Imagery obtained from unmanned aerial vehicle (UAV) is widely used for land surface modelling. Recent research prove that digital elevation models (DEMs) created from UAV imagery are characterized by a high rate of accuracy and reliability. Most of these studies are focused on assessing absolute elevation accuracy of the UAV DEMs, but the accuracy of relative elevations (i.e., accuracy of reproducing of local elevation differences within DEM) also should be considered. In this paper, we focus on the precision of replicating relative elevations in DEMs derived from imagery captured via UAVs without precise coordinate reference. To evaluate this accuracy, we use datasets of aerial images processed in two different methods: one with on-board coordinates obtained from a GNSS receiver, and the other based on precise coordinates calculated with the Post-Processing Kinematic (PPK) method. The sites selected for assessment are not look like each other in terms of terrain and forest cover characteristics to track the difference of modelling in the divergent areas. Constructed DEMs were compared with reference fragments of global DEMs by the statistical indices for the difference fields. The findings indicate that the absence of an accurate coordinate reference does not have a substantial impact on the precision of reproducing relative elevations in the DEM. This makes it possible to use UAV materials without precise coordinate reference for modelling in most geographical studies, where the error of terrain steepness values of 0.9° can be considered acceptable.