A Combined PCA-SIs Classification Approach for Delineating Built-up Area from Remote Sensing Data

Abstract

The aim of this study is to develop a method for delineating built-up areas based on remote sensing data. The proposed method evaluated 13 spectral indices (SIs) commonly used in assessing land use and land cover (LULC) and selected meaningful indices through a principle component analysis (PCA) and spectral separability analysis. These indices are combined into a built-up delineation index set (BDIS). The development was implemented at the example of the built-up area in Qena city, Egypt. The method was evaluated against ground truth data and one recently developed global product using confusion matrix statistics. The BDIS was computed from indices showing a high loading of each one of the most relevant principle components and high separability at the same time. Subsequently, the selected indices, i.e., the transformed difference vegetation index (TDVI), band ratio for a built-up area (BRBA), and a new built-up area index (NBI), was used as input variables for the supervised classification procedures. The results show an increase in the accuracy of the built-up area delineation using BDIS. The overall, producer’s, user’s accuracies, and Kappa coefficient were 96.3%, 96%, 93%, and 0.946, respectively. The results and a comparison with the global human settlement layer provided by the European Joint Research Center also verified the usefulness of the proposed method for utilizing Landsat 8 OLI imagery data in delineating a built-up area, providing a comprehensive view on built-up area at the local scale.