17 October marks the International Day for the Eradication of Poverty. While eradicating poverty is one of the top goals on the UN Sustainable Development Agenda, scarcity of reliable data hampers efforts to measure progress toward this goal. Space technology – especially when combined with mobile technology – has emerged as an effective poverty measuring tool, complementing and updating estimates from traditional methods. This blog demonstrates how integrating geospatial and mobile data into traditional mapping methods can provide novel insights into the spatial distribution of poverty. This analysis follows two previous blogs in the series exploring opportunities and challenges of frontier technologies in low- and middle-income countries (LMICs).
Measuring poverty with accuracy comes with challenges in LMICs
Measuring and understanding the impact of poverty is critical, especially in LMICs, where a large proportion of the population lives below the poverty line. However, assessing poverty can be complex, and requires the integration of multiple factors beyond the lack of income, such as access to education, sanitation, housing and nutrition. The complex combination of factors makes it challenging for governments to identify the groups most in need and design appropriate poverty mitigating policies.
Challenges exist in traditional poverty measurement methods in LMICs, with surveys and census generally available less frequently (every five to 10 years) and usually representative of large components of the population. Given the rapid nature of socioeconomic change, additional information is required between census periods to monitor socioeconomic indicators. Moreover, household surveys are an expensive exercise, with an average price of $322.99 per household in Sub-Saharan Africa as of 2014.
Poverty mapping provides a detailed description of the spatial distribution of poverty and inequality within a country, combining individual and household survey data and population census data to estimate welfare indicators. This type of mapping can foster greater awareness of poverty issues in a country, carrying critical information about the different poverty conditions that underlie the national average. The presentation as a map summarises a large volume of data concisely and facilitates the interpretation of that data by preserving the spatial relationships among different areas. As such, poverty maps are a vital source of information for actions to reduce poverty.
Satellite and mobile data can fill the poverty mapping gap
Satellite data is generated via remote sensing technologies, providing reliable and timely information about the Earth’s surface. Earth Observation through satellites relies on the use of their attached payloads, which collect and transmit images or other data of the Earth’s characteristics. The Earth Observation satellites are launched to relatively low altitudes, capturing image reflecting indicators of (or lack of) economic activities, urban areas, access to water, roads or electricity.
Most commercial high-resolution imagery comes from satellites operated by Digiglobe, GeoEye and ImageSat International. A large amount of satellite imagery is freely accessible to NGOs, researchers and start-ups through Google Earth (GE) or NASA’s Earthdata Search. CISGeography offers a list of major satellite imagery providers that can be accessed for free.
Satellite data provides accurate information about living conditions in rural areas. However, limitations exist in terms of densely populated areas. Therefore, combining mobile and satellite data can lead to accurate poverty predictions. Mobile data and networks provide information on individuals’ economic opportunities, carrying information such as levels of data usage or data linked to travelling. Furthermore, anonymised data on monthly mobile credit consumption, and the proportion of people in an area using them, can indicate household access to financial resources. This helps build a more accurate picture of poverty than traditional methods, when combined with the satellite-enabled data.
A case in point, NGO Flowminder Foundation, researches the potential of mobile and satellite data in poverty mapping. The organisation has modelled three traditional poverty measures by using aggregate data from mobile operators and widely available geospatial data. Models combining these data sources provide the best predictive power at granularity level. The findings indicate the possibility to estimate and continually monitor poverty rates at high spatial resolution in countries with limited capacity to support traditional methods of data collection.
Machine Learning (ML) algorithms provide effective tools for analysing satellite imagery, enabling the delivery of sharper insights on poverty measures. Machine Learning algorithms aggregate satellite data and data collected from other sources, facilitating the human interpretation. Algorithms enable us to scan a large amount of satellite imagery data and to develop models to extract certain features, detect changes and predict future situations with the help of Artificial Intelligence (AI).
Technology company, NVIDIA, in collaboration with Standford University’s research, leverages Google Earth images into a statistical poverty mapping model, using ML and survey data. Major NVIDIA mapping projects in Nigeria, Malawi and Rwanda are based on satellite images combined with Google Earth Images. The ML model learns to identify daytime satellite features associated with poverty from datasets such as roads, farmland and bodies of water, in order to achieve accurate poverty mapping (Figure 1).
Figure 1: Predicted poverty probabilities using satellite imagery and survey data
(left: fine-grained level (10km × 10km block level), middle: at the district-level; right: 2005 survey results for comparison). Source: World Resources Institute.
How can we bring satellite-enabled mapping to the next level in LMICs?
The increased availability of free satellite data, along with costs of satellite technology driven down by new technological developments and collaborations between local, national, regional and international stakeholders, have the potential to further drive the adoption of satellite measurements. However, in the endeavour of poverty mapping to establish partnerships with mobile operators, satellite imagery and digital mapping providers, it is essential to ensure that the right resource is allocated to the right issue.
The full development of the technology in LMICs is challenged by the lack of awareness of the benefits of space technologies, limited financial resources and technology and skills gaps in developing, using and adapting space technologies. Developing policies for open data and open science for geospatial data and leveraging public–private cooperation on common objectives to achieve the UN Sustainable Development Goals (SDGs) is critical. Likewise, the international community should build data capacities through training.
Satellite and mobile mapping methods cannot replace traditional methods such as census but can complement it successfully. The next stage will be to build mapping models leveraging AI and big data to help governments and organisations predict poverty and its future impact – one further step towards the goal of improving living conditions in LMICs.
The GSMA Mobile for Development (M4D) Central Insights Unit is an initiative supported by the UK Foreign, Commonwealth & Development Office (FCDO), the GSMA and its members.