Evaluation of Spectral Built-up Indices for Impervious Surface Extraction Using Sentinel-2A MSI Imageries: A Case of Addis Ababa City
No Thumbnail Available
Date
2021-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
An urban area extraction with a high degree of accuracy is important for a variety of applications, especially urban planning, natural resource management, and disaster prediction. In recent years, extraction of an impervious surface using spectral built-up indices has been extensively explored. However, a comprehensive evaluation of various built-up indices from high-resolution satellite images is still lacking. Hence, this study examines the performance of seven spectral built-up indices in the classification and change detection of impervious surfaces using sentinel-2A MSI imageries in Addis Ababa city. It includes Built-up Area Extraction Index (BAEI), Band Ratio for Built-up Area (BRBA), Modified Built-up Index (MBI), Normalized Built-up Area Index (NBAI), New Built-up Index (NBI), Normalized Difference Built-up Index (NDBI), and Urban Index (UI). All the built-up indices maps were classified into built-up and non-built-up areas and evaluated based on histogram overlap and statistical methods, namely spectral discrimination index (SDI). Simultaneously, a machine learning method called support vector machine (SVM) was employed to classify the imageries into five classes: bare land, built-up, forest, vegetation, and water bodies. The finding of the study indicated that NBAI, NBI, and NDBI have the highest SDI value (1.24, 1.23, and 1.54), (1.06, 1.08, and 1.23), and (1.16, 1.1, and 1.26) for the years 2016, 2018, and 2020, respectively. However, the other spectral built-up indices show unsatisfactory results for the study area. The LULC changes between 2016 and 2020 showed that the built-up area, bare land, and water bodies were increased by 9,084.5 ha, 813 ha, and 2 ha whereas vegetation and forest areas were declined by 9,279.3 ha and 620.2 ha respectively. According to this study, NBAI, NBI, and NDBI were more robust built-up indices with kappa coefficient, (90%, 87%, and 0.81%), (86%, 85%, and 80%) and (92%, 93%, and 86%) for the years 2016, 2018, and 2020, respectively. Also, the overall classification accuracy using SVM is 81%, 86%, and 82% for 2016, 2018, and 2020, respectively and it reveals that this algorism has potential in LULC classification using high-resolution satellite images. Therefore, this study concludes spectral built-up indices show a promising result in the extraction of impervious surfaces.
Description
Keywords
Impervious Surface, Machine Learning Algorithms, Sentinel-2A Imagery, Spectral Built-up Indices, Spectral Discrimination Index, Support Vector Machine