The changing surface of the world's roads.

Deep learning-based satellite analysis maps 95.5% of 9.2 million km of critical roads globally, revealing detailed infrastructure conditions for 2020 and 2024.

• Why it matters: Understanding the physical state of road surfaces at a global scale is essential for sustainable development, climate resilience, and equitable infrastructure planning, yet such data has been scarce and inconsistent.
• What they did: A deep learning framework was developed to extract road pavedness and width from satellite imagery, producing a comprehensive dataset that covers nearly half of previously unclassified roads worldwide.
• The result: This dataset enables detailed assessments of infrastructure investment, accessibility, and vulnerabilities, supporting humanitarian logistics and revealing governance and climate-related disparities across regions.