Leveraging Commercial Applications to Help the World Bank Map Poverty
Census and household survey data are often combined by economists to generate local estimates of poverty and economic activity, in order to target aid to those who need it the most. But the process of manually acquiring this type of information is expensive and time consuming and censuses are typically only conducted once every ten years.
For this reason, the World Bank and Orbital Insight partnered to research whether at-scale geospatial analysis of satellite imagery could replace census data to generate more timely local poverty estimates.
About the World Bank & the Current Challenge of Poverty Estimates:
The World Bank is a vital source of financial and technical assistance to developing countries around the world. For decades, the World Bank and the development community have relied heavily on data from household surveys to understand poverty and economic activity. At least 70 countries have combined household and census data to generate these poverty maps, to help policymakers focus interventions on the areas of greatest need. However, since economic conditions change quickly and census data is typically collected only ten years, these valuable maps can quickly become dated. Waiting for census and survey results causes delays of any analysis dependent on the data.
How Orbital Insight Contributed:
The World Bank, in partnership with Orbital Insight, undertook a project to explore how well geospatial features extracted from daytime satellite imagery could substitute for census data to generate local estimates of poverty in Sri Lanka. Orbital Insight automated the quantification and contextualization of several geospatial features, including:
- Buildings (mapping developed areas)
- Building height (estimating building height using building shadows)
- Car density (counting cars)
- Agriculture (mapping agricultural land)
The theory was that all of the “signals” above would provide useful indicators of economic activity and poverty at the local level. For example, mapping buildings and building height indicates development and construction and therefore, may be considered a proxy for population growth and migration. Cars can indicate where particularly neighborhoods are more prosperous. Agricultural analysis reveals information about food health and productivity and may help observers gain a deeper understanding of both rural and urban areas.
Orbital Insight acquires geospatial imagery for analysis from its growing ecosystem of satellite partners and has cultivated the largest virtual constellation of satellites in the world. Proprietary deep learning algorithms identify objects and patterns by mining massive amounts of satellite imagery. Recent developments in computer vision allow for classification of all kinds of objects, including buildings, cars, planes, trains, ships, water, land and much more. Cloud computing enables Orbital Insight to ingest and process imagery at scale, analyzing millions of images simultaneously.
The results of Orbital Insight’s collaboration with the World Bank indicate that analysis of satellite imagery has great potential for explaining poverty across countries. Satellite indicators alone explain about 60 percent of the variation in local poverty rates, in areas of a few square kilometers, and about 65 percent of the variation in the welfare of households living in those areas—the exact measure of welfare in this case is the logarithm of average household per capita consumption.
Of the “signals” and algorithms listed above, building density was the leading correlate of poverty in both urban and rural areas. Car counts were strongly correlated with poverty in urban areas. Mapping agricultural land (land classification) was weakly correlated with poverty, however, looking at the health of crops (through a technique called Normalized Difference Vegetation Index analysis) proved helpful with poorer areas having lusher vegetation. Interestingly, in rural areas, vegetation is associated with greater poverty, but the reverse is true in urban areas, as lusher vegetation indicates luxury such as public parks, private gardens, or lawns.
In their paper Poverty from Space: Using High Resolution Satellite Imagery for Estimating Economic Well-being and Geographic Targeting, World Bank senior economists Ryan Engstrom, Jonathan Hersh, and David Newhouse commented regarding the analysis, “These results suggest that features extracted from high spatial resolution imagery hold considerable promise for contributing to small area poverty estimates within areas covered by household surveys.”
The results provide one example of how geospatial analysis using satellite imagery can help organizations and businesses make better decisions. Due to the success of the engagement the World Bank and Orbital Insight are continuing their collaboration for another year of work focusing on Mexico. Orbital Insight continues to explore how commercially available satellite imagery and computer vision algorithms may be applied in both the development and commercial industries.
About Orbital Insight:
Orbital Insight is a geospatial analytics and software company applying computer vision and deep learning algorithms to satellite imagery at scale. Our goal is to understand and characterize socio-economic trends at global, regional, and hyper-local scales. Orbital Insight combines a unique technology stack with geospatial data to allow the company to automatically process petabytes of imagery through the use of customizable and scalable algorithms. Our current customers include Global 500 companies, financial services firms, U.S. Government agencies, the World Bank, and the World Resources Institute. Orbital is backed by top-tier venture capital firms, including Sequoia Capital and Google Ventures.