Methodological proposal for the construction and selection of digital high-precision elevation models

Abstract

Digital Elevation Models improve accuracy in terrain analysis. This work compared different interpolation methods using statistical criteria in a topographic survey of 339 points raised with a total station in the “San Juan de Lagunillas” area (Venezuela). The existence of a significant trend process with spatial autocorrelation modeled by an isotropic Gaussian semivariogram was documented. The kriging allowed to obtain residuals less than a minimum tolerable error equal to 0.1 m. The residual kriging was the best evaluated in cross validation with exclusion (absolute mean error of 0.71 m). The disjunctive kriging showed a lower error average but a wider error range and the highest mean absolute error (12.8 m). The other kriging showed mean error variance greater than 1.3 m and absolute minimum errors greater than 0.85 m. Due to the high dependence on the model specification form, graphics are recommended for a better fit.