An Assessment of the Accuracy of Triangulated Irregular Networks (TINs) and Lattices in ARC/INFO

Abstract

This paper evaluates the accuracy of triangulated irregular networks (TINs) and lattices in ARC/INFO. The results of an empirical comparison of the two models over two selected topographic sites are presented. Both vector and raster data were used to build the models. Three pairs of models were constructed based on 1,600, 4,000, and 9,000 sample points for the study area of the State Botanical Garden of Athens, Georgia, and 400, 800, and 1,600 sample points for the study area of Lake Lucerne of Wisconsin. The two models were assessed based on the same number of input sample points. Overall, TINs performed better than lattices. The quality of lattices decreased more dramatically than that of TINs when the number of sample points used for the construction of the models decreased. With an increase in the number of sample points used, the difference in performance between the two models decreased. The results of the evaluation directly depend on the comparison criteria and modeling algorithms. The evaluation is slightly sensitive to test indices used and the distribution of test points. The spatial pattern of residuals on spot heights was quite different from that on randomly selected test points. Users should choose models carefully based on the purpose of their application, the accuracy required, and the computer resources that are available.