Digital elevation model with regular grids simplifies the modeling and analyzing processes. However, it fails to adapt to different terrain shapes and it is hard to obtain high surface accuracy and small dataset in the meantime, which limits its application in areas such as vehicle dynamic simulation. To solve this problem, a modeling method with varying data resolution was proposed. The node density varied according to the local surface curvature so that it ensures high data resolution around sharp changing areas and low data resolution around relatively flat area. Then Delaunay gridding method together with cubic interpolation was applied to obtain a digital elevation model with high surface accuracy and small data size using the resulting irregular node set. The proposed method was testified with a complex terrain surface used in rover simulation, and compared with the traditional regular grid model, which verified its advantages in achieving the two major objectives. 