Geospatial Modeling of Carbon Stocks in Forest Litter in the Republic of Karelia and on the Karelian Isthmus (Leningrad Oblast)

A regional geospatial model of the variability of carbon stocks in forest litters has been constructed, and environmental factors, influencing soil organic carbon accumulation, have been assessed for the Republic of Karelia and the Karelian Isthmus, Leningrad Oblast. The modeling is based on 137 terrain samples taken in 2007–2010 within the framework of the International Co-operative Program on Assessment and Monitoring of Air Pollution Effects on Forests (ICP Forests). Spatial predictors, characterizing soil formation factors, are used for modeling soil organic carbon stocks according to the SCORPAN model. The set of independent predictors was formed based on the correlation analysis and the exclusion of multicollinear variables. Random Forest machine learning algorithm was applied for regression modeling. The resulting carbon stock model explains 46% of the variability in carbon stocks in the study area (R2 = 0.46; MAE = 1.84; RMSE = 2.59). The mean carbon stock in the forest litter is 3.9 kg/m2 with a minimum of 1.4 kg/m2 and a maximum of 7.4 kg/m2. It is shown that the most significant factors of the variability of carbon stocks in the forest litter in Karelia and on the Karelian Isthmus include climate (37.2%), spatial position (22.6%), and vegetation (17.9%).