A cascaded data fusion approach for extracting the rooftops of buildings in heterogeneous urban fabric using high spatial resolution satellite imagery and elevation data

Abstract

In this study, a cascaded data fusion approach was implemented to extract the rooftops of buildings from remote sensing data in a heterogeneous urban fabric in Irbid city, Jordan. The rooftops of buildings were extracted from Pleiades-B1 very high spatial resolution imagery (i.e., 2 m) along with LiDAR-derived normalized digital surface model (nDSM) data and normalized difference vegetation index (NDVI). The imagery from Pleiades-B1 satellite was classified using support vector machine (SVM), then a normalized digital surface model (nDSM) was calculated to generate the heights map. Then, the trees and vegetation cover were filtered using NDVI in which different threshold values were tested to achieve the best identification of the rooftops of buildings. The results of the SVM, nDSM, and NDVI maps were combined to obtain one layer representing the rooftops of buildings in the study area. Finally, the results were evaluated against reference data obtained from the municipality of Irbid and ground surveying. The correctness and completeness measures of the detected buildings’ footprints were 0.956, 0.854 and 0.902 for the precision, recall, and F-score respectively. The total area of the extracted rooftops of buildings was approximately 6.2 km2 when compared to the area of the registered buildings in the municipality of Irbid (i.e., 5.9 km2) in 2021. The results proved the possibility of using the implemented method in mapping the rooftops of buildings in such a heterogeneous urban fabric in the study area.