Evaluatiing the NDLI's Performance for Identifying Water Surface Using Sentinel-2 MSI Data

Abstract

Land cover and land use (LULC) are the key determinant factors that influence the regional climate. In this study, we present LULC classification for the Taipei City, Taiwan based on Sentinel-2B image acquired in 2018. A recently-proposed Nomalized Difference Laten Heat Index (NDLI) and Nomalized Difference Vegetation Index (NDVI) are ultilized and compared to derive LULC, in particular, water bodies. Validation is based on reference datasets collected from Google Earth and field survey. Overall accuracices of classification are about 76% for NDLI and 91% for NDVI. However, it is shown that NDLI is highly capable to distinguish the water bodies from the others, such as built-up and bareland with accuracies of 100% and 95%, respectively, while NDVI shows better perfomance on vegetation classificantion only. In addition, it is found that shortwave infrared (SWIR)-2 (band 12) is more sensitive to identify the water bodies in comparison to SWIR-1 (band 11) of Sentinel-2B image to compute NDLI for extracting water bodies. This result further demonstrates that NDLI can be used as an effective indicator to detect and map the water surface or built-up or bareland by using Sentinel-2 imagery as initially suggested by Liou et al. [1].