Object-oriented mapping of urban poverty and deprivation.

The eradication of urban poverty is a major global concern, stressed also by the Millennium Development Goals (MDG’s). In developing cities where urban poverty is increasingly concentrated, localizing areas of urban poverty and deprivation, while understanding their heterogeneity and developmental dynamics, is a major challenge for local authorities, Such understanding is a critical prerequisite for targeting interventions. The scarcity of relevant information, coupled with constraints in both human and financial resources for extensive field data collection, stimulates an evaluation of the utility of remote sensing tools to extract poverty relevant information.
In this paper, we discuss a conceptual frame of mapping spatial/physical variables, such as dwelling size, building density, green spaces, and lack of proper road network, of areas of urban poverty, while also capturing aspects of their factual heterogeneity. Conventional approaches to poverty mapping often rely on sample (household) data which are aggregated to administrative units for analysis, thus tend to neglect their spatial heterogeneity.

Using as a case study a selection of deprived versus non deprived wards in Delhi, India, we test the utility of the developed conceptual frame for objected based information extraction on a high resolution Ikonos image. Delhi, as most Indian cities, is characterized by fast developing slum areas scattered all over the municipal corporation areas, the speed of development (some areas can be occupied and settled within a couple of days) calls for fast and reliable tools to better monitor their development. The object-oriented approach allows a contextual characterisation of different settlement types, as well as the use of spatial metrics.

The results of mapping areas of urban poverty using objected based information extraction in combination with spatial metrics are compared with poverty information derived from census data of 2001, in particular with the Multiple Deprivation Index, recently developed by other researchers. The results clearly show that poverty is not spatially uniform within the selected wards, and hence demonstrate how data aggregation can hide spatial variation of poverty. Compared to traditional pixel-based image analysis methods, the strength of object-oriented processing lies in its ability to integrate context, multi-type data (both image and thematic) and a reasoning approach similar to that of an experienced analyst evaluating images visually. In a time series approach the object based approach further allows a detailed description of the type and, if multi-temporal data were to available the rate of change, both of which are critical but impractical to obtain with traditional, in particular field-based mapping methods.