Extraction of high-precision built-up areas from SENTINEL-2B imagery via multi-index approach and fuzzy C-means algorithm

Abstract

Accurate Urban built-up area information is required for a wide range of applications, in particular risk prevention and city planning. However, the extraction of built-up urban areas using high spatial resolution multispectral image, such as Sentinel-2, remains a significant challenge due to spectral confusion and intra-urban variability with other types of land cover, especially between built-up areas and bare land. As a result, in this work, we aim to increase built up accuracy mapping for Tan-Tan city (Southern of Morocco) by using six spectral indices, including Normalized Difference Building Index (NDBI), New Built Up (NBI), and Normalized Difference Tillage Index (NDTI) for urban area, and Normalized Difference Vegetation Index: NDVI, linked to vegetation, as well as the Bare Soil Index (BSI) and Dry Bare-Soil Index (DBSI) for bare land by means of Fuzzy C Means (FCM) algorithm. The six spectral indices were extracted from Sentinel-2 during the dry season and were combined to generate six multi-index datasets. Herein empirical results show that DBSI index works with NDBI, while BSI works better with NDTI. Therefore, the two multi-index datasets DBSI/ NDVI / NDBI and BSI/ NDVI / NDTI were suitable for built-up extraction in dry season in preferring order. Their overall accuracies were 85.28%, and 83.99%, respectively.