Enhanced Map Composition and Diachronic Land Cover Classification of Landsat Data in Google Earth Engine

Understanding how habitats in the European Natura 2000 network change over time and space is crucial to evaluating the effectiveness of protective measures and developing sustainable management practices. Satellite remote sensing through Earth observation offers a cost-effective, timely, reproducible vegetation analysis. This study aims to analyze the changes in grassland habitats (types 6210 and 6230*, Annex I, European Directive 92/43/EEC) within the Natura 2000 network’s areas in Umbria (Central Italy) between 2000 and 2020. The goal is to gather spatial information to develop more sustainable planning and management strategies. In this direction, we selected a three-year time window in two periods (1999–2001 and 2019–2021). We applied an enhanced dataset composition, including cloud-filtering, topography correction, and textural analysis on Landsat 7 and 8 surface reflectance images available in Google Earth Engine (GEE). Additional morphometric features were derived from the digital elevation model of the area. Using machine learning classification (Random Forest – RF), we obtained the land cover maps for the two periods under investigation with high overall accuracy (87 and 88%). Only shrublands showed a lower classification accuracy due to the varied nature of this land cover class and the limited resolution of Landsat data. On this basis, we identified transformations that occurred in the study areas. The results confirmed the effectiveness of enhanced map composition and RF in GEE for diachronic land cover classification. The final spatial information can help identify areas where conservation measures related to the Natura 2000 network have been more or less effective and develop more appropriate management strategies for grasslands of European concern.