Assessing the impact of input data incongruity in selected quantitative methods for modelling natural landscape typologies

With supervised classification methods, we can determine classification rules for landscape types of existing landscape typologies. In this article, we analyse whether supervised classification methods could also define adequate rules for landscape types determination in the case of poorly designed typologies. We tried to model two Slovenian intentionally distorted natural landscape typologies. We noted that due to the incongruity of the distorted typologies, decision tree methods were not capable of forming rules for determination of landscape types. Although we did manage to create modelled distorted typologies with minimum distance to means method, maximum likelihood method, and k -nearest neighbours method, they matched the basic distorted typology only slightly. Z metodami nadzorovane klasifikacije lahko za obstojece naravnopokrajinske tipizacije dolocimo klasifikacijska pravila za posamezne pokrajinske tipe. V prispevku razpravljamo, ali bi tudi v primeru zelo slabo zasnovanih tipizacij z metodami nadzorovane klasifikacije lahko izdelali dovolj natancna pravila za dolocanje pokrajinskih tipov. Poskusili smo modelirati dve namenoma popaceni naravnopokrajinski tipizaciji Slovenije. Opazili smo, da zaradi nesmiselnosti popacenih tipizacij metode odlocitvenih dreves sploh niso bile sposobne izdelati pravil za dolocanje pokrajinskih tipov. Z metodami najmanjse razdalje, najvecje verjetnosti in k najbližjih sosedov pa smo sicer uspeli izdelati modelirane popacene tipizacije, a so se te z osnovno popaceno tipizacijo le malo ujemale.