The COVID-19 pandemic has exposed long-standing inequalities in Pakistan, with dramatic differences between the country’s wealthy and poor populations in vulnerability to the virus and its impact. As development implementers work to reduce inequality while mitigating the disease and its secondary impacts, advances in machine learning (ML) provide a new localized understanding of differences that enable interventions to be sustainable, targeted, and impactful. Combining geospatial data with hundreds of indicators from satellite imagery, Fraym is leading the growing body of work to understand inequality among traditionally data-poor populations by maximizing existing data and remote analysis.

Fraym uses advanced ML to combine household surveys and satellite imagery – using harmonized survey data to ‘train’ a series of models with remotely sensed data as inputs to calculate population characteristics at 1km2 resolution. The final product is a consistent geospatial ‘surface’ that provides the information necessary to identify, target, and assist at-risk areas, assess their economic, health, and social characteristics, and quantify need at the community level. Using a series of data layers in Pakistan, Fraym analyzed three manifestations of inequality and avenues to address them: vulnerability to COVID-19, food insecurity, and access to finance.

Understanding a community’s vulnerabilities is critical to respond to disease outbreaks. Unfortunately, population and health data are often only available at aggregated levels that obscure differences in populations, towns, and neighborhoods. In Pakistan, Fraym developed profiles at the 1km2 level that measured COVID-19 related risk across five themes: exposure, comorbidities, health facility access, communication access, and socioeconomic vulnerability.

Fraym combined data layers to produce multi-faceted understanding of these five themes. For example, Fraym enhanced data on populations at-risk of exposure, including population density, proximity to others in the household, rates of adults working in essential and manual occupations, and sanitation. This data prioritizes provinces for assistance and quantifies this aspect of risk, finding that over 22 million people in Punjab are at the highest risk level for COVID-19 exposure, and only 35% of households in rural Sindh province have access to improved water sources.

While health and sanitation indicators provide insight into COVID-19’s immediate threats, understanding access to finance will tailor strategies to mitigate its economic impact and that of other future shocks. In June 2020, Pakistan’s government launched an emergency cash program that could be accessed using Cash-In-Cash-Out (CICO) services. Under these programs, individual agents use mobile money services to act as banks for communities where full-fledged, physical banks are too expensive to implement.

While health and sanitation indicators provide insight into COVID-19’s immediate threats, understanding access to finance will tailor strategies to mitigate its economic impact and that of other future shocks. In June 2020, Pakistan’s government launched an emergency cash program that could be accessed using Cash-In-Cash-Out (CICO) services. Under these programs, individual agents use mobile money services to act as banks for communities where full-fledged, physical banks are too expensive to implement.

ML approaches like Fraym’s enable development programming to be targeted, efficient, and data-driven in even traditionally data-poor and remote communities. In a new environment that increasingly limits on-the-ground actions like data collection, robust and remote analytical approaches will ensure interventions continue to act upon the latest population information for their strategy design.