Jobs cluster where roads are good and lights burn bright at night
In a country where national statistics have long obscured as much as they reveal, researchers from the ILO and partner universities have used satellite imagery and machine learning to render Ghana's labour market visible at a scale never before attempted. By mapping 17 employment indicators across half-kilometre grid cells, they have transformed a blurred national average into a precise geography of opportunity and exclusion. The work arrives as a quiet argument: that the tools to understand inequality have outpaced the will to act on it, and that developing nations need not wait for expensive surveys to know where their people are left behind.
- Ghana's headline employment rate of 48% conceals a fractured reality — youth employment sits at just 23%, gender gaps persist, and nearly two-thirds of all work is informal or self-employed.
- The north-south divide is not merely statistical but spatial: jobs cluster densely along the Accra-Kumasi corridor and coastal zones, while northern communities find opportunity only in scattered urban centres like Tamale and Bolgatanga.
- Researchers fused 64 layers of satellite and geographic data — nighttime light, road density, vegetation health, building coverage — with census records and trained an AI model to estimate employment patterns at a resolution no survey had ever reached.
- The resulting maps expose not just where jobs are absent, but why: poor infrastructure, low economic luminosity, and agricultural dependence define the regions most left behind.
- Policymakers now hold a tool that can direct infrastructure investment and job creation programmes with surgical precision — the question is whether the political will exists to use it.
Ghana's labour market has long been described in broad strokes — a 48 percent employment rate, roughly 9.6 million people with jobs across a population of 30.8 million. These numbers tell a story, but not the whole one. A team of researchers from the International Labour Organization, alongside colleagues at Nanjing University of Posts and Telecommunications, set out to look closer.
The problem is familiar to anyone studying developing economies: national statistics flatten the texture underneath. That 48 percent average conceals a youth employment rate of just 23 percent, significant gender gaps, and the reality that nearly two-thirds of Ghanaian jobs are self-employment — farming, small trading, informal work. No one had mapped these patterns with real granularity. The researchers changed that by dividing Ghana into grid cells roughly half a kilometre wide and estimating 17 different labour indicators for each: employment by age, gender, skill level, sector, and status.
They began with the 2021 Ghana Housing and Population Census and layered on 64 types of geographic and satellite data — land cover, rainfall, vegetation health, road networks, building density, and nighttime light intensity, a reliable proxy for economic activity. A machine-learning model called a random forest, trained on district-level employment statistics, translated all of this into a map of Ghana's labour market at unprecedented resolution.
What emerged was stark. Employment clusters heavily in the south, with the urban corridor between Accra and Kumasi as the country's dominant hub. The north, by contrast, is sparse — economic activity retreating to a handful of cities while surrounding regions offer far fewer opportunities. Salaried and skilled positions concentrate in large cities; rural areas depend on farming and informal enterprise. Local employment rates vary wildly across the country, shaped by the density of buildings, the brightness of lights at night, and the quality of roads beneath them.
For policymakers, these maps offer something concrete: the ability to identify regions with the fewest opportunities, target infrastructure investment, and design job creation programmes with precision. More broadly, the research demonstrates that countries lacking detailed economic surveys can still generate high-quality employment insights by combining census data with satellite imagery and AI. Ghana now has a clearer picture of where its labour market functions — and where it fails. Whether governments and development partners choose to act on that picture remains the open question.
Ghana's labour market has long been a blur of national statistics—a 48 percent employment rate, roughly 9.6 million people with jobs, 30.8 million people total. These numbers tell a story, but not the whole one. A team of researchers from the International Labour Organization, working with colleagues at Nanjing University of Posts and Telecommunications and Jiangsu Province's Smart Health Big Data Analysis and Location Services Engineering Lab, decided to look closer. Using satellite imagery, census data, and machine learning, they built something new: detailed maps showing not where Ghana works in aggregate, but where, precisely, jobs actually exist.
The problem they were solving is familiar to anyone who has tried to understand a developing economy. Labour statistics in countries like Ghana typically arrive at the national or district level—useful for broad strokes, useless for the texture underneath. A nation-wide employment rate of 48 percent conceals the fact that young people aged 15 to 24 face an employment rate of only 23 percent, while workers in their prime years, between 25 and 54, find work at much higher rates. It masks gender gaps, regional chasms, and the reality that nearly two-thirds of all jobs in Ghana are self-employment—farming, small trading, informal work. No one had mapped these patterns with real granularity before. The researchers changed that by breaking Ghana into grid cells roughly half a kilometre wide and estimating 17 different labour indicators for each one: employment by age, by gender, by skill level, by sector, by employment status.
To do this, they started with the 2021 Ghana Housing and Population Census and layered on 64 types of geographic and satellite data. They looked at land cover and rainfall, vegetation health and road networks, building density and nighttime light intensity—that last one a reliable proxy for economic activity. They fed all of this into a machine-learning model called a random forest, trained on district-level employment statistics, and let it estimate employment patterns down to the grid cell. The result was a map of Ghana's labour market at a resolution that had never existed before.
What the maps revealed was stark. Employment in Ghana clusters heavily in the south, with the urban corridor between Accra and Kumasi emerging as the country's dominant employment hub. Coastal areas and regions along major roads show concentrations of jobs. The north, by contrast, is sparse. Economic activity there clusters around a handful of urban centres—Tamale, Wa, Bolgatanga, Bawku—but the surrounding regions show far fewer opportunities. The quality of work follows the same geography. Salaried jobs and higher-skilled positions concentrate in large cities where government offices, industries, and services are located. Rural areas depend on farming and small self-employed businesses. Manufacturing, limited nationally, appears in small clusters around certain urban areas, suggesting the presence of local industrial centres.
When researchers mapped employment relative to the working-age population, the picture became even more complex. Local employment rates vary wildly across Ghana, ranging from very low in some places to near-full employment in others. Southern regions generally show higher employment rates than the north, though some rural areas demonstrate strong participation in farming and informal work. The maps also identified what drives these patterns. Areas with more buildings, brighter nighttime lights, and better road networks tend to have more jobs—these features reflect economic activity and the ease with which goods and people move. Agricultural employment tracks closely with environmental conditions: rainfall and vegetation health.
For policymakers, these maps offer something concrete. They can identify regions with the fewest job opportunities, highlight areas where infrastructure investment might unlock economic growth, and guide targeted job creation programmes. More broadly, the research demonstrates that countries lacking detailed economic surveys can still generate high-quality employment insights by combining census data with satellite imagery and machine learning. Ghana now has a clearer picture of where its labour market actually functions and where it fails. With that information in hand, governments and development partners can design policies that address regional inequality and support more inclusive economic growth—if they choose to use it.
Citações Notáveis
Labour data in many developing countries is usually available only at national or district levels, which often hides important local differences— Study methodology
A Conversa do Hearth Outra perspectiva sobre a história
Why does it matter that we can now see employment patterns at half-kilometre resolution instead of just at the district level?
Because the district average hides everything. You might know that a district has a 40 percent employment rate, but that tells you nothing about whether jobs are clustered in one town or spread across the region. With these maps, you can see that jobs are concentrated in specific places—along roads, near cities, in areas with electricity. That changes how you think about solutions.
The study found that youth employment is only 23 percent while prime-age workers exceed 50 percent. Is that a Ghana problem or a global pattern?
It's both. Youth unemployment is high everywhere. But what these maps show is that it's not evenly distributed. In some parts of southern Ghana, young people have better access to jobs. In the north, the problem is worse. That's the insight that matters for policy.
How reliable is a machine-learning model trained on district-level data when it's being asked to estimate employment in a half-kilometre grid cell?
That's a fair question. The model uses 64 different types of geographic and satellite data—building density, nighttime lights, road networks, vegetation. These are all correlated with economic activity. The researchers trained it on known district-level statistics, so it's grounded in real data. It's not perfect, but it's far better than assuming employment is evenly distributed across a district.
Self-employment accounts for two-thirds of jobs in Ghana. Does that change how we should read these maps?
Significantly. In rural areas, the maps are showing where farming happens and where small traders cluster. In cities, they're showing where formal employment concentrates. The maps don't distinguish between a salaried job and someone selling goods on the street, but the satellite data—building density, nighttime lights—gives you clues about which is which.
What happens next? Do these maps actually change policy?
That depends on whether governments use them. The maps exist. They show regional inequality clearly. Whether that translates into infrastructure investment in the north, or job creation programmes, or anything else—that's a political question, not a technical one. The research has done its part.