Application of Multi-Spectral Index from Sentinel-2 Data for Extracting Build-up Land of Hanoi Area in the Dry Season

Abstract

A remote sensing index is a simple and effective way to highlight a specific land cover. Therefore, in this study, we try to increase
the accuracy of the urban land map developed for Hanoi city by focusing on determining the appropriate combination of spectral
indices calculated from satellite image data. To conduct the study, four spectral indices were selected including namely normalized
difference tillage index (NDTI), bare soil index (BSI), dry bare soil index (DBSI) and the normalized difference vegetation index
(NDVI). All these spectral indices are calculated from Sentinel-2 data acquired in the dry season. The two combinations are created
from the superposition of NDTI/BSI/NDVI and NDTI/DBSI/NDVI spectral index layers. The use of the “K-means” algorithm as
an unsupervised classifier provides rapid and automatic urban land detection. The results show that the BSI index performs better
than using the DBSI index. As a result, the BSI index brings improvements: bare soil types and accumulation processes are better
differentiated, with overall accuracy increasing by 5.82% and Kappa coefficient increasing by 11.1%. The results show that the NDTI/
BSI/NDVI multi-spectral index dataset is suitable for mapping urban areas with the potential to help better urban management
during the dry season.