In emergencies like the Indonesian earthquake and tsunami, humanitarian actors are often confronted with a lack of maps, which are crucial to make quick decisions, ultimately saving lives. To address this issue, HI and crowdAI launched the Mapping Challenge. The five best solutions are being presented at the 5th International Conference on Data Science and Advanced Analytics, held in Turin on October 1st to 4th, 2018.
In emergency situations, humanitarians need recent and precise maps of the damaged or destroyed area.
“We need a clear picture of the situation so we can make the right decisions quickly,” explains Paul Vermeulen, Project manager for Strategic Innovation at HI. “We also need regular updates of these maps during our interventions.”
Many parts of the world have never been mapped. For instance, in developing countries, settlements with several thousand inhabitants might be marked by no more than simple roads. Such inadequate mapping is especially true in the most marginalized areas—those most vulnerable to natural disasters.
“Without accurate maps, it is a huge challenge for international organizations to identify resources and plan an effective emergency response,” explains Lars Bromley, principal analyst and research advisor at UNOSAT. “Obtaining maps of these potential crisis areas greatly improves the preparations, planning and implementation of emergency actions. UNOSAT and others have long worked hard to provide such information. More data for mapmaking is always a crucial need and one of our primary focuses.”
Saving lives with algorithms
Satellite imagery is readily available to humanitarian organizations, but translating these images into maps is an intensive effort. Today specialized organizations produce these maps, or they rely on events such as mapathons, where volunteers annotate satellite imagery with roads, buildings, farms, rivers, etc.
Images of the earth are increasingly available from a variety of sources, including nano-satellites, drones, and conventional high altitude satellites. “Today, the data market is booming and actors buy those images at unaffordable prices,” Vermeulen says.
“It is an imperative for NGOs and civil society actors to develop partnerships with the scientific and private sectors, in order to access maps at reasonable cost.”
HI launched its first Mapping challenge in partnership with crowdAI and with the support of UNOSAT and UN Global Pulse. The mission, for more than 50 participating researchers, was to produce maps of buildings using machine learning.
Over two months, the researchers developed and tested algorithms capable of accurately translating the pixels from satellite images into features on maps. crowdAI then evaluated each contribution. Out of 717 algorithms, they identified the five best performing ones for further testing.
Hope for populations living in unmapped zones
“This is a big hope for populations living in unmapped zones,” says Vermeulen. ”With the mapping challenge, we are identifying new actors and possible solutions to make recently updated maps more accessible. In addition, we are confident that innovation will allow new partnerships for increased access to updated maps.”
Platforms developed by start-ups for automated image analysis will soon be tested. What’s more, areas contaminated by landmines can be captured in high definition by cameras and various sensors attached to drones and analyzed by these same platforms. HI, a leading actor in humanitarian demining activities, will testing such technologies in 2019.