As road building expands globally, an automated system for detecting and mapping roads in near-real time is urgently needed to plan land use and conservation management. Machine-learning or artificial-intelligence (AI) specialists must help to meet this formidable challenge.
Current road data are grossly inadequate (see W. F. Laurance et al. Nature 513, 229–232; 2014), and most mapping techniques rely on visual interpretation by humans. Even the community-led OpenStreetMap initiative — which aims to maintain accurate maps of roads and many other built features — is patchy and suffers from systematic biases among nations, regions and biomes (www.openstreetmap.org). The freely available, high-resolution radar data sets being collected globally in all weather conditions under the European Union’s Copernicus Earth-observation programme are an important advance.
What we sorely need now is a road-detection algorithm that can discriminate between paved and unpaved roads. Crucially, it would need to operate consistently under varying topographical and environmental conditions, and be able to distinguish roads from other linear components such as low walls, irrigation ditches and natural features.
A road-detection algorithm would be instrumental in the discovery of illegal roads that are imperilling the world’s most vulnerable ecosystems and species. Authorities will then stand a fighting chance of tackling this huge problem effectively.
Nature 558, 30 (2018)