is shown. In a strict sense, the relevant quantity to evaluate the hydrostatic equilibrium condition of a subglacial lake is the actual geopotential at the anticipated lake level. Its computation requires a downward continuation of the disturbing potential inside the topography, which is closely related to the concept of the well-known geoid-quasigeoid separation term. In the past, this term was frequently described as an approximation by means of the Bouguer anomaly. However, considering the modern requirements of the “one-centimeter geoid” this approximation may be too coarse over rough terrain. Following recent works in this field, a generalized yet refined approach for practical implementation of the term is developed. The individual constituents of the term are quantified. In particular, their sensitivity against the radius up to which topography is taken into account is investigated. For this simulation study, the Himalaya mountain region served as test area. Furthermore, special focus is given to the indirect of topography on the potential which, contrary to applying a planar model, does not vanish in the spherical approach.