The spatial dynamics of poverty in South Africa: Assessing poverty using satellite technology and poverty maps

  • 0

Abstract

Poverty is antithetical to human well-being. Attempts are, therefore, continuously made to identify and quantify areas of poverty so that aid and developmental resources can be targeted most effectively and efficiently to assist populations living in poverty. Poverty maps, consequently, assume great importance. This article focuses on the spatial nature and dynamics of poverty in South Africa, as derived from the nightlight and infrared emissions sensed by American defence satellites. The spatial distribution of poverty has not yet been adequately studied, in contrast to the majority of poverty studies that focus on education, employment, income and health.

Within a South African context, the World Bank notes that poverty has a strong spatial dimension. Poverty, like many socio-economic problems, is a function of space, and spatial analysis can be a key to deeper understanding regarding the nature and dynamics of poverty. Geography and location have considerable explanatory power for the understanding of poverty.

This study primarily follows the methodology of other researchers, designing a granular poverty map for South Africa that indicates where the need is greatest. There have been promising signs in recent years that novel sources of high-resolution data can provide an accurate and up-to-date indication of living conditions. In particular, research illustrates the potential of features derived from remote sensing by satellites and geographic information system data (GIS).

The current study first creates spatially disaggregated 1 km2 resolution maps of the population in poverty. Secondly, spatially disaggregated data sources, i.e. the night-time stable lights and infrared satellite images along with the LandScan population grid, are employed for this purpose. Satellite images have the ability to update spatially disaggregated poverty maps annually, semi-annually, but also immediately in real time.

The use of satellite imagery and data addresses two fundamental constraints in poverty mapping thus far. It makes it possible to map poverty on a very detailed granular level and on a continuous basis, and also in much more detail. The primary data sources used are the LandScan global population distribution datasets for 2010 and 2015 augmented by the Socioeconomic Data and Applications Centre (SEDAC, 2020) of the National Aeronautics and Space Administration’s (NASA) Gridded Population of the World dataset for 2020 and the Visible Infrared Imaging Radiometer Suite (VIIRS) night-time dataset for 2015 and 2020, augmented by the Operational Linescan System (OLS) night-time dataset for 2010. These data are freely available on the internet website of the Colorado School of Mines.

Given the availability of the gridded population and night-time lights imagery and data from 2000 to 2020, it is possible to develop granular poverty maps for South Africa over these 20 years. For this study, the poverty map will, however, cover three periods, i.e. 2010, 2015 and 2020. Although it is possible to develop maps at the 1 km2 level, this study focuses on a 35 km2 level. This is because some grids do not align perfectly between the various datasets and years at lower levels. To ensure a more reliable alignment between the grids of the various datasets and years, the 1 km2 grids were aggregated to 35 km2 grids. This significantly improves the alignment of the various grids, which is important for comparative reasons within and between data areas.

A poverty index is also developed to assist in the creation of a colour ramp image that is adjusted for lower values in abundantly lit areas where economic activity is high. High poverty index values occur in areas with a high population count and dim (or no) lighting as detected by the VIIRS. The normality test and normal probability plots indicate that none of the poverty indexes are normally distributed, suggesting a significant unequal spatial distribution of poverty in South Africa and that this unequal distribution has not changed much over the ten years. Spatial distribution of poverty is therefore not randomly distributed, suggesting spatial dependence and/or spatial relevance. The Moran I-statistic had a value of 0,49 in 2010, decreasing to 0,45 in 2020. The associated z-values for the three periods suggest a strong rejection of the null hypothesis of spatial randomness.

In general, it can be stated that clusters of high values can be classified as hot spots with many people living in poverty and clusters of low values as cold spots with lower poverty levels. High poverty locations (high clusters) are concentrated in the north-eastern, eastern and south-eastern parts of the country, while the low poverty locations are concentrated in the south-western, western and central parts of the country. In 2010, there were 3 423 identified high-high cluster locations compared to 3 436 and 3 438 in 2015 and 2020, respectively. This represents on average around 7,5% of the total (46 194) 35 km2 locations or about 10% of the total land surface of South Africa. In contrast, the number of low-low locations (clusters) represents approximately 25% of the total 35 km2 locations or about 33% of the total land surface of South Africa.

The results also show that the locations of poverty have not changed much over the ten years. During the five years 2015 to 2020, there were around 1 447 locations that experienced high increases in poverty, while only 645 locations experienced low increases in poverty (i.e. a decrease in the number of poverty locations). The vast majority of these high and low locations were also statistically significant.

The results of the granular poverty maps developed in this research study suggest that poverty in South Africa is spatially clustered and not randomly distributed. Locations with high and low poverty are generally not in close proximity. Cross-referencing the derived granular poverty maps with previous studies and day-time imagery suggests that it is indeed possible to identify areas or locations with high or low poverty levels using the poverty index estimated using both the LandScan gridded population and VIIRS gridded night-time light imagery and data.

The essence of the spatial mapping of poverty at a granular level is that “space matters indeed”, in the sense that the poverty levels in one location are related to what happens in neighbouring locations. Not only do spatial relationships exist, but poverty seems to be fundamentally related to unique locations.

Keywords: light intensity; Moran I-statistics; night lights; population; poverty; remote sensing; satellite images

 

Lees die volledige artikel in Afrikaans

Assessering van die ruimtelike aard van armoede in Suid-Afrika met behulp van satelliettegnologie

  • 0

Reageer

Jou e-posadres sal nie gepubliseer word nie. Kommentaar is onderhewig aan moderering.


 

Top