Applying Multi-Index Approach from Sentinel-2 Imagery to Extract Urban Areas in Dry Season (Semi-Arid Land in North East Algeria)

The mapping of urban areas mostly presents a big difficulty, particularly, in arid and semi-arid environments. For that reason, in this research, we expect to increase built up accuracy mapping for Bordj Bou Arreridj city in semi-arid regions (North-East Algeria) by focusing on the identification of appropriate combination of the remotely sensed spectral indices. The study applies the 'k-means' classifier. In this regard, four spectral indexes were selected, namely normalized difference tillage index (NDTI) for built-up, and both bare soil index (BSI) and dry bare-soil index (DBSI), which are related to bare soil, as well as the normalized difference vegetation index (NOVI). All previous spectral indices mentioned were derived from Sentinel-2 data acquired during the dry season. Two combinations of them were generated using layer stack process, keeping both of NDTI and NDVI index constant in both combinations so that the multi-index NDTI/BSI/NDVI was the first single dataset combination, and the multi-index NDTI/DBSI/NDVI as the second component. The results show that BSI index works better with WTI index compared to the use of DBSI index. Therefore, BSI index provides improvements: bare soil classes and builtup were better discriminated, where the overall accuracy increased by 5.67% and the kappa coefficient increased by 12.05%. The use of k-means as unsupervised classifier provides an automatic and a rapid urban area detection. Therefore, the multi-index dataset NDTI/BSI/NDVI was suitable for mapping the cities in dry climate, and could provide a better urban management and future remote sensing applications in semi-arid areas particularly.