Aquaya receives Google’s AI for Social Good Contribution
Aquaya is excited to be part of Google’s AI for Social Good program for 2021-22. This contribution will help Aquaya examine more accurate and timely ways to map the location and size of urban and rural settlements.
Proposal ▶️ Leave No One Behind: Spatial AI-Enabled Settlement Mapping to Enhance WASH Access for Vulnerable Populations.
The imperative set by the United Nations Sustainable Development Goals (www.sdgs.un.org) to “leave no one behind” requires dedicated strategies to improve water, sanitation and hygiene (WASH) access amongst the most vulnerable and hard-to-reach populations. Local governments and implementers however lack critical information that they need to plan sanitation interventions targeted at the most vulnerable, both in rural and urban areas.
Existing maps of urban slums rely on field data collection that are quickly outdated as slums grow, densify, gentrify, or appear in new locations. As a result, when crises hit, the response is slower than it could be because decision-makers lack good information on where the poorest households are. Maps of impoverished settlements can help WASH programmers target tools such as subsidies to the households that most need them.
The ability to accurately map all settlements, including those which are very small and remote, will allow WASH implementers to design, customize, and budget programs, and ensure that no populations are missed. Settlement maps can be overlaid with other existing datasets, allowing implementers to identify areas that are both poor and meeting other important planning criteria (such as low WASH access, low education, population, etc.), ahead of sending staff into the field. Implementers often do not have time to collect this type of data with household surveys in time to apply for funding opportunities, leading to poorly targeted programs, and ultimately low success rates.
Research Question 1 :
Urban settlement mapping: Can high-resolution satellite imagery be used to map slums within secondary cities in Kenya? Can a spatial-heterogeneity-aware learning framework improve mapping quality?
Research Question 2 :
Rural and urban settlement clustering: Can advanced unsupervised clustering algorithms be used within existing datasets of populated places to accurately map the location and size of all rural settlements in Ghana? Can they be applied to group and locate poor urban settlements from RQ1 (assuming pixeled or rasterized output maps)?