Comparison of standardized methods (object-oriented vs. per pixel) to extract the urban built-up area: example of lubumbashi (DRC)

The delimitation of urban areas, and in Africa in particular, remains a challenging issue and is an important one for city management in a context of rapid urban growth. The objective of this study is to delineate the morphological limits of the urbanized city which is different from administrative boundaries. Two methods have been developed: the first one is using object-oriented classification (using eCognition software) while the second is based on a automated and customizable spectral threshold (by generating a large number of thresholding which are then combined to determine a probability of belonging to urban area) using GRASS (free GIS software). The main goal is to test if the classification procedure could be avoided (training and validation are time-consuming processes). Both methods are applied to 4 different dates, using SPOT-1 (1986), SPOT-2 (1995), ASTER (2001) and SPOT-5 (2006) images to test the reproducibility and robustness at different dates, and by using different sensors. The results show that on one hand the two methods produce fairly similar results for a given date, both from a standpoint of quantity and quality. On the other hand, for all dates, the results are consistent and very good (overall accuracy between 88% and 95%).