Tree Species Clasification in Natura 2000, Austria

Tree species mapping is essential for ecosystem services estimation, policymaking, and forest management. In this study, a Random Forest machine learning model was used to classify tree species based on multitemporal Sentinel-2 (S2) imagery within a Natura 2000 riparian site in Austria. Utilizing field-collected reference data, the potential of S2 timeseries was assessed to map native riparian tree species in the Danube-Auen National Park. Six cloud-free S2 scenes were combined into a 60-band multitemporal stack image used to capture phenological patterns crucial for tree species differentiation. An overall classification accuracy above 80% has been achieved, with individual species' accuracy ranging from 30% to 90%. The findings highlight the advantages of multitemporal S2 analyses over single date classifications and emphasize their relevance for habitat mapping and conservation efforts in heterogeneous riparian forests. The proposed approach provides a cost-efficient method to improve existing non-validated habitat maps in Natura 2000 sites, offering a replicable framework for similar ecological studies.