The process of geometric correction of a satellite image requires a network of ground control points (GCPs), the density of which depends on the accuracy required. In this study, an approach to determine the distribution of the GCPs was developed based on dividing the surface terrain into zones of low- and high-height variances. Comparative statistics were investigated to summarize the optimum number and distribution of the GCPs. For a uniform arrangement of 10 GCPs, the accuracy using root mean square error was 7.19 m. This accuracy was improved to 6.13 m following the inclusion of just two additional GCPs in the zone of high-surface height variance. Thus, the transformation using the polynomial model and a set of GCPs, for which the surface variance in height was considered, resulted in greater accuracy than using the conventional uniformly distributed method. In addition, the time required for the trial-and-error selection of the locations of the GCPs was reduced. Our results suggest that a method that considers the variance in height of the surface terrain could be applied to various types of images such as satellite or aerial photography.