Abstract
Abstract
Resilient road infrastructure is a cornerstone of the UN Sustainable Development Goals (SDGs), yet spatially consistent information on road surface conditions at the global scale remains limited. We overcome this gap based on a deep learning framework that extracts both road pavedness and road width from Planetscope satellite imagery. Thereby, mapping 95.5% of 9.2 million km of critical arterial roads for 2020 and 2024, nearly half of which was previously unclassified.The resulting dataset and multi-scale analytical framework enable assessments of road infrastructure in the context of development, climate adaptation, and equity.Unlike proxies such as nighttime lights, the data directly captures the physical infrastructure underpinning resilience at higher spatial resolution.This dataset reveals a multi-scale geography of human development, capturing infrastructure investment and accessibility at the planetary scale, enabling a global Humanitarian Passability Matrix for humanitarian logistics, while also providing local insights into road disparities—for example, revealing governance-related patterns in Ghana and highlighting climate-related vulnerabilities in Pakistan.