The Humanitarian Openstreetmap Team (HOT) has incorporated drone technology in mapping Liberia’s capital Monrovia, as part of a wider project to bring together the best machine learning practitioners from around the world to develop open-source algorithms and workflows that can generate geospatial data for cities.
HOT is an international team dedicated to humanitarian action and community development through open mapping. They provide map data which expedites responses and reduces risk during disasters and also enable government, communities and international organisations to contribute to opensource mapping for locally-relevant challenges through provision of training, equipment, knowledge exchange, and field projects. Their open source tools and maps are free and have been used by humanitarian organisation which include the Red Cross societies, Médecins Sans Frontières and UN agencies.
For the Monrovia mapping project, HOT worked in partnership with computer training hub iLab Liberia and with the support of the World Bank, roped in the expertise of Tanzania’s Uhurulabs to map the city and explore the potential for machine learning algorithms to augment the work of human mappers. The project was part of the Open Cities AI Challenge, where HOT and iLab Liberia created spatial data in Monrovia that was used to test and refine machine learning algorithms.
The goal of this Machine learning initiative is to offer HOT, humanitarian and development organisations, and city leaders a valuable toolkit to automate large-scale, repetitious data generation and mapping tasks while highlighting areas that need human attention. As urban populations grow, more people are exposed to the benefits and hazards of city life. To manage the risk of natural disasters in this dynamic built environment, buildings need to be mapped accurately, frequently and in enough detail to help communities prepare and respond. ML algorithms are becoming critical to scaling these mapping efforts by learning to use aerial imagery to automatically create building footprints.
“In the Liberian capital, project activities were aimed at addressing flooding and challenges relating to flooding through the provision of up-to-date Zone 300 data, the area of interest,” HOT said in a statement. “Flooding in Monrovia is cyclical and occurs during the April-October rainy season every year. Field mapping teams collected data during the rainy season, making way for interventions to occur during the next dry season.”
The team first worked with Uhurulabs to generate twenty square kilometres of drone imagery of Monrovia, which they did in six days with a SenseFly eBee drone. The drone data was then provided to the machine learning challenge for participants to develop machine learning algorithms to create building footprints, which were then compared to footprints for the same area created by human mappers based in Monrovia. This side-by-side exercise created valuable data to refine the algorithms’ capabilities.
“In addition to the machine learning benefits, the drone imagery was used to map roads, buildings, and drains for the twenty square kilometre area where the available satellite imagery did not have the resolution needed to make out these features. The drone imagery is also now publicly accessible online, via OpenAerialMap.”
When the field work was done, the team used software like OpenStreetMap (OSM) and a collection of open mapping tools including Java OpenStreetMap Editor (JOSM), the HOT Tasking Manager, OpenMapKit (OMK), Open Data Kit (ODK), Mapillary, and QGIS, to collect, validate, map, analyse and share accurate datasets from the target communities.
The HOT team also enlisted the help of government stakeholders during the mapping period, who included the Monrovia City Corporation (MCC), Liberia Institute of Statistics and Geo-Information Services (LSGIS), National Disaster Management Agency (NDM), the Ministry of Public Works, the Liberia Water and Sewer Corporation, and Ministry of Gender, Child and Social Protection.
Started in 2018, the Open Cities Challenge featured high-resolution drone imagery from 12 African cities and regions – Pointe Noire and Brazzaville in Congo, Monrovia, Ngaoundéré, Cameroon, Ggaba Parish, Uganda; St Louis, Senegal; Kinshasha, DRC; Antananarivo, Madagascar; Mahe, Praslin and La Digue in Seychelles; and Niamey in Niger – covering more than 700,000 buildings.
This imagery was then paired with building footprints annotated with the help of local OpenStreetMap communities.