A new combination of spectral indices derived from Sentinel-2 to enhance built-up mapping accuracy of cities in semi-arid land

Abstract

Accurate built-up extraction is important to land use planning. However, in semi-arid and arid environments, the accurate discrimination between bare soil and built-up area is challenging, due to their high spectral similarity. For that reason, the combination method of spectral indices was adopted from Sentinel-2 data to enhance built-up mapping of Ras El-Oued city (North-East Algeria). The spectral indices selected to be combined are mainly: the Normalized Difference Tillage Index (NDTI) and the Built-up Area Index (BAI) for built-up detection, and additionally, the Modified Bare Soil Index (MBI) for bare land extraction. Therefore, four combinations were developed and binarized via the Otsu algorithm to provide an automatic built-up mapping. The findings showed that the BAI index works better than the NDTI index in dry climates, since their overall accuracy (Oa) is about 92.00% and 86.33%, respectively. In contrast, the built-up mapping accuracy enhancement is noticed, when using the four combinations compared to the indices (NDTI and BAI); Com₁ (NDTI + MBI) and Com₂ (NDTI – BAI) have an identical (Oa) which is 93.00%. As for both Com₃ (MBI – BAI) and Com₄ (NDTI + MBI) – BAI), they produced approximately the same result, since they achieved an (Oa) which is 94.00% and 94.33%, respectively. Therefore, the four datasets created have revealed their positive behavior toward built-up detection in this area of semi-arid land, where both Com₃ and Com₄ were the best. The research results could, therefore, be suitable for mapping the cities in dry climates for better development in the future.