Boreal peatlands are wet ecosystems in which partly decomposed plant material accumulated in thick peat deposits. These deposits are rich in carbon and store more carbon than the Amazon rainforest. As a result of climate change, however, this carbon may be released in the atmosphere, accelerating climate change worldwide. Understanding the functioning of boreal peatlands, and specifically the peat-forming vegetation inhabiting these ecosystems, is of vital importance in predicting climate change impact on these carbon stores of global importance.
For many applications, such as estimating peat carbon stocks, forestry, and nature conservation it is important to identify different peatland vegetation units. Peatland vegetation is generally organized in a pattern of alternating dry elevated places (hummocks) and wet depressions (hollows). Digital aerial imagery is an excellent data source to classify peatland vegetation in hummocks and hollows. Additionally, the improved availability of raw LiDAR data unlocks new classification possibilities by combining both data sources.
Many classification procedures exist: either pixel-based or object-based image analysis, and ranging from maximum likelihood to advanced machine learning techniques like artificial neural networks, random forest. So far, it is unknown which classification method is suitable to classify vegetation of a set of boreal peatlands.
Your research topic
In this thesis you will develop, use and compare a broad range of classification algorithms to classify high-resolution aerial images of boreal peatlands. But contact us if you have additional ideas you would like to test!
Students: 1 or more
Required: GRS-20306 Remote sensing, ArcGIS. Experience with eCognition software and/or scripting in R or Python are valuable
Duration: 24 ECTS or more
Period: by mutual agreement
Supervision: Jelmer Nijp (SGL), Lammert Kooistra (GRS), Arnaud Temme (SGL)