Browsing by Subject "multivariate modelling"

Boreal soils are vital carbon reservoirs for the Earth, playing a crucial role in the global carbon cycle. The boreal landscape is a combination of small-scale variation of forests and wetlands with widely varying total carbon stocks. To understand these carbon dynamics, it is important to examine the factors that drive the C stock sizes. This thesis examines the variability of soil organic carbon, the environmental factors that explain it and the reliability of different modelling methods for SOC variation. The data consists of environmental data produced by the BioGeoClimate Modelling Lab, 129 soil samples collected from the study area, and their annealing results. SOC stocks have been modelled using three different methods: Generalized Linear Models, Generalized Additive Models, Generalized Boosted Methods. The results of the study show that the main factors explaining the variation in soil carbon are soil moisture and the presence of wetlands, and to a moderate extent slope, potential incoming solar radiation and canopy height. The estimated total carbon stock of the 130 km² study area is estimated to be about 1 858 620 016.3 Kg, with the largest stocks located in wetlands and GAM having the best predictive performance. Although the amount of carbon is not directly comparable to other studies, the results show similarities in the spatial distribution of carbon compared to other studies.