CLUSTER ANALYSIS FOR LANDSCAPE TYPOLOGY

Abstract

This paper deals with the classification of landscapes by means of cluster analysis, having as a theoretical basis a typology of landscapes, through the notion of zonality by L. S. Berg and the theory of geosystems by V. B. Sochava. Firstly landscape mapping was performed using geoprocessing techniques resulting in 272 landscape units for the municipality of Mineiros (Goiás State, Brazil). This units were defined by union of different elements of morphostructures, lithology, landforms, altitude, slope degree, drainage density, soils and land use. The objective of this work is to compare different forms of grouping to establish the typology of landscapes using an upscaling/bottom-up approach. The similarity coefficient Jaccard, Euclidean Distance metric and k-means algorithm were evaluated. Even though the field validations and statistical tests point to a different scenario, it was considered that the Jaccard and Euclidean Distance metrics presented satisfactory scenarios for the representation of the landscapes. The grouping was important in the optimization of processes, although there is a need to differentiate between statistical and spatial significance. Therefore, the relevance of this technique is in the collaboration to group and redefine a large amount of information that, by means of manual analysis and spatial regrouping, would present excessive delay.

Keywords:
Geosystems; Landscape taxonomy; Landscape mapping; Geoinformation, Cluster