Measuring Poverty from Space

How do we track economic progress and identify pockets of poverty? Often, the only way is to ask.

Census data is crucial to measuring national population, housing counts and even agriculture, business, and traffic. These are supplemented by household surveys that collect income or expenditure data. However, the process of manually acquiring this type of information is expensive and time consuming and is often most inaccurate precisely in the parts of the world most in need of international assistance. A recent World Bank study on data deprivation found that during the period from 2002 and 2011, 57 countries produced one or less poverty estimate. Surveys are processed slowly and poverty data often is released 12 to 18 months after it was collected. But census data is fraught in countries where violence makes data collection a dangerous mission.

Could we generate more frequent and timely data if we complement traditional surveys with satellite data?

The collection of objective and quantitative data is crucial in locating the poor and formulating policies to help them. To explore whether and how satellite data can improve traditional methods of poverty measurement, the World Bank has created a partnership with Orbital Insight, a geospatial analysis company that tracks socioeconomic trends at global, regional, and hyper-local scales.

Orbital Insight looks down from space and trains computers to measure population and indicators of economic growth from satellite images. Together with the World Bank, Orbital Insight will be testing the effectiveness of machine-analyzed satellite images for humanitarian data collection through deep learning artificial intelligence. If successful, this data could provide an important augmentation to census data in determining which areas are developing more slowly and whether localized interventions are effective.

A satellite image before and after analysis with the areas in red identified as agricultural land by algorithms.

This is Orbital Insight's second exploratory project of its kind, following a partnership with the World Resources Institute to not only track but also predict deforestation. If proven, the techniques that come out of the project have the potential to be applied across the globe.

Much faster than humans going from door-to-door, computers can scan thousands of pictures, counting any number of attributes. Orbital will be applying several of its successful deep learning applications such as car counting and shadow tracking to investigate the use of satellite images to measure poverty at the level of individual villages. Longer term, Orbital's technology could scale up poverty measurements at a countrywide, regional and even worldwide basis.

In a developing country, there are many visual measures of economic health that can best be seen from space. For instance, comparing satellite pictures of a city from 2010 to 2015, Orbital Insight's artificial intelligence might count more cars on the roads, which would suggest an increase in the income and prosperity of the area. Other indications of growth can come from monitoring the rate at which new buildings are constructed, roads are built, or cultivated land spreads.

With thousands of satellites orbiting our planet every day, new images are beaming down at an increasing rate, facilitating real-time monitoring of the world at large. With this information, the World Bank can find out what kind of aid succeeds in changing conditions in any given area.

The project will start in Sri Lanka, where pockets of poverty remain despite a brisk recovery following the end of the 25 year civil war and the 2004 Indian Ocean tsunami. This partnership's mission is to understand which satellite indicators best predict poverty estimates for small areas derived from census and survey data. The results will show how satellite imagery analysis can complement and enhance census-based techniques to help governments and nonprofit organizations make better decisions about resource allocation and the effectiveness of aid.