Advancing species-level vegetation mapping of alpine peat bogs using UAV imagery, machine learning and texture...

Alpine peat bogs are valuable, highly heterogeneous wetland ecosystems with specialized and often rare species, where spatial variation in species composition reflects fine-scale hydrological and ecological gradients. They play a key role as natural water and carbon reservoirs and contribute to landscape resilience under climate change, yet are highly vulnerable to drainage, droughts, and chemical changes, leading to vegetation shifts and biodiversity loss. Reliable species-level vegetation monitoring is therefore essential for understanding these dynamics and supporting conservation and management. This study evaluates the potential of UAV-based multispectral imagery combined with machine learning and texture features for species-level vegetation mapping in alpine peat bogs. Two representative peat bogs in the Krkonoše Mountains - the Úpské and Hraniční Meadow Peat Bogs, located within a Ramsar-listed wetland - were surveyed, with high-precision ground-truth data collected for 14 vegetation classes. Four classifiers (Random Forest, Gradient Boosting, XGBoost, and Multi-Layer Perceptron) were compared, and texture features derived from GLCM, GLDM, and GLSZM matrices were systematically assessed. XGBoost achieved the highest accuracy (weighted and macro F1-scores up to 94% and 79% respectively), while ensemble approaches enabled spatial consistency mapping. By jointly evaluating classifiers, texture features, and multi-temporal data, the study provides a comprehensive assessment of factors influencing species-level classification accuracy. Integrating texture features and multi-temporal UAV data significantly improves classification performance and enables detailed mapping of species composition, which serves as an indicator of ecosystem dynamics and associated environmental conditions. The proposed approach supports high-resolution, long-term monitoring of alpine peatland vegetation and improves the detection of fine-scale changes, contributing to more effective conservation and management of these vulnerable ecosystems.