This paper investigates the use of semi-automatic classification techniques applied to multispectral satellite imagery for the mapping and recognition of historical landscape structures. The research focuses on the Tratturo Magno, a traditional transhumance route, aiming to evaluate different image-processing approaches to detect its path and spatial influence within the territory.
The methodology integrates vegetation indices (NDVI, EVI), unsupervised clustering, and supervised machine learning classification (Random Forest), combined with GIS tools and Google Earth Engine for temporal and spatial analysis. Multitemporal Sentinel-2 imagery is analyzed to identify optimal seasonal conditions for feature detection, highlighting the tratturo as a linear green corridor during autumn and winter. Results show that supervised classification methods provide the most reliable identification, although challenges remain in distinguishing the tratturo from similar landscape elements. The study demonstrates the potential of AI-supported geospatial analysis to enhance landscape interpretation, while emphasizing the need for further refinement to achieve precise and unique feature extraction.