Mapping Every Building and Road in sub-Saharan Africa

Throughout the world we find huge variations in data quality and availability. Leveraging machine learning and very high-resolution satellite imagery, Maxar and Ecopia.ai are extracting every building and road throughout 51 countries in sub-Saharan Africa. This data will also be refreshed in year 2020, with a clear path towards annual refresh. This poster/presentation will be focused on the methodology employed to complete this massive project, which will produce more than 275 million buildings and 3 million km worth of roads.

Ecopia.ai has developed a feature extraction algorithm that, when applied to 30-50cm imagery captured by the Maxar satellite constellation, is able to extract buildings with a 90% accuracy while the roads have a minimum accuracy of 85%. This type of large-scale extraction is only made possible through the deep archive of high resolution imagery owned by Maxar, which has been turned into large scale 'Vivid' mosaics. These mosaics are at minimum 50cm in spatial resolution, color balanced across the entire AOI, and are constantly being updated based on the newest imagery captured. This is crucial due to the fact that this region has experienced significant change in the last decade.

This comprehensive dataset of building footprints and roads will directly support many of the Sustainable Development Goals. These data have already been validated and leveraged in-country by non-government organizations and governments on a wide array of use cases including polio eradication, malaria elimination, survey sampling, census campaigns, field operations, conservation programs, the Ebola response and more.