Abstract

ASTER DEM data was used to automate landform classification during soil survey in the Varamin area. For comparison, manual landform classification was done in the same area. Study area was located at South of Jajrood river watershed, Southeast of Tehran province (Iran). The main purpose of this study was to compare the effect of automated and manual landform classification methods in semi-detailed soil survey procedure. Eight geomorphometric parameters were extracted from DEM using the TAS and DiGem software. The Pearson correlation coefficient analysis elucidated that, the most effective of parameters were: analytical hill-shade, plan and profile curvature, and slope and divergence-convergence index. In addition to these terrain attributes, principal component analyses (PCA) of primary geomorphometric parameters were produced to increase the quality of classification and to reduce modeled data. First three PCAs cover 97% of variance of the data. These PCAs and mentioned terrain parameters were selected for performing of K-means unsupervised landform classification model. Results indicated that unsupervised and manual classification can be complemented, such that conflation of the final maps obtained by these methods can produce a more accurate map. Also, the K-means algorithm with correct iterations, tolerance and suitable number of classes can be used for automated landform classification as well. Hybrid landform classification method is useful for soil survey and soil mapping especially, in watersheds and natural resource fields.

Key words: Hybrid landform classification, geomorphometric parameters, K-means classifier, Pearson correlation coefficient.