GEOHUM: A Mountain of Geospatial Information For Humanitarian

Earth observation ( EO ), geographical information ( GI ) systems and geospatial data in general have evolved in the planning and operation of humanitarian assistance from a niche application to a centerpiece in the work of Doctors without Borders ( Médecins Sans Frontières, MSF) over the past decade. In complex, quickly evolving situations, with limited direct access, these are some of the questions where EO and GI can help.

  • What is the number of people in need?
  • How many vaccinations are needed?
  • How should we dimension our freshwater production?
  • Where are favorable  routes for water pipelines?
  • Which locations are more prone to diseases transmitted by mosquitoes  that breed in water bodies?

MSF is a technological leader in the field of EO and GI use in humanitarian action. Now, EO and geospatial data are becoming mainstream: more and more satellite launched, faster computer vision techniques developed, more other geospatial and statistical data available, and all kids of services offered. Therefore, in GEOHUM we investigate how MSF can make the best out of this “mountain of information”, so that their most pressing questions are answered fast, reliable, and clearly, enabling them to more effectively help the people in need. GEOHUM consists of three research areas, Img2Info , ConSense , Info2Comm , as interlocking fields of activities, and building on top of each other.

Research area Img2Info

In the first and fundamental research area Img2Info we look at methods to extract information from Earth Observation data. In particular:

  • Combining deep learning in convolutional neural networks with object based image analysis (OBIA) to extract buildings and other relevant features from optical satellite images
  • Producing 3D models of cities, using cross-track / cross-date stereo data, which help in the estimation of urban population numbers
  • Improving the analysis of Radar satellite data for mapping under cloud conditions
  • Exploring the potential of big EO data for land cover mapping, flood and fire monitoring, etc.

Research area ConSense

In ConSense , we look at how we combine EO primary data with data and information that are available from other sources, like OpenStreetMap, institutional actors, statistical data etc. We want to develop a toolset for data quality control, data aggregation, spatial regionalization, and validation, which allows us to employ robust data assimilation strategies, resulting in more accurate information products on a high semantic level.

Research area Info2Comm

Info2Comm is about the communication of scientific results to the recipients and decision makers at the level of MSF operations. Especially for information products, which build upon a multitude of input data, understanding and communicating the uncertainties involved in the production process is fundamental. The best and most sophisticated information product is inappropriate if users do not understand the inherent uncertainties, and therefore cannot make confident decisions based on it.

The use of geospatial data sometimes touches ethical issues of privacy as well. Development of guidelines in this regard is also part of this research.