The 2013 Ebola outbreak across West Africa brought into focus a major information gap that impeded effective response to the crisis – a lack of reliable population maps. Accurate and timely population distribution maps are critical to addressing health epidemics, but are also heavily used for natural disaster response and impact assessment, to track global changes for environmental conservation, and to address human rights issues. Many countries and territories across the globe lack the infrastructure and resources to map their population consistently and accurately, especially in remote rural areas. Geographical Sciences PhD Candidate Amanda Hoffman-Hall and other members of the Loboda Research Lab (Tatiana V Loboda, Joanne Hall, Mark Carroll, and Dong (Tony) Chen) recently published a new methodology that uses moderate resolution remotely sensed data to map human settlements across rural and isolated landscapes, titled "Mapping remote rural settlements at 30 m spatial resolution using geospatial data-fusion" in the journal Remote Sensing of Environment.
Moderate resolution remote sensing data, such as imagery acquired by NASA's Landsat Science Program, has generally been viewed as insufficient for rural mapping given its coarseness relative to the size of a rural dwelling. Landsat pixels are 30 m spatial resolution, which means that each pixel within a Landsat image corresponds to an area roughly 100 x 100 ft on the ground surface. This makes it virtually impossible to visually locate small isolated settlements, which sometimes consist of only 2-3 small dwellings. The figure to the right highlights the differences between Landsat and finer spatial resolution imagery.
The research presents an approach for detecting small rural settlements within Ann Township of Rakhine State, a remote region of Myanmar with a highly mobile population, by combining Landsat data with publicly available auxiliary geospatial data. The key to the success of detecting small settlements at the Landsat resolution was the inclusion of multi-date data sensitive to human activity patterns in the regions, such as distance to water, change in seasonal vegetation signals, and distance to a recent active fire. The use of active fire in this context is novel and provided a key metric that increased the accuracy of the resultant map.
The final dataset has a classification accuracy of 86.5% on a per-pixel basis and 93.1% on a location identification basis. The figure below shows the final dataset, including some portions overlaid finer resolution imagery. Numerous small settlements (on the order of 2–3 structures in some cases) not previously mapped by other datasets were identified, revealing that the population of Ann Township is far more dispersed and isolated than previously mapped. This study concludes that by incorporating regionally specific characteristics, moderate resolution remotely sensed data can be used to successfully map geographically marginalized communities so that services and aid are better able to reach them.