In this paper the accuracy of 6 spatial interpolation methods on geoid grid models are tested and compared: Inverse Distance Weighting, Kriging, Minimum Curvature, Natural Neighbour, Radial Basis Function and Moving Average. Grid models are derived from three global geopotential models: EGM2008, EIGEN-6C4 and GECO. Totally 18 geoid models in forms of equiangular grid models are created in GoldenSoftware Surfer 13 for this purpose, by combining 6 methods of interpolation and 3 global models, whose size and position cover a wide area of Croatia with parts of neighbouring countries. Default interpolation parameters were kept during the process. Models were created in a way that modelled values of geoid undulations can be subtracted from undulations of independent models in equal points. Residuals calculated in this way served as a base for quality assessment. Comparative analysis of quality indicators, statistics and graphical indicators of residuals led to a conclusion that the best-fitted methods of interpolation for this specific case are Radial Basis Function, Kriging, and Natural Neighbour, in that order.

Key words: geoid, geoid model, spatial interpolation, global geopotential model, interpolation, Croatia