A comparison of GIS landscape modeling and traditional landscape mapping (by the example of the Elbrus region)

The theory of natural landscapes is one of the central and most complex concepts of modern physical geography. As is well known, in Western science, the concept of “landscape” is recognized only as a general one and is usually used to designate geosystems that have been exposed to anthropogenic influence for a long time. In this regard, geoinformation modeling all over the world (outside Russia) in recent decades has been mainly devoted to obtaining the so-called “landscape cover” Landuse-Landcover, which represents some kind of land use types, fragments of cultural landscape and urbanized areas mixture. Attempts at geoinformation modeling aimed at delimiting territorial natural complexes in the West and developing predictive maps of vegetation, soil cover and “habitats” are similar in content and algorithms to the approaches used for semi-automated mapping of natural landscapes. The development of synthetic geoinformation modeling methods was largely associated with overcoming the theoretical difficulties and controversial “plots” of Russian landscape science, which include ideas about the role of the morpholithogenic basis and biota of the landscape, taking into account the “leading” factors of differentiation, the presence of objective spatial hierarchical levels of landscape differentiation, and others. In this article, using the example of a key area of the Elbrus Region National Park, the capabilities of the traditional technique of expert-manual mapping are compared with mapping in a geoinformation environment. It is shown that the intuitive actions taken by an expert drawing a landscape map, although not strictly algorithmic in reality, are nevertheless close in content to complex variants of cluster analysis and decision trees. It is substantiated that the best option for landscape synthesis is not an overlay of finite classes of the morpholithogenic base and biota, but a joint analysis (cluster or isocluster classification) of many initial variables, in particular, geomorphometric parameters and landscape-vegetation indices. Supervised classifications with the creation of training files based on the author’s manual landscape maps give the worst result compared to uncontrolled ones, which, firstly, indicates the inaccuracy of the drawn maps, and secondly, the authors’ failure to comply with any strict algorithms and phenomena, which may be labeled as “changing the rules on the fly”.