Multispectral Sentinel-2 and SAR Sentinel-1 Integration for Automatic Land Cover Classification

• Main Authors: Paolo De Fioravante, Tania Luti, Alice Cavalli, Chiara Giuliani, Pasquale Dichicco, Marco Marchetti, Gherardo Chirici, Luca Congedo, Michele Munafò
• Format: Article
• Language: English
• Published: MDPI AG 2021-06-01
• Series: Land
• Subjects: land monitoring; sentinel; Copernicus; change detection; EAGLE concept; land consumption
• Online Access: https://www.mdpi.com/2073-445X/10/6/611

The study of land cover and land use dynamics are fundamental to understanding the radical changes that human activity is causing locally and globally and to analyse the continuous metamorphosis of landscape. In Europe, the Copernicus Program offers numerous territorial monitoring tools to users and decision makers, such as Sentinel data. This research aims at developing and implementing a land cover mapping and change detection methodology through the classification of Copernicus Sentinel-1 and Sentinel-2 satellite data. The goal is to create a versatile and economically sustainable algorithm capable of rapidly processing large amounts of data, allowing the creation of national-scale products with high spatial resolution and update frequency for operational purposes. Great attention was paid to compatibility with the main activities planned in the near future at the national and European level. In this sense, a land cover classification system consistent with the European specifications of the EAGLE group has been adopted. The methodology involves the definition of distinct sets of decision rules for each of the land cover macro-classes and for the land cover change classes. The classification refers to pixels’ spectral and backscatter characteristics, exploiting the main multi-temporal indices while proposing two new ones: the NDCI to distinguish between broad-leaved and needle-leaved trees, and the Burned Index (BI) to identify burned areas. This activity allowed for the production of a land cover map for 2018 and the change detection related to forest disturbances and land consumption for 2017–2018, reaching an overall accuracy of 83%.