Large scale carbon mapping of forest-steppe ecotones using multispectral satellite data

Environmental changes, such as altered precipitation patterns and temperatures, but also the type of
land management, have strong impact on the vegetation structure and the associated soil carbon
storage. Vulnerable ecosystems that have always grown at the limits of system stability have small
resilience and therefore respond to the smallest changes. This is also true for the forest-steppe
ecotones at the southern border of the Mongolian taiga, with their two subtypes of light and dark
taiga. Due to drought stress, this forest only grows on the northern slopes, where the rate
evapotranspiration is smaller. Climatic change, which is very pronounced in these highly continental
areas, leads to water scarcity and thus to higher drought stress as well as an increased risk of forest
fires. In the forest-steppe ecotone in northern Mongolia, light taiga dominated by Betula is increasingly
spreading into areas previously covered by dark taiga representing coniferous forests dominated by
Pinus and Larix. Since soil organic carbon stocks are known to be related to vegetation, in this study
we aimed at assessing the spatial carbon stocks distribution of different forest-steppe ecotones
characterized by different tree compositions by using a multispectral satellite image approach. Based
on Sentinel-2 data, a supervised random forest classification was carried out using the MSAVI index
and carbon stocks from 50 soil profiles of these sites as training data. For the first time, a mean of
multi-year MSAVI was used to compensate the temporal gap between the actual image of vegetation
vitality and the comparatively inert soil organic carbon. The results were validated by ground truthing
on further 36 soil profile measurements. The validation confirmed the accuracy of the classification
and thus led to a valid area calculation. The map based on the measurement results, which was created
by the use of machine learning, illustrates that the significant differences in the spatial distribution of
the taiga subtypes and their soil organic carbon stocks balance each other out in the areas under
consideration. Since the resulting map could be validated by both soil investigations and field survey
experiences, we assume that the applied remote sensing method can be used as a basis for a realistic
area monitoring of the ecosystem under consideration to calculate the spatial change of the carbon
pool.