全球变化知识资源中心

The taiga-tundra ecotone (TTE) extends over 13.400 km around the northern hemisphere and marks the transition zone between taiga and tundra. Since this area is vulnerable to climate change, human activities and natural disturbances such as wildfires, monitoring of this region is extremely important. Dealing with the characterization of this transition zone, very few studies are available on a global scale. Existing approaches lack the thorough characterization of the ecotone as they capture just a part of the complex transition zone.

The objective of this research is to develop a monitoring concept for detecting the taiga-tundra interface for a designated study area in northern Siberia. We used optical and SAR remote sensing data products from 2007 to 2010, whereby the spatial resolution of all datasets were unified to 1 km. Earth Observation data and products from ALOS PALSAR, MODIS, AMSR-E were utilized in addition to digital elevation model data and the Percent Tree Cover product. Using random forest variable importance measure, most relevant statistics were selected and used for different model setups. The final random forest model setup included statistics of land surface temperature (LST), albedo, NDVI, EVI, Vegetation Continuous Fields (VCF), PALSAR HV-Polarization and LAI. Snow water equivalent (SWE) statistics had to be excluded for compensating artifacts.

Since sharp distinctions between forest and tundra do not occur in nature, class probability values were computed. This allowed the unique allocation per class for each grid cell. The model setup showed high performance values, which were cross validated, with accuracies of 95.6% ( +/- 1.5%). The model result was validated using high resolution Google Earth imagery, where an overall map accuracy of 93% (Kappa 0.89) was achieved. The assessment of the input data products revealed that statistics of LST, SWE and albedo were most important.

Compared to existing circumpolar products the advantage of this approach is the derivation of continuous probability values for the taiga-tundra transition zone instead of sharp borders between the biogeographic classes. The model provides the basis for a monitoring effort aiming at detecting changes over time. This would help to understand the dynamics of ecosystem changes with respect to influencing factors.