Research PaperMeasuring intra-urban poverty using land cover and texture metrics derived from remote sensing data
Introduction
The majority of the global population today is urban. The percentage of urban dwellers increased from 43% in 1990 to 52% in 2011, and it is expected to grow to 67% by 2050 (United Nations, 2007, United Nations, 2008, United Nations, 2012). All population growth from 2011 to 2050 is expected to be absorbed by urban areas, and most of this growth will occur in cities of less developed regions (United Nations, 2012). In developing countries, rapid urban growth normally exceeds the capacity for local governments to deliver services and infrastructure, which increases urban poverty and intra-urban inequalities (Duque, Royuela, & Noreña, 2013).
The monitoring of poverty is a key issue for policy makers because it can help prevent poverty traps and crime nests and allocate public investments where they are needed most (Duque et al., 2013). Urban poverty is a multidimensional phenomenon; as such, there are many ways to measure it. These measures usually include information from at least one of the following dimensions: income/consumption, health/education, and housing (Carr-Hill and Chalmers-Dixon, 2005, Moser, 1998). They are computed from survey or census data, which are quite expensive, time consuming, less frequently produced, and often statistically significant for spatial units that are too large to capture the intra-urban variability of phenomena. This last feature creates inference problems such as the ecological fallacy (Baud et al., 2010, Robinson, 1950) or aggregation bias (Fotheringham and Wong, 1991, Paelinck and Klaassen, 1979).
This study works toward overcoming these problems by exploring the possibility of using remote sensing imagery to measure urban poverty. This proposal is based on the premise that the physical appearance of a human settlement is a reflection of the society in which it was created and on the assumption that people living in urban areas with similar physical housing conditions have similar social and demographic characteristics (Jain, 2008, Taubenböck et al., 2009). The main advantage of using remote sensing imagery for urban poverty quantification is that this type of data can be obtained faster, at higher frequencies, and for a fraction of the cost required for field surveys and censuses. Poverty mapping usually follows two types of approaches: the expenditure-based econometric approach linked to a poverty line used by World Bank, and the value-focused approach used by United Nations Development Programme (UNDP) based on the Human Development Index (Baud, Pfeffer, Sridharan, & Nainan, 2009). The Index of Multiple Deprivations (Baud, Sridharan, & Pfeffer, 2008), the Slum Index (Weeks, Hill, Stow, Getis, & Fugate, 2007), and the Slum Severity Index (Patel, Koizumi, & Crooks, 2014) all follow the value-focused approach that integrates several dimensions of deprivation in one single measure.
We chose the Slum Index to corroborate this possibility because this measure is based on the physical aspects of dwelling units. A slum household is defined as a group of individuals living under the same roof in an urban area that lacks one or more of the following: durable housing of a permanent nature, sufficient living space (not more than three people sharing the same room), easy access to safe water at sufficient amounts and at an affordable price, access to adequate sanitation in the form of a private or public toilet shared by a reasonable number of people, and security of tenure (UN-Habitat, 2006). Weeks et al. (2007) presented the calculation of the Slum Index from census and survey data as the sum of the fractions of households that lack one or more of the five conditions mentioned above. The value can range from 0, meaning that no slum-like households are present in an area, to 5, where all households in an area lack all five of the features defined by UN-Habitat. The proportion of slum dwellers in cities is strongly correlated with the Human Development Index, which integrates three development indicators: per capita GDP, longevity and educational attainment (UN-Habitat, 2003). Thus, the presence of slums in a city is an indicator of poverty, and the Slum Index is a good proxy variable for urban poverty at the intra-urban level. This paper implements spatial econometric models using data from Medellin (one of the most unequal cities in the world) to assess whether the Slum Index can be estimated using image-derived measures.
Weeks et al. (2007) and Stoler et al. (2012) used land cover descriptors and texture measures from medium to very high spatial resolution satellite imagery to develop spatial econometric models for predicting the Slum Index as a function of remote sensing-derived variables. This work builds on these previous studies by analyzing a wider set of remote sensing variables on land cover composition, image texture and urban layout spatial pattern descriptors to provide empirical evidence that either supports or refutes the hypothesis that remote sensing could be used to estimate the Slum Index at the intra-urban scale. As our intention was to lower the costs of this approach as much as possible, we use data drawn from an RGB composition of a Quickbird scene with a spatial resolution of 0.60 m captured in May of 2008. The imagery is similar in color and spatial resolution to Google Earth and Microsoft Bing images (Quickbird is a commercial Earth-observation satellite that collects very high spatial resolution -VHR- imagery). Although the conclusions of this exercise may not be valid worldwide, we seek to present new, innovative and low-cost means of measuring urban poverty.
The structure of this paper is as follows: Section 2 describes the spatial unit of analysis, the socioeconomic data for Slum Index calculation, the remote sensing data and variables derived from it and the statistical analysis for model specification. The results are presented in Section 3, and the subsequent discussion is presented in Section 4. Section 5 presents the main conclusions and public policy implications of this line of research for local governments and authorities.
Section snippets
Spatial unit of analysis
Located in the northwestern Colombia (Fig. 1), Medellin is the second largest city in Colombia with a population of 2.4 million (DANE, 2012). The urban area of Medellin has two levels of administrative spatial units: communes (16) and neighborhoods (243). The Slum Index is typically reported at the commune level from socioeconomic data available in the Quality of Life Survey (whose sampling process is designed to be representative at this spatial scale). There are two main disadvantages to
Results
The principal components factor analysis of image structure and texture variables resulted in four retained factors with eigenvalues above 1. These four factors account for 95% of the variance for the original 21 variables. Table 4 shows the rotated factor analysis results. Factor 1 accounts for 57% of the variance among these variables. It was labeled overall complexity (OC) as it captures most of the structural variables with the exception of RSF and VFM as well as most of the texture
Discussions
We found spillover effects regarding the Slum Index in Medellin, which is in agreement with previous works that have found the same effect for different poverty measures (Duncan et al., 2012, Holt, 2007, Okwi et al., 2007, Orford, 2004, Sowunmi et al., 2012, Voss et al., 2006). The results of the statistical analysis indicate that the most important remote sensing predictors of the Slum Index at the analytical region level for Medellin include the percentage of impervious surfaces, the fraction
Conclusions
This paper seeks to estimate the Slum Index for one Latin American city solely using remote sensing data. The usefulness of remote sensing data for estimating the Slum Index at the intra-urban level was previously tested in the case of Accra, Ghana. The findings of the present work corroborate these earlier results in a city with a different geographical and economic setting. In this work, we tested a wider set of remote sensing variables related not only to land cover composition but also to
Acknowledgements
This research was made possible by funding from EAFIT University (EAFIT-342-000033; EAFIT-513-000084) and Medellin City Hall Enlaza-Mundos program. The authors thank the anonymous reviewers and Professor Gustavo Canavire for their insightful observations and suggestions. The usual disclaimer applies.
References (55)
- et al.
Definition of a comprehensive set of texture semivariogram features and their evaluation for object-oriented image classification
Computers & Geosciences
(2010) - et al.
Using semivariogram indices to analyse heterogeneity in spatial patterns in remotely sensed images
Computers & Geosciences
(2013) - et al.
Understanding heterogeneity in metropolitan India: The added value of remote sensing data for analyzing sub-standard residential areas
International Journal of Applied Earth Observation and Geoinformation
(2010) - et al.
Matching deprivation mapping to urban governance in three Indian mega-cities
Habitat International
(2009) - et al.
Specification searches in spatial econometrics: The relevance of Hendry's methodology
Regional Science and Urban Economics
(2003) - et al.
Civil conflict and forced migration: The micro determinants and welfare losses of displacement in Colombia
World Development
(2008) Remote sensing application for property tax evaluation
International Journal of Applied Earth Observation and Geoinformation
(2008)- et al.
Texture-based identification of urban slums in Hyderabad, India using remote sensing data
Applied Geography
(2012) The asset vulnerability framework: reassessing urban poverty reduction strategies
World Development
(1998)Identifying and comparing changes in the spatial concentrations of urban poverty and affluence: A case study of inner London
Computers, Environment and Urban Systems
(2004)