Horizontal resolution, one of dominant variables for grid based Digital Elevation Model (DEM), directly determines
topographic expression, the accuracy of terrain parameters and geosciences simulations based on DEM. Cell size is
determined on relationship between resolution and terrain parameters traditionally, without taking terrain variance
information content of the raw data into account. This paper puts forward two methods for suitable DEM horizontal
resolution by mining the input contour data based on geostatistics. One is a direct method considering internal and
external variance. Regularization variables of serial resolutions are calculated from the sampled data by regularization
theory in geostatistics. After the variance comparison between point and serial resolutions, the grid at which the external
variance between adjacent grids is larger than average internal variance in grid is named suitable resolution. The other
method, which combines macro-topographical variance with micro-topographical variance, is an indirect way. Various
large-scale supports and their regularization variables are made by dividing the sampled data using regularization theory.
In order to ascertain an optimal support size to express macroscopic spatial variance of terrain, semivariograms of
regularization variables are analyzed on various support sizes. Estimation of the optimal bin size that can estimate the
probability density function in non-parametric density estimation is referred to decide the microcosmic appropriate
resolution in the optimal support. Both methods were experimented in practice, and gave relatively consistent results.
The latter was commended considering the computational efficiency.