A new global road dataset built with artificial intelligence and high-resolution satellite imagery maps 9.2 million kilometers of roads, offering new insights into infrastructure, economic development and humanitarian access.
AI road mapping identifies 9.2 million kilometers of roads worldwide
From highways and paved streets to dirt roads, researchers have mapped 9.2 million kilometers, about 5.7 million miles, of roads worldwide using artificial intelligence and high-resolution satellite imagery. The result is a massive dataset that provides a detailed picture of the planet’s road network, classifying roads by width and pavement conditions.
The imagery was collected over four years, from 2020 to 2024, by researchers at the Institute of Geography at Heidelberg University. According to the research team, the new approach achieves roughly 20 percentage points greater accuracy than previously available datasets.
Such detailed AI road mapping could help researchers understand how transportation infrastructure changes over time as a result of climate change, natural disasters, conflicts and deterioration. The data could also support humanitarian operations in some of the world’s most difficult-to-reach areas and provide another way to assess socioeconomic development in regions where conventional indicators are scarce.
Although the United Nations considers access to safe and developed road infrastructure an important component of the Sustainable Development Goals, researchers have lacked a global benchmark capable of systematically assessing road conditions. A study published in Nature Communications aims to fill that gap.
AI road mapping combines satellite data and deep learning
The key methodological advance comes from combining Earth observation with advanced deep-learning techniques. Researchers used high-resolution imagery from PlanetScope satellites to train models capable of identifying major roads and classifying them according to two fundamental characteristics: road width and pavement level.
“Our model is not only significantly more accurate, but also makes it possible to monitor changes in road infrastructure over periods of several years,” said Sukanya Randhawa, head of the “GeoAI for Good” projects at HeiGIT.
The ability to repeatedly analyze the same areas could turn the dataset into a tool for tracking how infrastructure evolves rather than simply providing a static snapshot of the world’s road network.
Roads can reveal patterns of economic development
Understanding road characteristics and tracking how they change over time can provide valuable information about accessibility under different conditions, including during extreme weather events.
Road quality and pavement levels are also closely connected to economic development and future growth potential. The difference between a three-lane highway and a narrow, winding single-lane road can have major consequences for freight movement, passenger transportation and access to markets and essential services.
To turn the raw data into a practical tool for emergency responders, the research team developed a Humanitarian Passability Matrix. The system is designed to help humanitarian organizations plan emergency routes by estimating whether roads are likely to remain accessible as environmental conditions change.
Ghana shows how AI road mapping can support urban planning
The researchers tested the model in several real-world settings, including Ghana. In Accra, the country’s capital, geospatial analysis made it possible to identify infrastructure gaps within individual neighborhoods.
That level of detail could give local governments and policymakers a stronger evidence base for more equitable urban planning and more targeted allocation of public resources.
The approach could be particularly valuable in rapidly growing cities where conventional infrastructure data may be incomplete, outdated or unavailable.
As roads change, they reveal how societies are changing
Beyond emergency response, the dataset could serve as a powerful indicator of economic and social well-being. It may be especially useful in regions where traditional measures of development, including historical nighttime-light data, have significant limitations.
The research shows that countries with a higher share of paved roads tend to rank higher on the United Nations Human Development Index, while areas with more unpaved roads and slower rates of road construction generally show lower levels of development.
By combining satellite observations with artificial intelligence, researchers can therefore use roads not only to understand how people and goods move, but also to gain a clearer picture of how communities and economies are evolving.
