Monitoring changes in land cover induced by atmospheric pollution in the Kola Peninsula, Russia, using Landsat-MSS data

We present the results of hybrid unsupervised-supervised classification of a series of Landsat-MSS images, spanning the period from 1978 to 1992, to study the impact of SO2 emissions from the nickel smelter at Monchegorsk (67 degrees 55'N, 32 degrees 50' E) in the Kola Peninsula, Russia, on adjacent boreal forest and upland (lichen-dominated) tundra vegetation. Ground truth data were collected from a 2500 km(2) area during airborne and surface field campaigns in 1994 and 1995, and used to classify the 1992 image into 56 different surface types, including a characterisation of the level of vegetation damage. The pre-1992 images could not be classified by transferring the spectral signatures from the 1992 image, mainly as a result of phenological differences. Instead, they were classified using spatial context and a set of observationally-derived botanical rules governing the types of allowable land-cover change. A comparison of the classified images was performed by further combining the land-cover classes into groups representing forest areas with varying proportions of canopy damage, and upland tundra areas with varying degrees of lichen cover and damage level. Quantitative comparative results were obtained for a 22 225 km(2) area common to all images after 1978. Although damage levels were already significant by 1980, our results show that most of the increase in vegetation damage since 1980 has in fact occurred since 1989. We attribute this to a change in the local meteorological conditions. The method developed in this paper has the merit of revealing areas of upland tundra vegetation showing early effects of SO2-induced damage. It should thus have widespread applicability to the delineation and monitoring of areas of industrially affected arctic vegetation, especially in the former Soviet Union, where many such areas occur, often in regions where access is severely limited. However, it is likely that held validation will continue to be required, since the mechanism by which increasing damage levels are depicted in MSS images of lichen tundra areas remains unclear.