Mapping the Unmapped: Why Understanding Low-Income Areas is Essential.
Is understanding changes in low-income areas essential? This is a question that policymakers, researchers, funders, service implementers, and investors should ask themselves.
Low-income areas are neighborhoods or regions with a median household income significantly below the national or regional average. High poverty levels, poor infrastructure, and inadequate public services typically characterize these settlements. Many countries invest in basic infrastructure to support socio-economic growth in low-income and surrounding areas by ensuring everyone can access essential services such as shelter, roads, water, sanitation, and waste disposal.
Local governments often lack critical planning information in urban areas; most African cities have no publicly available high-resolution maps of informal settlements or low-income areas. Subsidies and services are, as a result, poorly targeted, and timely responses to needy households are delayed.
Therefore, the project aimed to create a novel machine learning solution, building on previously published methods to generate fine-scale maps of low-income areas in rapidly growing urban areas using multi-source public data coupled with each Airbus and Sentinel 2 satellite imagery. We conducted extensive field surveys in Kenyan cities (Nairobi, Kisumu, Malindi, Nairobi, and Nakuru) to generate high-quality, ground-truth datasets for model experimentation and deployment.
With support from the Google AI for Social Good program, the Aquaya Institute, in collaboration with the University of Maryland, harnessed machine learning techniques to automatically model several available low-income areas in Kenya using publicly available data sources. Using transect walks, we manually mapped the boundaries of select locations in Kisumu, Malindi, Nairobi, and Nakuru, Kenya. In Nairobi, shown in Figure 2, these areas included the Kamukunji settlement (Blue Estate, Katanga, Kiambio, etc.), Mukuru (Mukuru Kwa Njenga, Mukuru Kwa Ruben, Imara Daima, Viwandani), Kawangware (Kabiro, Magithundia, Muslim etc.), Ruaraka area (Korogocho, Kisumu Ndogo, Mathare, Ngomongoe, Dam Valley, Deep Sea, Kaptagat, etc.).
The United Nations Sustainable Development Goal’s directive to “leave no one behind” requires dedicated strategies to improve water, sanitation, and hygiene access among the most vulnerable and hard-to-reach populations. Our findings demonstrated that our novel approach shows promising performance improvements over existing machine learning methods. Water and sanitation providers, local governments, and research organizations for household targeting and service delivery projects can harness maps generated from the model. These maps can also be used for health, education, urban planning, and emergency response. The algorithms can be re-run often, allowing for the identification of low-income areas, which rapidly change, move, and grow, without the need for slow and resource-intensive manual mapping efforts. The goal is to replicate this idea in all cities in Africa.
This work was supported by Google, Airbus, and the University of Maryland. The Aquaya Institute is a nonprofit research and consulting organization dedicated to advancing global health through universal access to safe water, sanitation, and hygiene.
By Irene Atieno, Kara Stuart, and Karen Setty, The Aquaya Institute