Soil type and content of macro-elements determine hotspots of Cu and Ni accumulation in soils of subarctic industrial barren: inference from a cascade machine learning

Aerial technogenic pollution from the activity of ferrous and non-ferrous metallurgy resulting in degradation of vulnerable natural ecosystems is a principal environmental problem in Russian Arctic. The industrial barren in the vicinity of Monchegorsk (Kola Peninsula) has been forming since 1950-s in the impact zone of the copper-nickel smelter. Soil heterogeneity, complete or partial degradation of vegetation, and rugged terrain intensified by soil erosion result in complex lateral spatial redistribution patterns of aerial depositions of Cu and Ni emitted by the smelter. In this research, we applied cascade machine learning (gradient boosting machines) to quantitatively describe these patterns. An extensive soil sampling campaign (n=506) across an area of 343 ha has revealed an extremely high levels of contamination (max bulk concentrations of Cu and Ni - 29.87 and 30.12 g/kg). We showed that soil types and the content of macro-elements (Ca and Fe) mapped based on the conventional set of predictors (topography, hydrology, landscape’ spectral properties) explained spatial variability and especially hotspots of Cu and Ni contents with a higher accuracy compared to the models where interactions between macro-elements and heavy metals are not considered. This approach is a promising tool for mapping heavy metals’ distribution in eroded, degraded, and highly polluted areas, which can be very useful to support land reclamation plans and allocate bioremediation measures.