Wednesday, April 9, 2025
Urban Poverty Spatial Data Needs
Dennis Mwaniki, UN-Habitat
Dana R Thomson, CIESIN, Climate School, Columbia University
Nicera Wanjiru Kimani, Community Mappers / SDI-Kenya
Caroline Kaberia, African Population and Health Research Center
Peter Elias, University of Lagos
Urban Poverty Mapping Methods Expo
Alex de Sherbinin, CIESIN, Climate School, Columbia University
Kamwoo Lee, World Bank
Maxwell Owusu, George Washington University
Monika Kuffer, University of Twente - ITC
Francis Onyambu, African Population and Health Research Center
Ryan Engstrom, George Washington University
Dana R Thomson, CIESIN, Climate School, Columbia University
Dana R Thomson, CIESIN, Columbia University (POPGRID)
Jessica Espey, WorldPop, Univ. of Southampton (POPGRID)
Open Discussion
Dennis Mwaniki of UN-Habitat outlined the limitations of using household-level survey data to track slums and urban poverty, noting that such methods often misclassify households and overlook tenure insecurity and community-generated data. He emphasized the need for spatial mapping to identify where informal settlements are located, how severe conditions are, and how these areas change over time. Responding to growing demand from national and city governments, UN-Habitat is supporting efforts to develop disaggregated, settlement-level data, including a new mandate to incorporate slum-specific data in the 2030 census round. Dennis described a multi-level geospatial framework—ranging from global morphological indicators to locally validated community data—as a pathway to generate more accurate, scalable, and policy-relevant information on informal settlements.
Recording
Slides
Dana R Thomson introduced the IDEAMAPS Network, an informal collective of actors—from community members to UN agencies—collaborating to co-produce neighborhood-level data that is locally relevant and globally scalable. She emphasized that the true value of data lies not in the dataset itself, but in the trust and shared understanding it can foster between marginalized communities and local decision-makers. Dana highlighted that co-production, rather than just consultation, is essential to building equitable partnerships and improving outcomes. Drawing on lessons from the IDEAMAPS Network’s evolving practice, she critiqued the common approach of starting with large, multi-stakeholder workshops, arguing instead for a bottom-up process that begins with relationship-building and data work led by and for communities. She noted that both communities and city departments typically prioritize asset and service data—not slum maps—which are actionable for planning and feedback. This, she argued, can serve as a foundation for dynamic, validated, and scalable data systems that meet both local needs and global reporting demands, such as SDG monitoring.
Recording
Slides
Nicera Wanjiru Kimani, founder of Community Mappers and a resident of Kibera (Nairobi, Kenya), offered a powerful reflection on the lived realities of informal settlements and the disconnect between technical research and community experience. She challenged stereotypical portrayals of slums, describing them instead as places of resilience, creativity, and strong social bonds. While acknowledging the relevance of data-driven work, she stressed that it only becomes meaningful when grounded in transparency, consent, and genuine community involvement. Nicera raised concerns about data misuse, privacy, and exclusion from policymaking, and emphasized the importance of using localized language and participatory processes to ensure data is both accessible and empowering. Through Community Mappers, she and others equip residents to collect, analyze, and use their own data for advocacy—such as resisting unlawful evictions—thereby shifting power toward communities. She called for deeper inclusion of slum dwellers in decision-making, including in policy design and technical research, asserting that they are not just subjects of data but capable partners in its co-production and use.
Recording
Slides
Caroline Kabaria of the African Population and Health Research Center (APHRC) reflected on efforts to bridge community and city data systems in Nairobi, where informal settlements house an estimated 60% of the population. Despite decades of community data collection, city departments often remain unaware of or distrustful of these datasets, while communities, in turn, fear their data may be misused—such as for eviction. Caroline described APHRC’s role as a mediator, helping build trust and shared tools for data collection between community actors and city officials. She emphasized that slum-related issues cut across multiple departments—housing, sanitation, transport—yet data is fragmented and institutional coordination is weak. Through the IDEAMAPS project, APHRC works with communities to identify local priorities, such as water and sanitation, and align them with city department mandates. However, she noted challenges with data gaps across neighborhoods and called for co-produced foundational datasets and capacities to support planning and monitoring in the most deprived areas.
Recording
Slides
Peter Elias discussed persistent inequities in global research collaborations, particularly those involving data production and use in the Global South. Drawing on decades of experience, he highlighted how most partnerships remain structurally imbalanced: institutions in the Global North typically lead on funding, technology, modeling, and data storage, while partners in the Global South often serve as implementers or data collectors with limited decision-making power. This imbalance weakens long-term impact, as projects often end when external funding runs out, without building sustainable local capacity. Peter called for a shift toward locally led research and data systems, supported by meaningful investments in skills, infrastructure, and leadership among researchers and institutions in the Global South. He cited the LIRA 2030 program as a model that empowered African researchers to lead city-scale projects and emphasized the importance of co-design, co-production, and technology transfer. This approach, he argued, is essential to achieving equitable knowledge production and lasting data impact.
Recording
Slides
During the Q&A, participants focused on building trust, data protection, and equitable partnerships. Caroline Kabaria highlighted that trust between communities and city departments is built through aligned priorities and transparent data use, though unintended consequences like evictions remain a risk. Dennis Mwaniki noted growing national support for standardized slum mapping methods and emphasized the challenge of scaling community participation. Peter Elias stressed that trust-building is slow but achievable through deliberate facilitation by academic institutions, citing successes in Lagos and the relevance of the new Copenhagen Framework on citizen-generated data. Audience members raised questions about safeguarding granular community data through licensing and intellectual property rights, and how to improve funding models for Global South research institutions. Caroline responded by emphasizing the importance of anonymizing sensitive data and building internal capacity in local government. Peter recommended empowering African researchers through leadership roles in funding programs, institutional investment, and support for publication and international exchange. The session closed with a call to strengthen long-term, locally-led research infrastructure and inclusion in global data initiatives.
Recording
Alex de Sherbinin, Director of the Center for Integrated Earth System Information (CIESIN) at Columbia University, presented the development of the Global Relative Deprivation Index (GRDI), a gridded dataset integrating satellite-derived and socioeconomic indicators to estimate deprivation globally. The initial version combined metrics like child dependency ratios, subnational infant mortality, built-up areas, nighttime lights, and the Human Development Index. While the approach was intentionally simple and transparent, initial versions unintentionally biased results toward urban areas due to over-weighting satellite data. In version 1.1, the team refined the methodology by downscaling satellite indicators, replacing inconsistent inputs, and improving population metrics to correct urban bias. Alex acknowledged ongoing limitations, such as underestimating deprivation in well-lit, informal settlements, and outlined plans for a more theoretically grounded version 2, using principal component analysis to separately assess economic, social, and health dimensions.
Recording
Slides
Kamwoo Lee, a data scientist at the World Bank, presented his work generating a high-resolution poverty map for Sub-Saharan Africa, covering over one million populated areas at a 1 sq. mile resolution. Using only publicly available data, he combined feature-based and image-based machine learning approaches to estimate current welfare levels. Key inputs included OpenStreetMap features, population density, nighttime lights, and satellite imagery. To address gaps in training data and the limitations of each method, he developed a self-refining mechanism that iteratively improved predictions by integrating outputs between models. By harmonizing wealth data using the International Wealth Index (IWI) and applying a cross-country training approach, he validated the model across 25 countries, showing it performs as well as or better than single-country models. The most predictive single feature was the share of area with zero nighttime light, though Kamwoo emphasized the risks of relying on any one metric. The resulting map is publicly available through the World Bank Microdata Library, and his ongoing work focuses on detecting intra-urban inequalities and slum conditions at finer scales.
Recording
Slides
Dana Thomson presented an overview of the IDEAMAPS Network’s evolution and its diverse approaches to mapping urban poverty and slums. She traced the development of multiple projects within the network, and highlighted key conceptual frameworks guiding the work, including UN-Habitat’s policy/planning framework on slums, Divyani Kholi's geospatial ontology of slums, and IDEAMAPS’ Domains of Deprivation framework. She also outlined the IDEAMAPS Network’s model for iterative feedback and co-production, where community and government users help validate and improve model outputs, which are then reintegrated as training data.
Recording
Slides
Maxwell Owusu, a former MSc student at the University of Twente and George Washington University, presented findings from his research on user-driven Earth observation approaches to slum mapping in Accra, Ghana. Conducted as part of the SLUMAP project, the study explored how to develop deprivation maps that are not only methodologically robust but also meaningful and usable for end users, such as planners and community stakeholders. The research combined data-driven modeling with participatory fieldwork to assess ground-level characteristics and geoethical concerns, particularly the risks of eviction linked to public data release. Using high-resolution imagery and contextual features, Maxwell produced slum classifications aggregated to street blocks, incorporating model confidence levels and validation markers to guide user interpretation. The final map product, hosted online, reflects these insights and emphasizes transparency about model uncertainties.
Recording
Slides
Monika Kuffer presented a user- and data-centric approach to mapping informal settlements using artificial intelligence, as part of the IDEAMAPS initiative and in support of SDG 11.1.1 monitoring. She emphasized that deprivation is not a binary condition and criticized the common practice of training models on limited or unrepresentative data, which can produce poor results. To address this, her team started with pilot cities—ranging from easy-to-map (like Nairobi and Mumbai) to more complex cases (like Buenos Aires)—and then expanded to under-mapped, diverse cities. Using cost-effective Sentinel-2 imagery and cloud infrastructure, they developed scalable models for mapping settlement extent, spatial diversity, and changes over time, while preserving anonymity through 100-meter grid outputs. Crucially, they involved local stakeholders in validating model results via an interactive portal, where users could flag errors and annotate deprivation characteristics. This iterative process improved reference data and mapping accuracy. The project aims to scale to national levels and has already supported policy processes like Argentina’s slum registry. Future goals include building continental-scale models and transferring knowledge to local governments to support long-term data use and policy engagement.
Recording
Slides
Dana R Thomson summarized Francis Onyambu's presentation on the IDEAMAPS Data Ecosystem project, which is focused on building a platform and process for routine, accurate mapping of slums, informal settlements, and other deprived urban areas. Funded by the Gates Foundation and led from the University of Glasgow with hubs at the University of Lagos and APHRC in Nairobi, the project emphasizes co-creation and accessibility of modeled data.
The platform displays 100-meter gridded outputs with two- or three-level classifications (e.g., low/medium/high deprivation or binary presence/absence) using color-coded maps. Community members and local stakeholders can interact with the data via mobile devices, tapping grid cells to validate or correct the information. The site—ideamapsdataecosystem.org—logs user feedback through visible hash marks and tick marks.
The current version is being used to validate a model of morphological informality, initially developed by Angela Abascal and recently refined by Sebastian Hafner, with input from others in the network.
Recording
Slides
Ryan Engstrom presented IDEAMAPS Network research aimed at modeling not only slum versus non-slum areas but also degrees of informality, based on participatory action research in Nairobi, Lagos, and Accra. Responding to community input for more nuanced classifications, his team is testing a high–medium–low informality model using Sentinel-2 imagery, machine learning, and 53 global geospatial layers spanning multiple deprivation domains. Contextual features derived from spatial patterns—such as the histogram of oriented gradients—are central to the analysis. Binary slum models have shown strong accuracy (F1 ~0.94), but extending to severity levels is limited by the small number of labeled training points in the "medium" category.
Ryan emphasized persistent challenges in defining medium informality, managing multicollinearity across input datasets, and balancing model generalizability with local relevance. He highlighted the importance of qualitative feedback loops and community validation to improve model interpretability and trust. Ultimately, he called for new strategies to address training data gaps and overfitting risks while maintaining global scalability—acknowledging that meaningful slum mapping depends on both technical rigor and collaborative engagement.
Recording
Slides
Dana R Thomson reflected on equity challenges within the IDEAMAPS Network, emphasizing the need to rebalance technical roles and decision-making between Global North and South partners. She highlighted examples where Northern institutions handled data processing and modeling while Southern partners managed logistics, calling for more equitable sharing of skills and responsibilities—especially to support local leadership in slum and deprivation mapping.
Second, she described a top-down, global slum-proxy mapping project conducted at CIESIN, funded by USAID, Mercy Corps, and NASA. The project used open global datasets (e.g., GHSL Urban Centers, Overture building footprints, OSM roads) to define urban segments and identify socially and economically vulnerable areas based on indicators like road access and parcel size. Though not a direct slum map, this dataset was used to assess overlapping vulnerabilities with environmental hazards (e.g., floods, pollution) across 5,000+ cities, and was showcased at the World Urban Forum. Dana noted the importance of vector datasets (point, line, polygon) as a common ground among diverse global data co-producers that include communities and local governments.
Recording
Slides
The Q&A session focused on the real-world application of slum mapping data and the power of community involvement. Dana R Thomson provided context for current evictions in Nairobi and Lagos, which were accelerated by floods and new riparian zone policies. Nicera Wanjiru K and Monika Kuffer shared how community-generated maps have empowered residents to advocate against illegal evictions, with some maps even used in court proceedings.
Participants discussed how mapping tools are being used in real time to document evictions and support affected communities. Nicera emphasized the importance of community-led data collection, transparency, and documentation through data journalism, sharing examples of vulnerability mapping and locally developed adaptation plans.
In response to questions about integrating complex data into planning, Maxwell Owusu and Monika Kuffer explained current limitations (e.g., lack of building height data) and future directions, such as incorporating LiDAR or Street View data and using natural features (e.g., slopes, waterways) as proxies. They stressed the need for models that can transfer across diverse cities with varying morphologies. Dana concluded with reflections on operationalizing these data for government and community use, advocating for spatial units like city blocks over arbitrary grids and for attributes tied to real-world features (e.g., roads, buildings, settlement age).
Recording
Official statistics are facing an unprecedented crisis. More than 100 countries across Africa, the Caribbean, Asia and the Pacific rely on international support for approximately half of their census and household survey costs, and that funding is rapidly dissolving. The termination of USAID in February 2025 ended the Demographic and Health Survey (DHS) program. Dynamic geopolitics of the US and Europe point to severe cuts to UNFPA (which supports national censuses) and UNICEF (for the Multiple Indicator Cluster Surveys - MICS) in the near future.
Accurate census and survey data underpin democratic governance, fair resource allocation, and evidence-based policymaking, informing decisions about health, education, housing, and economics. Half of the Sustainable Development Goal (SDG) indicators depend on population denominators, and one-third rely directly on survey-generated data. At the launch of the 2030 census round (2025–2034), lower-resourced NSOs grapple with this new stark reality that their next census and surveys will need to be almost entirely domestically funded.
This session was not recorded to enable participants to speak more openly about the massive challenges and tidal shifts that are unfolding for organizations involved with censuses and surveys.
The discussion focused on how non-traditional data (e.g., geospatial) and methods from data science and statistics might be leveraged by NSOs to meet population data needs in the next decade, and how some pioneering NSOs may need to rapidly innovate.