Individual Tree Species Classification in a Mediterranean Ecosystem using UAV-Acquired Multispectral Images and Machine Learning

This study investigates the classification of individual tree and shrub species within a complex Mediterranean ecosystem using UAV-acquired multispectral imagery and machine learning techniques. Conducted on Culuccia Island (Sardinia, Italy), the research integrates high-resolution photogrammetric products with Object-Based Image Analysis (OBIA) and a Random Forest classifier to delineate and identify vegetation species at the crown level. A total of 272 geo-referenced samples from 16 species were used to train and validate the model, which achieved an overall accuracy of 0.71. The workflow demonstrates the effectiveness of single tree segmentation in highly heterogeneous environments and highlights the potential of phenology-informed feature sets for improving classification. The results underscore the value of UAV-based methods in conservation monitoring, ecological assessment, and habitat management. Future research directions include the integration of LiDAR data, deep learning architectures, and multitemporal observations to enhance scalability and model interpretability.