MAPPING OF SOIL ORGANIC CARBON CONTENT AND STOCKS AT THE REGIONAL AND LOCAL LEVELS: THE ANALYSIS OF MODERN METHODOLOGICAL PPROACHES

This paper provides an overview of scientific publications in Russia and other countries devoted to the soil organic carbon (SOC) content and stocks mapping at the regional and local levels. The analysis showed that the cartographic assessment of the SOC content and stocks was conducted using various approaches chosen depending on the multiple factors: the size of the territory (continental, national, regional, local levels); the cartographic basis availability (maps of soil types, landscapes, and vegetation formations, remote sensing data, etc.) and laboratory and field survey findings. Two main approaches were generally used for SOC content and stocks mapping: (1) based on available thematic maps; (2) digital soil mapping. The review also provides a set of spatial data that characterize the soil forming factors according to the SCORPAN model, which is widely used in digital soil mapping. Spatial terrain data was one of the most commonly used predictors, followed by the vegetation and climate variables. The mapping accuracy significantly increased by adding spatial data on classification units of the soils to the spatial data models. The authors of the publications noted that the climate variables had a significant effect on the spatial variation of the SOC content and stocks at the regional level, while at the local level the influence of climatic variables was less significant. The analysis showed that the most common methods used in digital mapping were machine learning algorithms, among which the Random Forest method often showed the best results. The plotted maps were cross-validated almost in all studies. Tests of the maps’ accuracy using an external independent validation dataset were rare, although this was the most important stage of digital soil mapping. R was the most popular software used for modeling the SOC content and stocks. SAGA GIS, QGIS, ArcGIS, and the cloud platform Google Earth Engine were most commonly used to prepare predictors.