The Hidden Burden of Morphological Deprivation in Small and Medium Cities

The Hidden Burden of Morphological Deprivation in Small and Medium Cities | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article The Hidden Burden of Morphological Deprivation in Small and Medium Cities Sai Ganesh Veeravalli, Alejandro Blei, John Friesen, Bedru Tareke, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8189204/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract In growing cities, deprived neighborhoods house large numbers of residents, yet their extent and distribution remain poorly quantified, complicating implementation of SDG 11.1.1. We present the first global, neighborhood-scale spatial estimates of morphological deprivation, covering 5,132 cities in 103 countries across Africa, Asia, and Latin America & the Caribbean (LAC) home to 3.2 billion people. Neighborhood units and built-environment indicators from the City Segments v1 dataset were combined with segment-level labels from the eight-city IDEABench benchmark to train a supervised model, which was then applied to classify each segment as morphologically deprived or non-deprived. The mapped cities contained 1.96 billion residents, of whom 349 million (17.8%) lived in deprived segments, with the highest regional shares in Africa and substantial burdens in Asia and LAC. Morphologically deprived populations spanned the urban hierarchy, with about one-third living in small and medium cities, revealing important gaps in current deprivation monitoring. Earth and environmental sciences/Environmental social sciences/Sustainability Social science/Geography Social science/Environmental studies Scientific community and society/Geography Scientific community and society/Developing world Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Rapid urbanization across low- and middle-income countries (LMICs) is transforming the spatial and social fabric of cities, yet deprivation remains widespread. More than one billion people, around a quarter of the world’s urban population, are estimated to live in slums and informal settlements lacking secure tenure, durable housing, and reliable access to basic services 1 , and 1.6 billion people are estimated to live in inadequate housing 2 . Addressing these conditions is central to Sustainable Development Goal (SDG) 11.1, which calls for adequate, safe, and affordable housing for all and the upgrading of slums by 2030. Despite decades of policy attention, the spatial distribution and intensity of urban deprivation remain poorly understood, particularly in small and medium-sized cities that now absorb a substantial, though not yet well quantified, share of urban growth across the Global South 3–5 . These settlements are often invisible in official statistics due to coarse administrative reporting units and their small spatial footprint, constraining government’s capacity to target infrastructure, housing, and climate-adaptation investments 6,7 . Advances in Earth observation and machine learning techniques have made it possible to identify the physical signatures of deprivation from space 8–11 , however, global evidence remains fragmented, locally calibrated, and seldom comparable across regions, a gap this study seeks to address. The term deprivation has been used in different ways across urban research and policy, increasingly as a broad concept aligned with SDG 11.1.1 inclusive of ‘slums’, ‘informal settlements’, and inadequate housing. The UN-Habitat definition of a slum household, based on deficits in water, sanitation, durability, crowding, and tenure, remains the official SDG and most widely applied standard 12 . While valuable for national monitoring, this household-based measure does not capture the spatial or morphological dimensions of deprivation within cities 6,13 . Informal settlements, in UN-Habitat’s sense, are areas where residents lack secure tenure and adequate services and where the built environment often falls outside planning and building regulations, even if their physical form varies 12 . At the same time, many formally recognized neighborhoods remain under-serviced and thus deprived. To encompass these variations, recent work conceptualizes Deprived Urban Areas (DUAs) as the spatial manifestation of overlapping social, infrastructural, and environmental disadvantages 6,9 . In this study, we adopt an area-based, spatially modelled deprivation, an expression of disadvantage in urban morphology, services, and infrastructure consistent with the IDEAMAPS ‘Domains of Deprivation’ framework and related work on area-based inequality 6,11,14 . Conventional estimates of urban deprivation rely heavily on population censuses and household surveys that inform UN-Habitat’s global household-based slum indicator. In its updated framing of SDG 11.1.1, UN-Habitat distinguishes three related aspects of slums, informal settlements, and inadequate housing and recommends that national statistical systems use census and survey data to monitor household-level deficits, while acknowledging the need for spatial approaches to capture area-based manifestations such as informal settlements 12 . While household datasets are invaluable for assessing living-condition deficits, they are not designed for intra-urban spatial analysis. Demographic Health Surveys (DHS) cluster coordinates, for example, are intentionally displaced by up to two kilometers in urban areas to protect respondent privacy, introducing spatial uncertainty that limits their compatibility with satellite imagery and local mapping 15–17 . Even where household data are available, the scale mismatch between individual surveys and neighborhood morphology remains substantial. The UN-Habitat definition identifies slum households, whereas mapping deprivation requires delineating deprived neighborhoods shaped by collective built environment, infrastructure and access patterns 6,13 . Moreover, censuses are infrequent and often exclude informal settlements, leaving large populations statistically invisible 18 . These challenges are likely to intensify as large-scale survey programmes face cuts and uncertainty, including the recent termination of the long-running DHS programme 19 . Consequently, survey-derived indicators do not capture where deprivation concentrates or how it evolves, highlighting the need for spatially modelled approaches that integrate Earth observation (EO) and morphometric data to characterize the built characteristics of disadvantage. Several recent initiatives have employed Earth observation and geospatial methods to capture different spatial dimensions of urban deprivation, ranging from access to services to infrastructural inclusion and built-form informality. The Slum Severity Index (SSI) 20 models household-level service deprivation using machine learning on DHS indicators, extending the UN-Habitat framework to produce national-scale estimates across Sub-Saharan Africa (SSA), but its primary focus is on service and housing conditions rather than neighborhood morphology. The Million Neighborhoods (MN) project 7 also focusing on SSA, instead emphasizes infrastructure inclusion, mapping block-level connectivity based on the accessibility of buildings to the road network. The World Resource Institute (WRI) Intra-urban LULC dataset 21 offers a morphology-driven perspective, classifying residential areas into formal, informal-subdivision, and atomistic (small, disconnected housing clusters typically lacking organized street networks) typologies. Collectively, these datasets represent complementary approaches to quantifying urban deprivation, yet differ in their conceptual scope, spatial coverage, data sources, and validation. In this study, we compare our results with these three products to situate our model within this evolving landscape of multi-dimensional deprivation mapping. To address these limitations, we developed a standardized, spatial deprivation model reflecting aspects of unplanned urbanization, service access, and infrastructure for cities across Africa, Asia, and Latin America & the Caribbean (LAC). Across these regions, 3.2 billion people live in cities (2.5 billion, excluding China) according to the most recent Global Human Settlement Urban Centers Database 22 . Our approach combines globally consistent neighborhood-scale spatial units from the City Segments Layer v1 23 , which contains building, road, and population indicators, with validated reference labels from IDEABench 24 , the first multi-city benchmark of morphologically deprived and non-deprived areas derived from Sentinel imagery, Google Open Buildings, and locally validated field data. Using these datasets, we trained a Random Forest classifier on segment-level data from eight IDEABench cities and then applied the resulting model to City Segments v1 across 5,132 cities in 103 countries to generate globally comparable estimates of morphological deprivation. The model relied exclusively on open, spatially aggregated, anonymized data and focused on physical structure rather than individual households or named communities, which is consistent with emerging standards for responsible and equitable use of AI in urban mapping 25 . By covering small, medium, large, very large cities and megacities, the analysis provided, to our knowledge, the first spatially explicit, globally comparable evidence on the distribution of morphological deprivation across the full urban hierarchy in the Global South. The overarching objective of this study was to develop and apply a globally consistent spatial model of morphological deprivation (City Segment Morphological Deprivation, CSMD) using harmonized geospatial data and supervised learning (Figure 1). Specifically, we asked: (1) what are the global, regional, and national patterns of morphological deprivation when assessed consistently at the city-segment scale across 5,132 cities in 103 countries?; (2) how is morphological deprivation distributed across the urban hierarchy, from small and medium-size cities to large cities, very large cities, and megacities?; (3) how do CSMD classifications align with existing large-scale products (SSI, MN, WRI) and what do these overlaps and divergences imply for monitoring SDG 11.1.1 and guiding equitable urban interventions, particularly in smaller cities that are under-represented? Results We spatially modelled urban morphological deprivation using a Random Forest model trained on built-environment indicators from City Segments v1 and labelled reference data from IDEABench. The resulting city segment morphological deprivation (CSMD) model was summarized at global, regional, national, and city-size levels to quantify how morphological deprivation is distributed across the urban system. We also compared CSMD classifications with three independent products, the Slum Severity Index (SSI), Million Neighborhoods (MN) project, and World Resources Institute (WRI) Intra-Urban Land Use product, to assess alignment with alternative representations of deprivation. Global and regional spatial patterns The analysis covers 5,132 cities across 103 countries, representing a combined urban population of approximately 1.96 billion people. Of these, an estimated 349 million people (17.8% of the total) resided within morphologically deprived city segments. Figure 2 maps the locations of the analyzed city segments and reports the percentage of population living in these deprived segments at regional and country scales. Regional shares differed markedly. Africa accounted for about 144 million people living in morphologically deprived segments, or 28.2% of the 512 million residents covered. In Asia, an estimated 147 million people lived in morphologically deprived segments, representing 13.3% of the 1.1 billion people analysed. In LAC, roughly 57 million people (17.0% of the 338 million) were estimated to live in morphologically deprived segments. Together, these regional aggregates indicate that morphological deprivation was most extensive in African cities and substantial in Asia and LAC cities. At the national level, the share of population residing in morphologically deprived city segments varied widely within and across regions (Figure 3). In Africa, national averages ranged from below 10% in countries such as Ghana to around 60% in Somalia, while countries such as Egypt (~19 million) and the Democratic Republic of the Congo (~18 million) contained the largest morphologically deprived populations in absolute terms. In Asia, country shares remained below 35% everywhere, while India (~68 million) accounted for the largest absolute counts. In LAC, Haiti (~69%) exhibited very high national shares, whereas several other countries recorded values below 20%. Country-level statistics, including ISO3 codes, country names, regional classifications, and corresponding area and population aggregates, are provided in Extended Data Table 2. City-size structure of morphological deprivation We summarized populations residing within morphologically deprived city segments by city-size class to examine how deprivation was distributed across the urban hierarchy. The 5,132 cities were grouped into five United Nations size categories 26 : 4,549 small cities (< 500,000 inhabitants), 283 medium (500,000 – 1 million), 252 large (1 – 5 million), 25 very large (5 – 10 million), and 23 megacities (≥ 10 million). Collectively, these cities contained 1.96 billion residents. Of these, approximately 82 million in small cities (12.8% of their population), 29 million in medium (14.7%), 93 million in large (17.8%), 43 million in very large (25.0%), and 102 million in megacities (24.0%) people lived in deprived segments. Taken together, small and medium cities accounted for roughly one-third (32%; 111M of 349 M) of all residents in deprived segments, while large and megacities comprised more than half (56%; 195M of 349 M), showing that deprivation spanned the full urban hierarchy rather than being confined to the largest metropolitan areas. Patterns differed markedly across regions (Figure 3 insets). In Africa, morphological deprivation was observed across all size classes, with the largest absolute numbers in large (42 million) and small (38 million) cities, even though relative shares peak in very large (41.5%) and megacities (38.9%). In Asia, morphological deprivation was heavily concentrated in the largest urban centres: megacities alone accounted for 66 million people (23.2%), almost double the combined total of small and medium cities (36 million), while large and very large cities added further substantial numbers. In LAC, morphologically deprived populations were relatively evenly distributed across size categories, with large cities contributing the highest absolute totals (22 million), followed by small (18 million) and megacities (12 million). Across regions, the magnitude of morphologically deprived populations in smaller cities remained substantial, motivating a closer look at how these contributions varied by country (Figure 4). At the country level, the composition of populations in morphologically deprived city segments varied considerably across size classes (Figure 4). In several cases, including Djibouti, Maldives, and Belize, all observed these deprived populations were located in small cities. In others, such as Nigeria, India, and Brazil, networks of small and medium cities collectively hosted millions of residents in morphologically deprived segments, in some instances matching or exceeding megacity totals. These patterns underscored that national deprivation burdens often reflected distributed systems of smaller and intermediate cities rather than a single dominant metropolis. To provide additional context on small-city patterns, Figure 5 presents satellite-based examples from six small cities across Africa, Asia, and LAC, showing the spatial distribution of morphologically deprived and non-deprived segments. In some cities, such as Juba (South Sudan) and Ouanaminthe (Haiti), deprived segments covered larger portions of the built-up area, whereas in others, including Sambhal (India) and Myaungmya (Myanmar), they formed more localized clusters. Across these examples, morphologically deprived segments ranged from compact pockets to more contiguous expanses, but in all cases corresponded to densely built areas with limited visible open space. Comparative alignment with existing datasets We overlaid the CSMD model with three independent products that captured different dimensions of urban deprivation. Figure 6 reported country-level precision, recall, and F1 for one representative configuration per dataset; additional boxplots are provided in supplementary material. SSI showed the strongest alignment with CSMD. The SpaceDef indicator, which captures overcrowding and inadequate living space, achieved high recall because almost all CSMD-deprived segments also exhibited SpaceDef deprivation (recall = 0.88), while precision was moderate (0.56) because SpaceDef classified a broader set of segments as deprived. Consequently, the share of segments and population classified as deprived was higher in SSI than CSMD (~43% vs ~35% of segments; ~43% vs ~31% of population). MN showed more modest alignment, precision and recall were low (0.35 & 0.32), indicating that only a minority of segments labelled deprived by MN were also labelled deprived by CSMD. MN identified a slightly smaller share of segments as deprived than CSMD but in more populous areas, consistent with its focus on street-network accessibility rather than broader built-form characteristics. WRI exhibited high recall but low precision (0.66 & 0.28). Many segments containing informal-subdivision or atomistic land-use classes also appeared in CSMD’s deprived set (recall =0.66), but WRI assigned these classes across a much wider portion of the urban region, leading to low precision (0.28) and substantially higher deprived coverage (~49% of segments; ~46% of population). Discussion Global picture Our results delivered comparable, neighborhood-scale estimates of spatially modelled morphological deprivation across 5,132 cities in 103 countries, showing how morphological deprivation was distributed across the urban hierarchy. By combining standardized city-segment units with simple morphology and access indicators 23 and harmonized, field-validated reference labels of morphologically deprived areas from IDEABench 24 , the analysis moved beyond case-specific studies that have constrained cross-city synthesis and largely focused on a handful of major cities 13,27–29 . In doing so, it complemented household-based slum indicators and survey-driven monitoring by emphasizing the spatial manifestation of disadvantage in urban form and access, responding to calls to bridge household conditions and neighborhood morphology through spatially modelled morphological deprivation 6,30 . Small and medium cities emerge as major hotspots of spatially modelled morphological deprivation once absolute population was taken into account, particularly in Africa and LAC. Globally, residents in morphologically deprived segments are present across the full urban hierarchy: small and medium cities together accounted for roughly one-third of all people living in deprived segments, while large cities and megacities comprised most of the remainder. In Africa, large and small cities (42M & 38M) host more residents in morphologically deprived segments than megacities (24M), even though relative shares peaked in the very largest agglomerations. In LAC, residents in morphologically deprived segments were relatively evenly distributed across size classes, with large and small cities contributing comparable totals. In Asia, by contrast, morphological deprivation was dominantly observed in megacities, which led both in share and absolute numbers. At the country level, many cases nevertheless showed most morphologically deprived segments concentrated in networks of small and intermediate cities rather than in a single dominant metropolis. This was consistent with national-scale analyses from Argentina, which found that extra-small and small urban areas collectively contain more informal-settlement area and experienced faster growth than the main metropolitan region 31 , and with evidence from secondary cities in Sub-Saharan Africa that small urban centers carried a disproportionate burden of overlapping deprivations in services, housing, and employment 32 . Together with long-standing reviews showing that EO-based slum mapping has concentrated on a handful of large or iconic settlements 13,27 , these patterns indicated that morphological deprivation was not exclusively a megacity phenomenon but a distributed condition across national settlement systems, reinforcing the need to examine neighborhood-scale conditions in secondary and intermediate cities using standardized spatial units. Alignment with existing datasets Comparisons with the three external products showed clear, interpretable patterns rather than random agreement. The SSI SpaceDef indicator 20 , which captures overcrowding and lack of living space, aligned most closely with CSMD, consistent with the idea that densely occupied housing often coincides with the built-form conditions the model learns. The Million Neighborhoods (MN) project 7 , which measures street-access complexity, overlapped partly with CSMD, suggesting that limited road access captures some but not all of the areas classified as morphologically deprived. The WRI intra-urban land-use product 21 , where we treated informal-subdivision and atomistic classes as proxies for informal built areas 33 , showed high recall but low precision relative to CSMD, reflecting a broader footprint of these land-use classes than the more selective set of segments that model labelled as morphologically deprived. Together, these results highlighted that agreement with CSMD depended on what each dataset was designed to measure; overcrowding, access, or land-use morphology, rather than implying a single, ordered ranking of deprivation maps. SSI, MN, WRI, and CSMD are part of the new generation of global products made possible by near-complete building, road, and EO datasets; CSMD added a city-segment-based, multi-city model spanning more than 5,000 cities, including many that rarely appeared in deprivation case studies. None of these layers should be treated as a definitive ground truth; instead, their overlaps and divergences provided a contextual triangulation of complementary perspectives on urban deprivation. Implications for policy makers The global and cross-product patterns above have direct implications for where action on morphological deprivation is needed and who is expected to act. A substantial share of residents living in morphologically deprived segments are located in small and medium cities rather than only in the largest metropolitan areas (Figures 3,4), meaning that local governments with relatively narrow tax bases and limited technical staff are responsible for these regions. Evidence from secondary cities in Sub-Saharan Africa shows that such centers often combine rapid growth with high levels of multiple deprivation but attract less policy and investment attention than primary cities 32 , while global assessments of basic service provision, such as the What a Waste 2.0 report, highlight how core urban functions like solid waste management can already strain municipal budgets and administrative capacity in low- and middle-income settings 34 . In this context, acting on the morphological deprivation patterns revealed by CSMD is less a matter of simply identifying hotspots than of recognizing that many of them are located in cities that face tight fiscal, staffing, and data constraints and are only weakly represented in national and international urban agendas. Within these constraints, the main value of the CSMD model and the City Segments v1 layer for policy makers lies in providing both a first-pass picture of intra-urban inequality and a practical spatial scaffold for further work. At the global and national scales, CSMD makes visible where morphologically deprived segments are concentrated within and across cities, including many small and medium cities that currently lack any neighborhood-scale mapping, helping ministries, development partners, and civil-society organizations to identify which places and city types carry the largest burdens. At the city scale, the standardized segment layer offers a common spatial frame into which local actors can plug existing information on services, infrastructure, risk, and community priorities. In regions where SSI, MN, WRI or similar products are available, CSMD can be read alongside them to build a richer picture of deprivation; in many other cities and countries, it will be the only large-scale, neighborhood-level representation currently available. In this role, CSMD is best understood as a screening and prioritization tool, not a substitute for local assessments, slum enumeration, or participatory planning, with areas of agreement between CSMD, other datasets, and available local data indicating high-confidence concentrations of deprivation and discrepancies pointing to where closer investigation is needed. Used in this way, the CSMD model and City Segments v1 can complement household-based indicators and official statistics by helping to locate and characterize morphologically deprived areas in support of monitoring and targeting efforts under SDG 11.1.1, especially in the many smaller cities that are currently under-represented in global reporting. Conclusion We developed a city-segment deprivation (CSMD) model that combines standardized built-environment indicators with multi-city reference labels to map spatially modelled morphological deprivation across 5,132 cities in 103 countries, providing the first large-scale, neighborhood-scale view of morphological deprivation across the urban hierarchy in the Global South. The resulting patterns show that morphological deprivation is a distributed condition present in all city-size classes, with small and medium cities accounting for a substantial share of residents in morphologically deprived segments rather than megacities alone. Comparisons with existing products indicate that alignment with CSMD follows each product’s conceptual focus, overcrowding, access, or land-use morphology, highlighting that no single representation captures all facets of urban deprivation and that CSMD adds a complementary, multivariate built-environment lens. At the same time, many of the morphologically deprived segments highlighted by the model fall under the responsibility of small and intermediate cities with constrained fiscal and technical capacity, where detailed local mapping is scarce or absent. In this context, CSMD model and City Segments v1 layer are best viewed as a screening and prioritization baseline that can be combined with local assessments and sectoral data where available, helping to locate and characterize deprived areas and to support more spatially explicit monitoring and targeting. Limitations Our estimates inherit constraints from the City Segments v1 dataset and its inputs. City Segments v1 spans 107 countries but omits several substantial urban populations (e.g., China, Chile, and parts of Asia), so the underlying spatial coverage is not globally exhaustive. From within this dataset, our analysis focuses on the 103 countries across Africa, Asia, and Latin America and the Caribbean, reflecting the Global South emphasis of the study. These 103 countries contain 5,132 cities for which City Segments v1 provides polygons and indicators. Importantly, these 5,132 cities represent only those urban centers for which the City Segments workflow could generate segments from available road data; many small cities with sparse or incomplete road networks were excluded upstream. As a result, our estimates for small cities apply to this subset rather than to all small urban settlements present in the underlying urban center database GHSL-UCDB 2019. Inputs to the City segments layer vary in completeness and may propagate into large and/or heterogeneous segments. In particular, OpenStreetMap roads and rivers are used to define segment boundaries, but coverage is uneven across regions and city sizes, leading in some cases to very large polygons or awkward delineations and skewed derived indicators. Recent global analyses of OpenStreetMap show that, although some regions exceed 80% completeness, many cities remain below 20% coverage, especially outside Europe and North America, highlighting spatial bias risks in global applications 35 . Population counts from GHS-POP 2023 may also underestimate residents in informal areas; multiple gridded products systematically undercount slum populations, sometimes capturing only a fraction of residents 36,37 implying that figures for population residing within deprived segments likely represent lower bounds. We also note a small number of countries (Laos, Turkmenistan, Kazakhstan) in our outputs with near-zero shares of population in deprived segments; given local knowledge, these are likely false negatives driven by data gaps or segmentation artefacts (e.g., very large mixed polygons) and should be interpreted cautiously. Model training relied on IDEABench labels from eight cities whose population sizes span approximately three to 25 million. This concentrates model learning on the upper end of the urban hierarchy and may under-represent settlement patterns typical of small and medium cities, although megacities also contain diverse neighbourhood morphologies, which partially mitigates this bias. Our label aggregation (> 30% segment overlap with deprived class) and the binary simplification (deprived/non-deprived, with non-built-up grouped with non-deprived) abstract complex realities, particularly in mixed segments where deprived and non-deprived areas co-exist. As an ensemble, the Random Forest can capture non-linearities but does not yield causal relations. The CSMD model itself is built solely from built-environment and access proxies (morphology, roads, population density) and does not directly observe income, tenure security, or service provision, so it should be interpreted as capturing the spatial imprint of morphological deprivation rather than all of its socio-economic dimensions. National and peripheral coverage is constrained by GHSL-UCDB 2019 urban extents used in the construction of City Segments v1; all segment footprints and population counts are defined within these 2019 urban-centre boundaries. Urban expansion beyond the 2019 extents, including more recent peri-urban growth, is not included and therefore is not reflected in our deprivation estimates. Automated segmentation can produce very large polygons in complex areas (e.g., airports), potentially blending adjacent deprived and non-deprived areas. Comparisons with SSI, MN, and WRI are contextual rather than validation exercises given different concepts, units, and geographies; reported overlap depends on aggregation choices (e.g., thresholds, metrics) and each dataset’s spatial coverage. Methods Overview This study developed a globally consistent framework for mapping spatially modelled morphological deprivation using harmonized geospatial datasets and supervised learning. Two complementary data sources supported the analysis. The City Segments v1 dataset 23 provided standardized neighborhood-scale spatial units and associated built-environment indicators across thousands of cities, while IDEABench 24 offered field-validated labels of morphological deprived and non-deprived areas derived from multi-sensor Earth observation imagery. We used IDEABench labels from eight cities to train a Random Forest classifier on the City Segments indicators, including a categorical regional variable to capture broad contextual variation, and then applied the resulting model to 5,132 cities in 103 countries across Africa, Asia, and Latin America and the Caribbean (LAC). Feature selection was performed using the Variable Selection Using Random Forests (VSURF) algorithm 38 . Predicted segment-level morphological deprivation was aggregated to city, national, region, and city-size classes to summarize how the population residing in morphologically deprived segments is distributed across the urban hierarchy. Predicted morphological deprivation patterns were subsequently compared, in overlapping geographies with three existing spatial datasets, the Slum Severity Index 20 , Million Neighborhoods project 7 , and WRI Intra-Urban LULC dataset 21 , to contextualize spatial patterns of deprivation across different conceptualizations. Detailed model parameters, evaluation metrics, and sensitivity analyses are presented in the Supplementary Material. City segments dataset The City Segments v1 dataset 23 formed the spatial foundation of this study. City segments represent standardized neighborhood-scale polygons that delineate coherent urban units bounded by navigable roads and major water bodies. Each segment contained a minimum population of approximately 400 residents, ensuring demographic significance while maintaining sufficient spatial granularity to capture intra-urban variation. The segmentation approach combined population, infrastructure, and hydrographic data to produce spatially meaningful and internally comparable units across diverse city contexts. Urban extents were first defined using the Global Human Settlement Layer Urban Centre Database (GHSL-UCDB 2019), after which OpenStreetMap (via GeoFabrik) road and water networks were used to partition these areas into contiguous blocks. Iterative adjustments based on the GHS-POP 2023 population grid ensured that each segment met the demographic threshold, while the inclusion of building footprints from the Overture Maps Foundation refined the geometry and representation of the built environment. City Segments v1 spans 107 countries globally, but the present analysis focused on cities in Africa, Asia, and LAC. This restriction within these three regions resulted in 103 countries and 5,132 cities, reflecting both the Global South focus of the study and the fact that all IDEABench training cities are located in Africa, Asia, or LAC. Urban centers in four countries outside these regions were not considered further. A complete list of countries included is provided in Extended Data Table 2. For each city segment, we considered a set of 21 built-environment indicators available in the City Segments dataset, comprising 11 absolute variables and 10 ratio indices that characterize the demographic, infrastructural, and morphological structure of the urban environment. These indicators are grouped into four domains: Population and Area (e.g., total population, segment area, population density), Roads and Connectivity (e.g., total road length, population-to-road ratios), Parcels and Access (e.g., number of parcels, proportion without road access), and Buildings and Morphology (e.g., total built area, average building size). The derived ratios (i 1 – i 10 ) normalize these measures by population, area, or parcel count, capturing relative density, accessibility, and variability in urban form. Together, these indicators served as predictor variables for modelling CSMD across the Global South. Full definitions, computation formulas, and units of all 21 indicators are provided in Extended Data Table 1. IDEABench benchmark dataset IDEABench 24 provided the labelled reference data required to train and validate the supervised model of spatially modelled morphological deprivation (CSMD). Developed under the European Space Agency-funded IDEAtlas project, the dataset was co-designed with local partners and community stakeholders to support open, transparent mapping of morphologically deprived urban areas. It integrates multi-sensor Earth observation imagery with expert- and community-validated annotations to serve as a benchmark for developing and testing AI-based methods for deprivation mapping. The dataset spanned eight cities in the Global South across Africa, Asia, and LAC: Nairobi, Lagos, Mexico City, Buenos Aires, Salvador, Medellin, Jakarta, and Mumbai, capturing diverse morphological and socio-economic conditions. For each city, IDEABench provided 128 x 128 pixel image patches derived from Sentinel-1 and Sentinel-2 satellites, supplemented with a built-up density layer generated from the Google Open Buildings dataset. Each patch was accompanied by a three-class reference label distinguishing deprived urban areas, non-deprived urban areas, and non-built-up areas. For this study, IDEABench reference labels were aggregated to the city-segment scale to match the spatial framework of the City Segments dataset. Patch-based labels were first converted into city-wide rasters for each of the eight training cities and overlaid with the corresponding city-segment polygons. A segment was assigned a label of deprived (1) if more than 30% of its area was classified as a deprived urban area in IDEABench; otherwise, it was labelled non-deprived (0). Non-built-up areas were grouped with non-deprived segments, as the objective was to distinguish morphologically deprived versus all other urban areas within the urban extent. This integration produced a harmonized segment-level training set linking the 21 built-environment indicators from City Segments to binary deprivation labels across eight cities and three world regions. Modelling Framework A supervised learning framework was employed to predict spatially modelled morphological deprivation at the city-segment scale. The model integrated built-environment indicators from the City Segments dataset with deprivation labels derived from IDEABench, enabling a globally consistent yet regionally adaptive analysis. We trained a Random Forest (RF) classifier on labelled segments from the eight IDEABench cities (refer to Supplementary Table 1) and then applied the resulting model to all city segments in the 5,132 cities and 103 countries in Africa, Asia, and LAC described above. This RF classifier is referred to as the city-segment deprivation (CSMD) model throughout the paper. A categorical regional variable (Africa, Asia, LAC) was included as an additional predictor to capture broad contextual variation across world regions (refer to Supplementary Table 1). Feature selection was performed using the Variable Selection Using Random Forests (VSURF) algorithm 38 . VSURF was run on the full set of 21 built-environment indicators plus regional variable (total 22) from City Segments and identified a parsimonious subset of ten indicators that contributed most strongly to predicting morphological deprivation (refer to Supplementary section 1.2). The final set of ten predictors used in the CSMD model are i5_par_area (average parcel area), i1_pop_area (population density), B_AVG_SEG (average building area), i9_roads_par (road length per parcel), i6_paru_area (average area of untouched parcels), i8_paru_par (proportion of untouched parcels), PARU_A_SEG (median area of untouched parcels), B_AREA_SEG (total building footprint area), B_CV_SEG (coefficient of variation of building area) and REG1_GHSL (the regional code variable). The Random Forest was implemented in Python using the scikit-learn library. We fitted the model using a fixed 80/20 train-test split at the segment level within the eight IDEABench cities, with balanced class weighting to mitigate the minority status of morphologically deprived segments (refer to Supplementary section 1.3). Model performance on the held-out test data was evaluated using standard classification metrics, including precision, recall, F1-score, and balanced accuracy (refer to Supplementary section 1.4). The trained CSMD model was then applied to all city segments in the 5,132 cities to generate binary deprivation classifications. Full details of the RF steup, hyper-parameter tuning, ROC curves, threshold sweeps, confusion matrices, and feature-importance plot are provided in Supplementary section 1. City-size analysis To examine how spatially modelled morphological deprivation varied with urban scale, we aggregate segment-level predictions from the CSMD model to the level of individual cities and classified each city according to its population size. Following the World Urbanization Prospects 2018 classification, cities were grouped into five size categories: small (< 500,000 inhabitants), medium (500,000 – 1 million), large (1 – 5 million), very large (5 – 10 million), and megacity (≥ 10 million) 26 . Population per segment was taken from the City Segments dataset, where it is one of the 21 built-environment indicators derived from the GHS POP 2023 dataset. For every city, total population was computed as the sum of segment-level populations, and deprived population was calculated as the sum of population within segments classified as morphologically deprived by the CSMD model. The proportion of morphologically deprived population was then derived as the ratio of deprived to total population. For each country, regional, and global summary, we aggregated total and deprived populations across cities within each size class. These statistics provided the basis for the regional and national patterns reported in the Results section, including comparisons of the magnitude and share of morphologically deprived populations across the urban hierarchy (Figures 2, 3 & 4). Comparative analyses To contextualize the spatial patterns produced by the CSMD model, we compared its segment-level classifications with three existing datasets that capture different dimensions of urban deprivation: the Slum Severity Index (SSI) 20 , the Million Neighbourhoods (MN) project 7 , and the World Resource Institute (WRI) Intra-Urban Land-Use/Land-Cover dataset 21 . These comparisons were restricted to cities and countries where both CSMD and the respective external dataset were available; sub-Saharan Africa for SSI and MN, and a set of over 100 cities worldwide for WRI and were interpreted as contextual alignment across concepts rather than formal validation. For each dataset, we derived a binary ‘deprived/non-deprived’ label at the city-segment level, intersected these labels with the CSMD classifications, and computed country-level precision, recall, and F1 scores, as well as the proportions of segments and population classified as deprived by each source. These metrics formed the basis of the comparative alignment summaries reported in the results (Figure 6). The SSI dataset was produced by operationalizing the four UN-Habitat slum indicators; water, sanitation, housing durability, and living space adequacy (SpaceDef) and provided each component and a combined SSI index (0-4) at 100 m resolution across sub-Saharan Africa. Country-level SSI raster’s (five bands: water, sanitation, housing, SpaceDef, combined SSI) were clipped to the extent of the City Segments polygons, and segments were retained only if they contained at least one valid SSI pixel. For each segment, we calculated the proportion of valid pixels with SpaceDef > 0 and used this component as the SSI-based indicator of deprivation. Segments were classified as deprived by SSI when this proportion exceeded a threshold τ; we evaluated τ = 0.1, 0.2, and 0.3 and used τ = 0.1 as the representative configuration in our comparative analyses. Detailed results for alternative SSI components and thresholds are provided in the Supplementary material (Section 2.3 & Figure 6). The MN dataset reports a building-to-street depth index k at the block level across sub-Saharan Africa, where higher k values indicate poorer street access and greater infrastructural inaccessibility. MN blocks were harmonized to WGS84 and overlaid with the City Segments polygons after geometry checks. Blocks with k > 0.3 were treated as access-deficit, and for each city segment we calculated the proportion of intersecting blocks above this threshold using a centroid-majority rule with area-based fallbacks where needed (refer to Supplementary section 2.4.2). Segments were then classified as deprived by MN when this proportion exceeded a threshold τ, taken as τ = 0.1 for the comparative analyses. Sensitivity of the alignment metrics to alternative k cut-off and segment-level thresholds is reported in the Supplementary Material (Section 2.4 & Figure 7). The WRI Intra-Urban LULC dataset has coverage beyond sub-Saharan Africa (over 100 countries in 48 countries that overlap with City Segments) and followed Friesen et al. (2025) in treating the ‘informal subdvision’ and ‘atomistic’ classes as proxies for informal built areas. WRI rasters (5 m) were mosaicked and clipped to the city-segment boundaries for all cities where both datasets were available. For each segment, we computed the proportion of valid pixels belonging to these informal subdivision or atomistic classes and used this proportion as the WRI-based indicator of deprivation. Segments were classified as deprived by WRI when this proportion exceed τ = 0.1, Detailed sensitivity analyses exploring alternative thresholds are provided in the Supplementary Material (Section 2.5 & Figure 8). Declarations Acknowledgements This research was supported by funding from FORMAS (Swedish Research Council for Sustainable Development) under grant no. 2023-01210 (DEPRIMAP). The computation (model training) was partly enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by the Swedish Research Council through grant agreement no. 2022-06725. The IDEABench benchmark dataset is an output enabled by the foundational support of the IDEAtlas project (https://ideatlas.eu/). We formally acknowledge the essential contributions of the local co-anchors and collaborating institutions, who were responsible for the detailed data curation, collection, and validation necessary for the IDEABench work. Author contributions SGV, SG, DRT conceptualized the study. SGV led the data analysis and drafted the paper. SGV, SG, DRT, JF contributed to the methodological design and paper revisions. AMB, DRT contributed to City Segments v1 dataset. BT, MK, CP, RVM, AA contributed to IDEABench dataset. SGV, JF prepared the figures. DRT, JF, SG supported data interpretation. SG and DRT supervised the overall project. All authors contributed to discussions, provided critical feedback and approved the final version of the paper. Data availability All data and code used for preprocessing, modelling, and figure generation are publicly available in the project’s GitHub repository (https://github.com/saiga143/citysegmentdeprivation). A Zenodo archive linked to this repository (https://doi.org/10.5281/zenodo.17637298) provides all large files, including the trained Random Forest model and full prediction outputs that cannot be hosted on GitHub due to size limits. The City Segments v1 dataset is publicly available at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/XLRSF0. The IDEABench Reference dataset are available at https://phys-techsciences.datastations.nl/dataset.xhtml?persistentId=doi:10.17026/PT/X4NJII. External products used for comparative analysis were obtained from their respective sources: SSI from https://zenodo.org/records/14998570, MN from https://www.millionneighborhoods.africa/download and WRI Urban Land Use dataset via Google Earth Engine at https://code.earthengine.google.com/?asset=projects/wri-datalab/urban_land_use/V1. Code availability All code is publicly available via GitHub (https://github.com/saiga143/citysegmentdeprivation) and Zenodo (https://doi.org/10.5281/zenodo.17637264). References UN Habitat. UN Habitat’s 2024 Annual Report, Adequate Housing for All . UN-Habitat https://unhabitat.org/annual-report-2024 (2025). UN-Habitat. Implementation of the Strategic Plan for the Period 2020–2025 . United Nations https://unhabitat.org/sites/default/files/2025/04/2503673e.pdf (2025). Kundu, D. & Pandey, A. K. World Urbanisation: Trends and Patterns. in Developing National Urban Policies 13–49 (Springer Nature Singapore, Singapore, 2020). doi:10.1007/978-981-15-3738-7_2. Grossmann, K. & Mallach, A. The small city in the urban system: complex pathways of growth and decline. Geogr Ann Ser B 103 , 169–175 (2021). Wagner, M. & Growe, A. Research on Small and Medium-Sized Towns: Framing a New Field of Inquiry. World 2 , 105–126 (2021). Abascal, A. et al. “Domains of deprivation framework” for mapping slums, informal settlements, and other deprived areas in LMICs to improve urban planning and policy: A scoping review. Comput Environ Urban Syst 93 , 101770 (2022). Bettencourt, L. M. A. & Marchio, N. Infrastructure deficits and informal settlements in sub-Saharan Africa. Nature 645 , 399–406 (2025). Georganos, S., Vanhuysse, S., Abascal, A. & Kuffer, M. Extracting Urban Deprivation Indicators Using Superspectral Very-High-Resolution Satellite Imagery. in 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS 2114–2117 (IEEE, 2021). doi:10.1109/IGARSS47720.2021.9554849. Kuffer, M. et al. Mapping the Morphology of Urban Deprivation. in Urban Remote Sensing 305–323 (Wiley, 2021). doi:10.1002/9781119625865.ch14. Wang, J. et al. EO + Morphometrics: Understanding cities through urban morphology at large scale. Landsc Urban Plan 233 , 104691 (2023). Owusu, M., Engstrom, R., Thomson, D., Kuffer, M. & Mann, M. L. Mapping Deprived Urban Areas Using Open Geospatial Data and Machine Learning in Africa. Urban Science 7 , 116 (2023). UN Habitat. The Urban SDG Monitoring Series | Monitoring SDG Indicator 11.1.1 . UN-Habitat www.unhabitat.org. (2019). Kuffer, M., Pfeffer, K. & Sliuzas, R. Slums from Space—15 Years of Slum Mapping Using Remote Sensing. Remote Sens (Basel) 8 , 455 (2016). do Nascimento, G. A., Giannotti, M., Regueira, T. A. & Tomasiello, D. B. Identifying slum areas: A multidimensional analysis leveraging with explanatory machine learning techniques. Sustain Cities Soc 131 , 106645 (2025). Warren, J. L., Perez-Heydrich, C., Burgert, C. R. & Emch, M. E. Influence of Demographic and Health Survey Point Displacements on Point-in-Polygon Analyses. Spat Demogr 4 , 117–133 (2016). Perez-Heydrich, C., Warren, J. L., Burgert, C. R. & Emch, M. E. Influence of Demographic and Health Survey Point Displacements on Raster-Based Analyses. Spat Demogr 4 , 135–153 (2016). Pettersson, M. B. & Daoud, A. Leveraging Compact Satellite Embeddings and Graph Neural Networks for Large-Scale Poverty Mapping . arXiv http://arxiv.org/abs/2511.01408 (2025). Georganos, S., Hafner, S., Kuffer, M., Linard, C. & Ban, Y. A census from heaven: Unraveling the potential of deep learning and Earth Observation for intra-urban population mapping in data scarce environments. International Journal of Applied Earth Observation and Geoinformation 114 , 103013 (2022). Tatem, A. & Espey, J. Global population data is in crisis – here’s why that matters. The Conversation Preprint at https://doi.org/10.64628/AB.c64mdcqqs (2025). Li, C. et al. Mapping urban slums and their inequality in sub-Saharan Africa. Nature Cities 1–12 (2025) doi:10.1038/s44284-025-00276-0. Guzder-Williams, B., Mackres, E., Angel, S., Blei, A. M. & Lamson-Hall, P. Intra-urban land use maps for a global sample of cities from Sentinel-2 satellite imagery and computer vision. Comput Environ Urban Syst 100 , 101917 (2023). Mari Rivero, I. et al. GHS-UCDB R2024A - GHS Urban Centre Database 2025. European Commission, Joint Research Centre (JRC) Preprint at https://doi.org/10.2905/1a338be6-7eaf-480c-9664-3a8ade88cbcd (2024). Thomson, D. R. & Blei, A. City Segments Layer V1 . Harvard Dataverse https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/XLRSF0 (2025) doi:10.7910/DVN/XLRSF0. Tereke, B. W. et al. IDEABench: Benchmark Dataset for Mapping Deprived Urban Areas . DANS Data Station Physical and Technical Sciences https://phys-techsciences.datastations.nl/dataset.xhtml?persistentId=doi:10.17026/PT/X4NJII (2025) doi:10.17026/PT/X4NJII. Kochupillai, M., Kahl, M., Schmitt, M., Taubenbock, H. & Zhu, X. X. Earth Observation and Artificial Intelligence: Understanding emerging ethical issues and opportunities. IEEE Geosci Remote Sens Mag 10 , 90–124 (2022). United Nations. World Urbanization Prospects 2018 Highlights . Department of Economic and Social Affairs, Population Division (2019). Mahabir, R., Croitoru, A., Crooks, A., Agouris, P. & Stefanidis, A. A Critical Review of High and Very High-Resolution Remote Sensing Approaches for Detecting and Mapping Slums: Trends, Challenges and Emerging Opportunities. Urban Science 2 , 8 (2018). Veeravalli, S. G. et al. Towards a Spatial Measure of SDG 11.1.1: Open Data for Urban Deprivation Mapping. in 2025 Joint Urban Remote Sensing Event (JURSE) 1–4 (IEEE, 2025). doi:10.1109/JURSE60372.2025.11076033. Veeravalli, S. G., Haas, J., Friesen, J. & Georganos, S. Understanding Informal Settlement Transformation through Google’s 2.5D Dataset and Street View based Validation. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences XLVIII-M-7–2025 , 245–251 (2025). Thomson, D. R. et al. Need for an Integrated Deprived Area “Slum” Mapping System (IDEAMAPS) in Low- and Middle-Income Countries (LMICs). Soc Sci 9 , 80 (2020). Samper, J. et al. Spatiotemporal Dynamics of Informal Settlements Across Argentine Cities: A National-Scale Analysis . https://www.ssrn.com/abstract=5588589 (2025) doi:10.2139/ssrn.5588589. Githira, D. et al. Analysis of Multiple Deprivations in Secondary Cities in Sub-Saharan Africa . UN-Habitat https://unhabitat.org/sites/default/files/2021/04/analysis_of_multiple_deprivations_in_secondary_cities_-_analysis_report.pdf (2020). Friesen, J., Georganos, S. & Haas, J. Differences in walking access to healthcare facilities between formal and informal areas in 19 sub-Saharan African cities. Communications Medicine 5 , 41 (2025). Kaza, S., Yao, L., Bhada-Tata, P. & Woerden, F. Van. What a Waste 2.0: A Global Snapshot of Solid Waste Management to 2050. Urban Development Series . World Bank https://openknowledge.worldbank.org/handle/10986/2174. (2018). Herfort, B., Lautenbach, S., Porto de Albuquerque, J., Anderson, J. & Zipf, A. A spatio-temporal analysis investigating completeness and inequalities of global urban building data in OpenStreetMap. Nat Commun 14 , 3985 (2023). Breuer, J. H. P., Friesen, J., Taubenböck, H., Wurm, M. & Pelz, P. F. The unseen population: Do we underestimate slum dwellers in cities of the Global South? Habitat Int 148 , 103056 (2024). Thomson, D. R. et al. Evaluating the Accuracy of Gridded Population Estimates in Slums: A Case Study in Nigeria and Kenya. Urban Science 5 , 48 (2021). Genuer, R., Poggi, J.-M., Tuleau-Malot, C. & Tuleau, C. VSURF: An R Package for Variable Selection Using Random Forests. R J 7 , 19–33 (2015). Additional Declarations There is NO Competing Interest. Supplementary Files ExtendedDataTable1.xlsx Extended Data Table 1 ExtendedDataTable2.xlsx Extended Data Table 2 VSGNatureCitiesSuppMaterialv3.docx Supplementary Material Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8189204","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":558242669,"identity":"2e125ac9-6fb4-4a5c-835b-481f0cc6799e","order_by":0,"name":"Sai Ganesh Veeravalli","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5klEQVRIie3RPQrCMBTA8UhBl4KrRbBXeEUQhOJZUgp11NHBIZMuBde4eAdv8KBgl6BrwKW9QcTFoajRQcQhfkwO+U/J8OO9EEJstr+sxh4nLCCErwhBCskn5CmkJHtP/HnGHHcSjpo8LpCOd90mp45SBgIi0kQkfS4T0Ivtey1J6x43ERKx2nKm9xF4JyHRxHFNiy1KTc4X8EWuNNmGvqTOsTI9RkaMHBgC5OltCvZAUtI2CZAlQ7WJIcjTsSZxNxDlzEuNiw3Lgk4H0Mka60JVg2CVx5k6mcbo8OX+9Lk2m81m+60riPRVwWr1TfkAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0002-1670-8703","institution":"Karlstad University","correspondingAuthor":true,"prefix":"","firstName":"Sai","middleName":"Ganesh","lastName":"Veeravalli","suffix":""},{"id":558242670,"identity":"54a52e87-836e-4ebc-bf02-36a0b21c53b9","order_by":1,"name":"Alejandro Blei","email":"","orcid":"","institution":"Independent Consultant","correspondingAuthor":false,"prefix":"","firstName":"Alejandro","middleName":"","lastName":"Blei","suffix":""},{"id":558242671,"identity":"b0e48fab-c1ab-4fed-a665-d89d1a04b73a","order_by":2,"name":"John Friesen","email":"","orcid":"https://orcid.org/0000-0003-2530-1363","institution":"University of Würzburg","correspondingAuthor":false,"prefix":"","firstName":"John","middleName":"","lastName":"Friesen","suffix":""},{"id":558242672,"identity":"571d7b89-d964-43a8-9ea2-7c3226302fd0","order_by":3,"name":"Bedru Tareke","email":"","orcid":"","institution":"University of Twente","correspondingAuthor":false,"prefix":"","firstName":"Bedru","middleName":"","lastName":"Tareke","suffix":""},{"id":558242673,"identity":"7ae0497b-d746-4d78-9829-762522e89fc8","order_by":4,"name":"Monika Kuffer","email":"","orcid":"https://orcid.org/0000-0002-1915-2069","institution":"twente","correspondingAuthor":false,"prefix":"","firstName":"Monika","middleName":"","lastName":"Kuffer","suffix":""},{"id":558242674,"identity":"69aaca16-986f-4ba2-a7a9-dba4b22dcf4e","order_by":5,"name":"Claudio Persello","email":"","orcid":"https://orcid.org/0000-0003-3742-5398","institution":"University of Twente","correspondingAuthor":false,"prefix":"","firstName":"Claudio","middleName":"","lastName":"Persello","suffix":""},{"id":558242675,"identity":"bcf214f9-fe62-4b3b-93ea-ac1457ccebe7","order_by":6,"name":"Raian Maretto","email":"","orcid":"","institution":"University of Twente","correspondingAuthor":false,"prefix":"","firstName":"Raian","middleName":"","lastName":"Maretto","suffix":""},{"id":558242676,"identity":"0c9a112f-01d7-4143-a6b0-4a6fd4b94678","order_by":7,"name":"Angela Abascal","email":"","orcid":"","institution":"Public University of Navarra","correspondingAuthor":false,"prefix":"","firstName":"Angela","middleName":"","lastName":"Abascal","suffix":""},{"id":558242677,"identity":"81f0ee92-e9b5-4049-aa62-ef48e1aa5476","order_by":8,"name":"Stefanos Georganos","email":"","orcid":"","institution":"Karlstad University","correspondingAuthor":false,"prefix":"","firstName":"Stefanos","middleName":"","lastName":"Georganos","suffix":""},{"id":558242678,"identity":"b5c0292b-81d3-4423-ba33-b574acc313f9","order_by":9,"name":"Dana R. Thomson","email":"","orcid":"","institution":"Columbia University","correspondingAuthor":false,"prefix":"","firstName":"Dana","middleName":"R.","lastName":"Thomson","suffix":""}],"badges":[],"createdAt":"2025-11-24 04:50:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8189204/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8189204/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":98423733,"identity":"5f2162a3-3303-4318-9cca-869dddf914c4","added_by":"auto","created_at":"2025-12-17 16:32:34","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1427621,"visible":true,"origin":"","legend":"\u003cp\u003eOverview of the City-Segment Deprivation (CSMD) Model\u003c/p\u003e","description":"","filename":"Figure1OverviewCSMD.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8189204/v1/b3df03ad4595e045a1d81add.jpg"},{"id":97948757,"identity":"f250579d-bd24-43f7-9c77-12e2e0cd0fc8","added_by":"auto","created_at":"2025-12-11 06:29:49","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":893645,"visible":true,"origin":"","legend":"\u003cp\u003eGlobal and regional distribution of population residing within morphologically deprived city segments. Share of population living in morphologically deprived segments across 103 countries (\u0026amp; 5,132 cities) spanning Africa, Asia, and Latin America and the Caribbean (LAC). The map depicts national-level percentages, with darker shades indicating higher shares. Regional summaries below the map show the proportion and total population residing in deprived versus non-deprived segments. Africa records the highest relative share (28.2%, 144 M of 512 M), followed by LAC (17.0%, 57 M of 338 M) and Asia (13.3%, 147 M of 1108 M).\u003c/p\u003e","description":"","filename":"Figure2GlobalDeprivedShareZoomed.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8189204/v1/3e233a61b93accf2c8c63824.jpg"},{"id":97948752,"identity":"d85942fb-d309-4250-9557-8253b915fc6b","added_by":"auto","created_at":"2025-12-11 06:29:49","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":690101,"visible":true,"origin":"","legend":"\u003cp\u003eNational variation in the share of population residing within deprived city segments across Africa, Asia, and Latin America and the Caribbean (LAC). Each panel shows country-level percentages of population residing in deprived city segments, with point size proportional to the absolute population (in million). Vertical dashed lines indicate regional averages. Insets summarize the distribution of total population by city-size (small to megacity) within each region, distinguishing between populations residing in deprived and non-deprived segments.\u003c/p\u003e","description":"","filename":"Figure3Lollipopswithcitysize.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8189204/v1/143b910a8fa27b9bc92f83d7.jpg"},{"id":98422609,"identity":"fc4ee534-c02e-42b5-b274-cd6f043b5af9","added_by":"auto","created_at":"2025-12-17 16:31:14","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":303712,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of populations residing within deprived city segments by city size and region. Each horizontal bar represents a country, partitioned by the share of population in deprived segments across five UN-defined city-size categories (small, medium, large, very large, megacity). The numbers represent absolute counts in either thousands (k) or millions (M).\u003c/p\u003e","description":"","filename":"Figure4DepAcrossSizes.png","url":"https://assets-eu.researchsquare.com/files/rs-8189204/v1/610f003ec9ee8f84a7855fe8.png"},{"id":97948758,"identity":"7d8539a8-b62d-46fe-94a4-eadaf429b81b","added_by":"auto","created_at":"2025-12-11 06:29:49","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":2473670,"visible":true,"origin":"","legend":"\u003cp\u003eSpatial distribution of morphologically deprived and non-deprived city segments in six small cities. Examples from Africa (Kisumu, Juba), Asia (Sambhal, Myaungmya), and LAC (Ocumare del Tuy, Ouanaminthe) highlight the diversity of spatial arrangements, with deprived segments ranging from localized pockets to larger contiguous areas.\u003c/p\u003e","description":"","filename":"Figure5Smallcityplots.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8189204/v1/fa520bc7f7fe5273de336bc5.jpg"},{"id":98421878,"identity":"5de78d99-5c06-4e29-a3c8-2b08687938fd","added_by":"auto","created_at":"2025-12-17 16:29:50","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1336289,"visible":true,"origin":"","legend":"\u003cp\u003eComparative alignment between CSMD labels and three external datasets. Country-level boxplots show precision, recall, and F1 (at τ = 0.1) for: (top) SSI using the SpaceDef (overcrowding) indicator; (middle) MN using k \u0026gt; 0.3; and (bottom) WIR using the share of ‘informal subdivision’ + ‘atomistic’ pixels. Bars to the right report, for each dataset, the proportion of segments (upper bar) and population (lower bar) classified as deprived, shown alongside the corresponding CSMD proportions for direct comparison.\u003c/p\u003e","description":"","filename":"Figure6ThreeComparisons.png","url":"https://assets-eu.researchsquare.com/files/rs-8189204/v1/be942777a757df4a2d7d6e91.png"},{"id":98623012,"identity":"29d76ff5-70b9-46ac-a006-a93b4bd92afa","added_by":"auto","created_at":"2025-12-19 17:04:02","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":7480299,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8189204/v1/c0017317-d23a-42d3-b9f9-27e397cd620e.pdf"},{"id":97948751,"identity":"d0bebc6c-3eb4-492a-8ad6-be8bbbaaa865","added_by":"auto","created_at":"2025-12-11 06:29:49","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":10379,"visible":true,"origin":"","legend":"\u003cp\u003eExtended Data Table 1\u003c/p\u003e","description":"","filename":"ExtendedDataTable1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8189204/v1/a8f609e2305bded73c4bd1a2.xlsx"},{"id":97948754,"identity":"0ff4f877-a364-4c9a-af9d-8b65506860cd","added_by":"auto","created_at":"2025-12-11 06:29:49","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":16735,"visible":true,"origin":"","legend":"\u003cp\u003eExtended Data Table 2\u003c/p\u003e","description":"","filename":"ExtendedDataTable2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8189204/v1/50d3c838335cec7fb3d1488e.xlsx"},{"id":97948759,"identity":"25887a77-1d05-422b-97ad-3a11319bed69","added_by":"auto","created_at":"2025-12-11 06:29:49","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":607129,"visible":true,"origin":"","legend":"Supplementary Material","description":"","filename":"VSGNatureCitiesSuppMaterialv3.docx","url":"https://assets-eu.researchsquare.com/files/rs-8189204/v1/f52c66acbaafb0a59adb12c4.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"The Hidden Burden of Morphological Deprivation in Small and Medium Cities","fulltext":[{"header":"Introduction","content":"\u003cp\u003eRapid urbanization across low- and middle-income countries (LMICs) is transforming the spatial and social fabric of cities, yet deprivation remains widespread. More than one billion people, around a quarter of the world\u0026rsquo;s urban population, are estimated to live in slums and informal settlements lacking secure tenure, durable housing, and reliable access to basic services\u003csup\u003e1\u003c/sup\u003e, and 1.6 billion people are estimated to live in inadequate housing\u003csup\u003e2\u003c/sup\u003e. \u0026nbsp;Addressing these conditions is central to Sustainable Development Goal (SDG) 11.1, which calls for adequate, safe, and affordable housing for all and the upgrading of slums by 2030. Despite decades of policy attention, the spatial distribution and intensity of urban deprivation remain poorly understood, particularly in small and medium-sized cities that now absorb a substantial, though not yet well quantified, share of urban growth across the Global South\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e3\u0026ndash;5\u003c/span\u003e\u003c/sup\u003e. These settlements are often invisible in official statistics due to coarse administrative reporting units and their small spatial footprint, constraining government\u0026rsquo;s capacity to target infrastructure, housing, and climate-adaptation investments\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e6,7\u003c/span\u003e\u003c/sup\u003e. Advances in Earth observation and machine learning techniques have made it possible to identify the physical signatures of deprivation from space\u003csup\u003e8\u0026ndash;11\u003c/sup\u003e, however, global evidence remains fragmented, locally calibrated, and seldom comparable across regions, a gap this study seeks to address.\u003c/p\u003e\n\u003cp\u003eThe term deprivation has been used in different ways across urban research and policy, increasingly as a broad concept aligned with SDG 11.1.1 inclusive of \u0026lsquo;slums\u0026rsquo;, \u0026lsquo;informal settlements\u0026rsquo;, and inadequate housing. The UN-Habitat definition of a slum household, based on deficits in water, sanitation, durability, crowding, and tenure, remains the official SDG and most widely applied standard\u003csup\u003e12\u003c/sup\u003e. While valuable for national monitoring, this household-based measure does not capture the spatial or morphological dimensions of deprivation within cities\u003csup\u003e6,13\u003c/sup\u003e. Informal settlements, in UN-Habitat\u0026rsquo;s sense, are areas where residents lack secure tenure and adequate services and where the built environment often falls outside planning and building regulations, even if their physical form varies\u003csup\u003e12\u003c/sup\u003e. At the same time, many formally recognized neighborhoods remain under-serviced and thus deprived. To encompass these variations, recent work conceptualizes Deprived Urban Areas (DUAs) as the spatial manifestation of overlapping social, infrastructural, and environmental disadvantages \u003csup\u003e6,9\u003c/sup\u003e. In this study, we adopt an area-based, spatially modelled deprivation, an expression of disadvantage in urban morphology, services, and infrastructure consistent with the IDEAMAPS \u0026lsquo;Domains of Deprivation\u0026rsquo; framework and related work on area-based inequality\u003csup\u003e6,11,14\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eConventional estimates of urban deprivation rely heavily on population censuses and household surveys that inform UN-Habitat\u0026rsquo;s global household-based slum indicator. In its updated framing of SDG 11.1.1, UN-Habitat distinguishes three related aspects of slums, informal settlements, and inadequate housing and recommends that national statistical systems use census and survey data to monitor household-level deficits, while acknowledging the need for spatial approaches to capture area-based manifestations such as informal settlements\u003csup\u003e12\u003c/sup\u003e. While household datasets are invaluable for assessing living-condition deficits, they are not designed for intra-urban spatial analysis. Demographic Health Surveys (DHS) cluster coordinates, for example, are intentionally displaced by up to two kilometers in urban areas to protect respondent privacy, introducing spatial uncertainty that limits their compatibility with satellite imagery and local mapping\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e15\u0026ndash;17\u003c/span\u003e\u003c/sup\u003e. Even where household data are available, the scale mismatch between individual surveys and neighborhood morphology remains substantial. The UN-Habitat definition identifies slum households, whereas mapping deprivation requires delineating deprived neighborhoods shaped by collective built environment, infrastructure and access patterns\u003csup\u003e6,13\u003c/sup\u003e. Moreover, censuses are infrequent and often exclude informal settlements, leaving large populations statistically invisible\u003csup\u003e18\u003c/sup\u003e. These challenges are likely to intensify as large-scale survey programmes face cuts and uncertainty, including the recent termination of the long-running DHS programme\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e19\u003c/span\u003e\u003c/sup\u003e. Consequently, survey-derived indicators do not capture where deprivation concentrates or how it evolves, highlighting the need for spatially modelled approaches that integrate Earth observation (EO) and morphometric data to characterize the built characteristics of disadvantage.\u003c/p\u003e\n\u003cp\u003eSeveral recent initiatives have employed Earth observation and geospatial methods to capture different spatial dimensions of urban deprivation, ranging from access to services to infrastructural inclusion and built-form informality. The Slum Severity Index (SSI)\u003csup\u003e20\u003c/sup\u003e models household-level service deprivation using machine learning on DHS indicators, extending the UN-Habitat framework to produce national-scale estimates across Sub-Saharan Africa (SSA), but its primary focus is on service and housing conditions rather than neighborhood morphology. The Million Neighborhoods (MN) project\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e7\u003c/span\u003e\u003c/sup\u003e also focusing on SSA, instead emphasizes infrastructure inclusion, mapping block-level connectivity based on the accessibility of buildings to the road network. The World Resource Institute (WRI) Intra-urban LULC dataset\u003csup\u003e21\u003c/sup\u003e offers a morphology-driven perspective, classifying residential areas into formal, informal-subdivision, and atomistic (small, disconnected housing clusters typically lacking organized street networks) typologies. Collectively, these datasets represent complementary approaches to quantifying urban deprivation, yet differ in their conceptual scope, spatial coverage, data sources, and validation. In this study, we compare our results with these three products to situate our model within this evolving landscape of multi-dimensional deprivation mapping. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo address these limitations, we developed a standardized, spatial deprivation model reflecting aspects of unplanned urbanization, service access, and infrastructure for cities across Africa, Asia, and Latin America \u0026amp; the Caribbean (LAC). Across these regions, 3.2 billion people live in cities (2.5 billion, excluding China) according to the most recent Global Human Settlement Urban Centers Database\u003csup\u003e22\u003c/sup\u003e. Our approach combines globally consistent neighborhood-scale spatial units from the City Segments Layer v1\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e23\u003c/span\u003e\u003c/sup\u003e, which contains building, road, and population indicators, with validated reference labels from IDEABench\u003csup\u003e24\u003c/sup\u003e, the first multi-city benchmark of morphologically deprived and non-deprived areas derived from Sentinel imagery, Google Open Buildings, and locally validated field data. Using these datasets, we trained a Random Forest classifier on segment-level data from eight IDEABench cities and then applied the resulting model to City Segments v1 across 5,132 cities in 103 countries to generate globally comparable estimates of morphological deprivation. The model relied exclusively on open, spatially aggregated, anonymized data and focused on physical structure rather than individual households or named communities, which is consistent with emerging standards for responsible and equitable use of AI in urban mapping\u003csup\u003e25\u003c/sup\u003e. \u0026nbsp;By covering small, medium, large, very large cities and megacities, the analysis provided, to our knowledge, the first spatially explicit, globally comparable evidence on the distribution of morphological deprivation across the full urban hierarchy in the Global South.\u003c/p\u003e\n\u003cp\u003eThe overarching objective of this study was to develop and apply a globally consistent spatial model of morphological deprivation (City Segment Morphological Deprivation, CSMD) using harmonized geospatial data and supervised learning (Figure 1). Specifically, we asked: (1) what are the global, regional, and national patterns of morphological deprivation when assessed consistently at the city-segment scale across 5,132 cities in 103 countries?; (2) how is morphological deprivation distributed across the urban hierarchy, from small and medium-size cities to large cities, very large cities, and megacities?; (3) how do CSMD classifications align with existing large-scale products (SSI, MN, WRI) and what do these overlaps and divergences imply for monitoring SDG 11.1.1 and guiding equitable urban interventions, particularly in smaller cities that are under-represented?\u0026nbsp;\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eWe spatially modelled urban morphological deprivation using a Random Forest model trained on built-environment indicators from City Segments v1 and labelled reference data from IDEABench. The resulting city segment morphological deprivation (CSMD) model was summarized at global, regional, national, and city-size levels to quantify how morphological deprivation is distributed across the urban system. We also compared CSMD classifications with three independent products, the Slum Severity Index (SSI), Million Neighborhoods (MN) project, and World Resources Institute (WRI) Intra-Urban Land Use product, to assess alignment with alternative representations of deprivation.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eGlobal and regional spatial patterns\u003c/h2\u003e\n\u003cp\u003eThe analysis covers 5,132 cities across 103 countries, representing a combined urban population of approximately 1.96 billion people. Of these, an estimated 349 million people (17.8% of the total) resided within morphologically deprived city segments. Figure 2 maps the locations of the analyzed city segments and reports the percentage of population living in these deprived segments at regional and country scales.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRegional shares differed markedly. Africa accounted for about 144 million people living in morphologically deprived segments, or 28.2% of the 512 million residents covered. In Asia, an estimated 147 million people lived in morphologically deprived segments, representing 13.3% of the 1.1 billion people analysed. In LAC, roughly 57 million people (17.0% of the 338 million) were estimated to live in morphologically deprived segments. Together, these regional aggregates indicate that morphological deprivation was most extensive in African cities and substantial in Asia and LAC cities.\u003c/p\u003e\n\u003cp\u003eAt the national level, the share of population residing in morphologically deprived city segments varied widely within and across regions (Figure 3). In Africa, national averages ranged from below 10% in countries such as Ghana to around 60% in Somalia, while countries such as Egypt (~19 million) and the Democratic Republic of the Congo (~18 million) contained the largest morphologically deprived populations in absolute terms. \u0026nbsp;In Asia, country shares remained below 35% everywhere, while India (~68 million) accounted for the largest absolute counts. In LAC, Haiti (~69%) exhibited very high national shares, whereas several other countries recorded values below 20%. Country-level statistics, including ISO3 codes, country names, regional classifications, and corresponding area and population aggregates, are provided in Extended Data Table 2. \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eCity-size structure of morphological deprivation\u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eWe summarized populations residing within morphologically deprived city segments by city-size class to examine how deprivation was distributed across the urban hierarchy. The 5,132 cities were grouped into five United Nations size categories \u003csup\u003e26\u003c/sup\u003e: 4,549 small cities (\u0026lt; 500,000 inhabitants), 283 medium (500,000 \u0026ndash; 1 million), 252 large (1 \u0026ndash; 5 million), 25 very large (5 \u0026ndash; 10 million), and 23 megacities (\u0026ge; 10 million). Collectively, these cities contained 1.96 billion residents. Of these, approximately 82 million in small cities (12.8% of their population), 29 million in medium (14.7%), 93 million in large (17.8%), 43 million in very large (25.0%), and 102 million in megacities (24.0%) people lived in deprived segments. Taken together, small and medium cities accounted for roughly one-third (32%; 111M of 349 M) of all residents in deprived segments, while large and megacities comprised more than half (56%; 195M of 349 M), showing that deprivation spanned the full urban hierarchy rather than being confined to the largest metropolitan areas.\u003c/p\u003e\n\u003cp\u003ePatterns differed markedly across regions (Figure 3 insets). In Africa, morphological deprivation was observed across all size classes, with the largest absolute numbers in large (42 million) and small (38 million) cities, even though relative shares peak in very large (41.5%) and megacities (38.9%). In Asia, morphological deprivation was heavily concentrated in the largest urban centres: megacities alone accounted for 66 million people (23.2%), almost double the combined total of small and medium cities (36 million), while large and very large cities added further substantial numbers. In LAC, morphologically deprived populations were relatively evenly distributed across size categories, with large cities contributing the highest absolute totals (22 million), followed by small (18 million) and megacities (12 million). Across regions, the magnitude of morphologically deprived populations in smaller cities remained substantial, motivating a closer look at how these contributions varied by country (Figure 4).\u003c/p\u003e\n\u003cp\u003eAt the country level, the composition of populations in morphologically deprived city segments varied considerably across size classes (Figure 4). In several cases, including Djibouti, Maldives, and Belize, all observed these deprived populations were located in small cities. In others, such as Nigeria, India, and Brazil, networks of small and medium cities collectively hosted millions of residents in morphologically deprived segments, in some instances matching or exceeding megacity totals. These patterns underscored that national deprivation burdens often reflected distributed systems of smaller and intermediate cities rather than a single dominant metropolis.\u003c/p\u003e\n\u003cp\u003eTo provide additional context on small-city patterns, Figure 5 presents satellite-based examples from six small cities across Africa, Asia, and LAC, showing the spatial distribution of morphologically deprived and non-deprived segments. In some cities, such as Juba (South Sudan) and Ouanaminthe (Haiti), deprived segments covered larger portions of the built-up area, whereas in others, including Sambhal (India) and Myaungmya (Myanmar), they formed more localized clusters. Across these examples, morphologically deprived segments ranged from compact pockets to more contiguous expanses, but in all cases corresponded to densely built areas with limited visible open space.\u003c/p\u003e\n\u003ch2\u003eComparative alignment with existing datasets\u003c/h2\u003e\n\u003cp\u003eWe overlaid the CSMD model with three independent products that captured different dimensions of urban deprivation. Figure 6 reported country-level precision, recall, and F1 for one representative configuration per dataset; additional boxplots are provided in supplementary material.\u003c/p\u003e\n\u003cp\u003eSSI showed the strongest alignment with CSMD. The SpaceDef indicator, which captures overcrowding and inadequate living space, achieved high recall because almost all CSMD-deprived segments also exhibited SpaceDef deprivation (recall = 0.88), while precision was moderate (0.56) because SpaceDef classified a broader set of segments as deprived. Consequently, the share of segments and population classified as deprived was higher in SSI than CSMD (~43% vs ~35% of segments; ~43% vs ~31% of population). MN showed more modest alignment, precision and recall were low (0.35 \u0026amp; 0.32), indicating that only a minority of segments labelled deprived by MN were also labelled deprived by CSMD. MN identified a slightly smaller share of segments as deprived than CSMD but in more populous areas, consistent with its focus on street-network accessibility rather than broader built-form characteristics. WRI exhibited high recall but low precision (0.66 \u0026amp; 0.28). Many segments containing informal-subdivision or atomistic land-use classes also appeared in CSMD\u0026rsquo;s deprived set (recall =0.66), but WRI assigned these classes across a much wider portion of the urban region, leading to low precision (0.28) and substantially higher deprived coverage (~49% of segments; ~46% of population).\u003c/p\u003e"},{"header":"Discussion","content":"\u003ch2\u003eGlobal picture\u003c/h2\u003e\n\u003cp\u003eOur results delivered comparable, neighborhood-scale estimates of spatially modelled morphological deprivation across 5,132 cities in 103 countries, showing how morphological deprivation was distributed across the urban hierarchy. By combining standardized city-segment units with simple morphology and access indicators\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e23\u003c/span\u003e\u003c/sup\u003e and harmonized, field-validated reference labels of morphologically deprived areas from IDEABench\u003csup\u003e24\u003c/sup\u003e, the analysis moved beyond case-specific studies that have constrained cross-city synthesis and largely focused on a handful of major cities\u003csup\u003e13,27\u0026ndash;29\u003c/sup\u003e. In doing so, it complemented household-based slum indicators and survey-driven monitoring by emphasizing the spatial manifestation of disadvantage in urban form and access, responding to calls to bridge household conditions and neighborhood morphology through spatially modelled morphological deprivation\u003csup\u003e6,30\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eSmall and medium cities emerge as major hotspots of spatially modelled morphological deprivation once absolute population was taken into account, particularly in Africa and LAC. Globally, residents in morphologically deprived segments are present across the full urban hierarchy: small and medium cities together accounted for roughly one-third of all people living in deprived segments, while large cities and megacities comprised most of the remainder. In Africa, large and small cities (42M \u0026amp; 38M) host more residents in morphologically deprived segments than megacities (24M), even though relative shares peaked in the very largest agglomerations. In LAC, residents in morphologically deprived segments were relatively evenly distributed across size classes, with large and small cities contributing comparable totals. In Asia, by contrast, morphological deprivation was dominantly observed in megacities, which led both in share and absolute numbers. At the country level, many cases nevertheless showed most morphologically deprived segments concentrated in networks of small and intermediate cities rather than in a single dominant metropolis. This was consistent with national-scale analyses from Argentina, which found that extra-small and small urban areas collectively contain more informal-settlement area and experienced faster growth than the main metropolitan region\u003csup\u003e31\u003c/sup\u003e, and with evidence from secondary cities in Sub-Saharan Africa that small urban centers carried a disproportionate burden of overlapping deprivations in services, housing, and employment\u003csup\u003e32\u003c/sup\u003e. Together with long-standing reviews showing that EO-based slum mapping has concentrated on a handful of large or iconic settlements \u003csup\u003e13,27\u003c/sup\u003e, these patterns indicated that morphological deprivation was not exclusively a megacity phenomenon but a distributed condition across national settlement systems, reinforcing the need to examine neighborhood-scale conditions in secondary and intermediate cities using standardized spatial units.\u003c/p\u003e\n\u003ch2\u003eAlignment with existing datasets\u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eComparisons with the three external products showed clear, interpretable patterns rather than random agreement. The SSI SpaceDef indicator\u003csup\u003e20\u003c/sup\u003e, which captures overcrowding and lack of living space, aligned most closely with CSMD, consistent with the idea that densely occupied housing often coincides with the built-form conditions the model learns. The Million Neighborhoods (MN) project\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e7\u003c/span\u003e\u003c/sup\u003e, which measures street-access complexity, overlapped partly with CSMD, suggesting that limited road access captures some but not all of the areas classified as morphologically deprived. The WRI intra-urban land-use product\u003csup\u003e21\u003c/sup\u003e, where we treated informal-subdivision and atomistic classes as proxies for informal built areas\u003csup\u003e33\u003c/sup\u003e, showed high recall but low precision relative to CSMD, reflecting a broader footprint of these land-use classes than the more selective set of segments that model labelled as morphologically deprived. Together, these results highlighted that agreement with CSMD depended on what each dataset was designed to measure; overcrowding, access, or land-use morphology, rather than implying a single, ordered ranking of deprivation maps. SSI, MN, WRI, and CSMD are part of the new generation of global products made possible by near-complete building, road, and EO datasets; CSMD added a city-segment-based, multi-city model spanning more than 5,000 cities, including many that rarely appeared in deprivation case studies. None of these layers should be treated as a definitive ground truth; instead, their overlaps and divergences provided a contextual triangulation of complementary perspectives on urban deprivation.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eImplications for policy makers\u003c/h2\u003e\n\u003cp\u003eThe global and cross-product patterns above have direct implications for where action on morphological deprivation is needed and who is expected to act. A substantial share of residents living in morphologically deprived segments are located in small and medium cities rather than only in the largest metropolitan areas (Figures 3,4), meaning that local governments with relatively narrow tax bases and limited technical staff are responsible for these regions. Evidence from secondary cities in Sub-Saharan Africa shows that such centers often combine rapid growth with high levels of multiple deprivation but attract less policy and investment attention than primary cities\u003csup\u003e32\u003c/sup\u003e, while global assessments of basic service provision, such as the \u003cem\u003eWhat a Waste 2.0\u0026nbsp;\u003c/em\u003ereport, highlight how core urban functions like solid waste management can already strain municipal budgets and administrative capacity in low- and middle-income settings\u003csup\u003e34\u003c/sup\u003e. In this context, acting on the morphological deprivation patterns revealed by CSMD is less a matter of simply identifying hotspots than of recognizing that many of them are located in cities that face tight fiscal, staffing, and data constraints and are only weakly represented in national and international urban agendas.\u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWithin these constraints, the main value of the CSMD model and the City Segments v1 layer for policy makers lies in providing both a first-pass picture of intra-urban inequality and a practical spatial scaffold for further work. At the global and national scales, CSMD makes visible where morphologically deprived segments are concentrated within and across cities, including many small and medium cities that currently lack any neighborhood-scale mapping, helping ministries, development partners, and civil-society organizations to identify which places and city types carry the largest burdens. At the city scale, the standardized segment layer offers a common spatial frame into which local actors can plug existing information on services, infrastructure, risk, and community priorities. In regions where SSI, MN, WRI or similar products are available, CSMD can be read alongside them to build a richer picture of deprivation; in many other cities and countries, it will be the only large-scale, neighborhood-level representation currently available. In this role, CSMD is best understood as a screening and prioritization tool, not a substitute for local assessments, slum enumeration, or participatory planning, with areas of agreement between CSMD, other datasets, and available local data indicating high-confidence concentrations of deprivation and discrepancies pointing to where closer investigation is needed. Used in this way, the CSMD model and City Segments v1 can complement household-based indicators and official statistics by helping to locate and characterize morphologically deprived areas in support of monitoring and targeting efforts under SDG 11.1.1, especially in the many smaller cities that are currently under-represented in global reporting.\u0026nbsp;\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eWe developed a city-segment deprivation (CSMD) model that combines standardized built-environment indicators with multi-city reference labels to map spatially modelled morphological deprivation across 5,132 cities in 103 countries, providing the first large-scale, neighborhood-scale view of morphological deprivation across the urban hierarchy in the Global South. The resulting patterns show that morphological deprivation is a distributed condition present in all city-size classes, with small and medium cities accounting for a substantial share of residents in morphologically deprived segments rather than megacities alone. Comparisons with existing products indicate that alignment with CSMD follows each product\u0026rsquo;s conceptual focus, overcrowding, access, or land-use morphology, highlighting that no single representation captures all facets of urban deprivation and that CSMD adds a complementary, multivariate built-environment lens. At the same time, many of the morphologically deprived segments highlighted by the model fall under the responsibility of small and intermediate cities with constrained fiscal and technical capacity, where detailed local mapping is scarce or absent. In this context, CSMD model and City Segments v1 layer are best viewed as a screening and prioritization baseline that can be combined with local assessments and sectoral data where available, helping to locate and characterize deprived areas and to support more spatially explicit monitoring and targeting.\u003c/p\u003e\n\u003ch1\u003eLimitations\u003c/h1\u003e\n\u003cp\u003eOur estimates inherit constraints from the City Segments v1 dataset and its inputs. City Segments v1 spans 107 countries but omits several substantial urban populations (e.g., China, Chile, and parts of Asia), so the underlying spatial coverage is not globally exhaustive. From within this dataset, our analysis focuses on the 103 countries across Africa, Asia, and Latin America and the Caribbean, reflecting the Global South emphasis of the study. These 103 countries contain 5,132 cities for which City Segments v1 provides polygons and indicators. Importantly, these 5,132 cities represent only those urban centers for which the City Segments workflow could generate segments from available road data; many small cities with sparse or incomplete road networks were excluded upstream. As a result, our estimates for small cities apply to this subset rather than to all small urban settlements present in the underlying urban center database\u0026nbsp;GHSL-UCDB 2019.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eInputs to the City segments layer vary in completeness and may propagate into large and/or heterogeneous segments. In particular, OpenStreetMap roads and rivers are used to define segment boundaries, but coverage is uneven across regions and city sizes, leading in some cases to very large polygons or awkward delineations and skewed derived indicators. Recent global analyses of OpenStreetMap show that, although some regions exceed 80% completeness, many cities remain below 20% coverage, especially outside Europe and North America, highlighting spatial bias risks in global applications \u003csup\u003e35\u003c/sup\u003e. Population counts from GHS-POP 2023 may also underestimate residents in informal areas; multiple gridded products systematically undercount slum populations, sometimes capturing only a fraction of residents\u003csup\u003e36,37\u003c/sup\u003e implying that figures for population residing within deprived segments likely represent lower bounds. We also note a small number of countries (Laos, Turkmenistan, Kazakhstan) in our outputs with near-zero shares of population in deprived segments; given local knowledge, these are likely false negatives driven by data gaps or segmentation artefacts (e.g., very large mixed polygons) and should be interpreted cautiously.\u003c/p\u003e\n\u003cp\u003eModel training relied on IDEABench labels from eight cities whose population sizes span approximately three to 25 million. This concentrates model learning on the upper end of the urban hierarchy and may under-represent settlement patterns typical of small and medium cities, although megacities also contain diverse neighbourhood morphologies, which partially mitigates this bias. Our label aggregation (\u0026gt; 30% segment overlap with deprived class) and the binary simplification (deprived/non-deprived, with non-built-up grouped with non-deprived) abstract complex realities, particularly in mixed segments where deprived and non-deprived areas co-exist. As an ensemble, the Random Forest can capture non-linearities but does not yield causal relations. The CSMD model itself is built solely from built-environment and access proxies (morphology, roads, population density) and does not directly observe income, tenure security, or service provision, so it should be interpreted as capturing the spatial imprint of morphological deprivation rather than all of its socio-economic dimensions. National and peripheral coverage is constrained by GHSL-UCDB 2019 urban extents used in the construction of City Segments v1; all segment footprints and population counts are defined within these 2019 urban-centre boundaries. Urban expansion beyond the 2019 extents, including more recent peri-urban growth, is not included and therefore is not reflected in our deprivation estimates. Automated segmentation can produce very large polygons in complex areas (e.g., airports), potentially blending adjacent deprived and non-deprived areas. Comparisons with SSI, MN, and WRI are contextual rather than validation exercises given different concepts, units, and geographies; reported overlap depends on aggregation choices (e.g., thresholds, metrics) and each dataset\u0026rsquo;s spatial coverage. \u0026nbsp;\u0026nbsp;\u003c/p\u003e"},{"header":"Methods","content":"\u003ch2\u003eOverview\u003c/h2\u003e\n\u003cp\u003eThis study developed a globally consistent framework for mapping spatially modelled morphological deprivation using harmonized geospatial datasets and supervised learning. Two complementary data sources supported the analysis. The City Segments v1 dataset\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e23\u003c/span\u003e\u003c/sup\u003e provided standardized neighborhood-scale spatial units and associated built-environment indicators across thousands of cities, while IDEABench\u003csup\u003e24\u003c/sup\u003e offered field-validated labels of morphological deprived and non-deprived areas derived from multi-sensor Earth observation imagery. We used IDEABench labels from eight cities to train a Random Forest classifier on the City Segments indicators, including a categorical regional variable to capture broad contextual variation, and then applied the resulting model to 5,132 cities in 103 countries across Africa, Asia, and Latin America and the Caribbean (LAC). Feature selection was performed using the Variable Selection Using Random Forests (VSURF) algorithm\u003csup\u003e38\u003c/sup\u003e. Predicted segment-level morphological deprivation was aggregated to city, national, region, and city-size classes to summarize how the population residing in morphologically deprived segments is distributed across the urban hierarchy. Predicted morphological deprivation patterns were subsequently compared, in overlapping geographies with three existing spatial datasets, the Slum Severity Index\u003csup\u003e20\u003c/sup\u003e, Million Neighborhoods project\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e7\u003c/span\u003e\u003c/sup\u003e, and WRI Intra-Urban LULC dataset\u003csup\u003e21\u003c/sup\u003e, to contextualize spatial patterns of deprivation across different conceptualizations. Detailed model parameters, evaluation metrics, and sensitivity analyses are presented in the Supplementary Material.\u003c/p\u003e\n\u003ch2\u003eCity segments dataset\u003c/h2\u003e\n\u003cp\u003eThe City Segments v1 dataset\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e23\u003c/span\u003e\u003c/sup\u003e formed the spatial foundation of this study. City segments represent standardized neighborhood-scale polygons that delineate coherent urban units bounded by navigable roads and major water bodies. Each segment contained a minimum population of approximately 400 residents, ensuring demographic significance while maintaining sufficient spatial granularity to capture intra-urban variation. The segmentation approach combined population, infrastructure, and hydrographic data to produce spatially meaningful and internally comparable units across diverse city contexts. Urban extents were first defined using the Global Human Settlement Layer Urban Centre Database (GHSL-UCDB 2019), after which OpenStreetMap (via GeoFabrik) road and water networks were used to partition these areas into contiguous blocks. Iterative adjustments based on the GHS-POP 2023 population grid ensured that each segment met the demographic threshold, while the inclusion of building footprints from the Overture Maps Foundation refined the geometry and representation of the built environment.\u003c/p\u003e\n\u003cp\u003eCity Segments v1 spans 107 countries globally, but the present analysis focused on cities in Africa, Asia, and LAC. This restriction within these three regions resulted in 103 countries and 5,132 cities, reflecting both the Global South focus of the study and the fact that all IDEABench training cities are located in Africa, Asia, or LAC. Urban centers in four countries outside these regions were not considered further. A complete list of countries included is provided in Extended Data Table 2.\u003c/p\u003e\n\u003cp\u003eFor each city segment, we considered a set of 21 built-environment indicators available in the City Segments dataset, comprising 11 absolute variables and 10 ratio indices that characterize the demographic, infrastructural, and morphological structure of the urban environment. These indicators are grouped into four domains: Population and Area (e.g., total population, segment area, population density), Roads and Connectivity (e.g., total road length, population-to-road ratios), Parcels and Access (e.g., number of parcels, proportion without road access), and Buildings and Morphology (e.g., total built area, average building size). The derived ratios (i\u003csub\u003e1\u003c/sub\u003e \u0026ndash; i\u003csub\u003e10\u003c/sub\u003e) normalize these measures by population, area, or parcel count, capturing relative density, accessibility, and variability in urban form. Together, these indicators served as predictor variables for modelling CSMD across the Global South. Full definitions, computation formulas, and units of all 21 indicators are provided in Extended Data Table 1.\u003c/p\u003e\n\u003ch2\u003eIDEABench benchmark dataset\u003c/h2\u003e\n\u003cp\u003eIDEABench\u003csup\u003e\u003cspan lang=\"EN-IN\"\u003e24\u003c/span\u003e\u003c/sup\u003e provided the labelled reference data required to train and validate the supervised model of spatially modelled morphological deprivation (CSMD). Developed under the European Space Agency-funded IDEAtlas project, the dataset was co-designed with local partners and community stakeholders to support open, transparent mapping of morphologically deprived urban areas. It integrates multi-sensor Earth observation imagery with expert- and community-validated annotations to serve as a benchmark for developing and testing AI-based methods for deprivation mapping.\u003c/p\u003e\n\u003cp\u003eThe dataset spanned eight cities in the Global South across Africa, Asia, and LAC: Nairobi, Lagos, Mexico City, Buenos Aires, Salvador, Medellin, Jakarta, and Mumbai, capturing diverse morphological and socio-economic conditions. For each city, IDEABench provided 128 x 128 pixel image patches derived from Sentinel-1 and Sentinel-2 satellites, supplemented with a built-up density layer generated from the Google Open Buildings dataset. Each patch was accompanied by a three-class reference label distinguishing deprived urban areas, non-deprived urban areas, and non-built-up areas.\u003c/p\u003e\n\u003cp\u003eFor this study, IDEABench reference labels were aggregated to the city-segment scale to match the spatial framework of the City Segments dataset. Patch-based labels were first converted into city-wide rasters for each of the eight training cities and overlaid with the corresponding city-segment polygons. A segment was assigned a label of deprived (1) if more than 30% of its area was classified as a deprived urban area in IDEABench; otherwise, it was labelled non-deprived (0). Non-built-up areas were grouped with non-deprived segments, as the objective was to distinguish morphologically deprived versus all other urban areas within the urban extent. This integration produced a harmonized segment-level training set linking the 21 built-environment indicators from City Segments to binary deprivation labels across eight cities and three world regions.\u003c/p\u003e\n\u003ch2\u003eModelling Framework\u003c/h2\u003e\n\u003cp\u003eA supervised learning framework was employed to predict spatially modelled morphological deprivation at the city-segment scale. The model integrated built-environment indicators from the City Segments dataset with deprivation labels derived from IDEABench, enabling a globally consistent yet regionally adaptive analysis. We trained a Random Forest (RF) classifier on labelled segments from the eight IDEABench cities (refer to Supplementary Table 1) and then applied the resulting model to all city segments in the 5,132 cities and 103 countries in Africa, Asia, and LAC described above. This RF classifier is referred to as the city-segment deprivation (CSMD) model throughout the paper. A categorical regional variable (Africa, Asia, LAC) was included as an additional predictor to capture broad contextual variation across world regions (refer to Supplementary Table 1).\u003c/p\u003e\n\u003cp\u003eFeature selection was performed using the Variable Selection Using Random Forests (VSURF) algorithm\u003csup\u003e38\u003c/sup\u003e. VSURF was run on the full set of 21 built-environment indicators plus regional variable (total 22) from City Segments and identified a parsimonious subset of ten indicators that contributed most strongly to predicting morphological deprivation (refer to Supplementary section 1.2). The final set of ten predictors used in the CSMD model are i5_par_area (average parcel area), i1_pop_area (population density), B_AVG_SEG (average building area), i9_roads_par (road length per parcel), i6_paru_area (average area of untouched parcels), i8_paru_par (proportion of untouched parcels), PARU_A_SEG (median area of untouched parcels), B_AREA_SEG (total building footprint area), B_CV_SEG (coefficient of variation of building area) and REG1_GHSL (the regional code variable).\u003c/p\u003e\n\u003cp\u003eThe Random Forest was implemented in Python using the scikit-learn library. We fitted the model using a fixed 80/20 train-test split at the segment level within the eight IDEABench cities, with balanced class weighting to mitigate the minority status of morphologically deprived segments (refer to Supplementary section 1.3). Model performance on the held-out test data was evaluated using standard classification metrics, including precision, recall, F1-score, and balanced accuracy (refer to Supplementary section 1.4). The trained CSMD model was then applied to all city segments in the 5,132 cities to generate binary deprivation classifications. Full details of the RF steup, hyper-parameter tuning, ROC curves, threshold sweeps, confusion matrices, and feature-importance plot are provided in Supplementary section 1.\u003c/p\u003e\n\u003ch2\u003eCity-size analysis\u003c/h2\u003e\n\u003cp\u003eTo examine how spatially modelled morphological deprivation varied with urban scale, we aggregate segment-level predictions from the CSMD model to the level of individual cities and classified each city according to its population size. Following the World Urbanization Prospects 2018 classification, cities were grouped into five size categories: small (\u0026lt; 500,000 inhabitants), medium (500,000 \u0026ndash; 1 million), large (1 \u0026ndash; 5 million), very large (5 \u0026ndash; 10 million), and megacity (\u0026ge; 10 million)\u003csup\u003e26\u003c/sup\u003e. Population per segment was taken from the City Segments dataset, where it is one of the 21 built-environment indicators derived from the GHS POP 2023 dataset. For every city, total population was computed as the sum of segment-level populations, and deprived population was calculated as the sum of population within segments classified as morphologically deprived by the CSMD model. The proportion of morphologically deprived population was then derived as the ratio of deprived to total population. For each country, regional, and global summary, we aggregated total and deprived populations across cities within each size class. These statistics provided the basis for the regional and national patterns reported in the Results section, including comparisons of the magnitude and share of morphologically deprived populations across the urban hierarchy (Figures 2, 3 \u0026amp; 4).\u003c/p\u003e\n\u003ch2\u003eComparative analyses\u003c/h2\u003e\n\u003cp\u003eTo contextualize the spatial patterns produced by the CSMD model, we compared its segment-level classifications with three existing datasets that capture different dimensions of urban deprivation: the Slum Severity Index (SSI)\u003csup\u003e20\u003c/sup\u003e, the Million Neighbourhoods (MN) project\u003csup\u003e7\u003c/sup\u003e, and the World Resource Institute (WRI) Intra-Urban Land-Use/Land-Cover dataset\u003csup\u003e21\u003c/sup\u003e. These comparisons were restricted to cities and countries where both CSMD and the respective external dataset were available; sub-Saharan Africa for SSI and MN, and a set of over 100 cities worldwide for WRI and were interpreted as contextual alignment across concepts rather than formal validation. For each dataset, we derived a binary \u0026lsquo;deprived/non-deprived\u0026rsquo; label at the city-segment level, intersected these labels with the CSMD classifications, and computed country-level precision, recall, and F1 scores, as well as the proportions of segments and population classified as deprived by each source. These metrics formed the basis of the comparative alignment summaries reported in the results (Figure 6).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe SSI dataset was produced by operationalizing the four UN-Habitat slum indicators; water, sanitation, housing durability, and living space adequacy (SpaceDef) and provided each component and a combined SSI index (0-4) at 100 m resolution across sub-Saharan Africa. Country-level SSI raster\u0026rsquo;s (five bands: water, sanitation, housing, SpaceDef, combined SSI) were clipped to the extent of the City Segments polygons, and segments were retained only if they contained at least one valid SSI pixel. For each segment, we calculated the proportion of valid pixels with SpaceDef \u0026gt; 0 and used this component as the SSI-based indicator of deprivation. Segments were classified as deprived by SSI when this proportion exceeded a threshold \u0026tau;; we evaluated \u0026tau; = 0.1, 0.2, and 0.3 and used \u0026tau; = 0.1 as the representative configuration in our comparative analyses. Detailed results for alternative SSI components and thresholds are provided in the Supplementary material (Section 2.3 \u0026amp; Figure 6).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;The MN dataset reports a building-to-street depth index \u003cem\u003ek\u0026nbsp;\u003c/em\u003eat the block level across sub-Saharan Africa, where higher \u003cem\u003ek\u0026nbsp;\u003c/em\u003evalues indicate poorer street access and greater infrastructural inaccessibility. MN blocks were harmonized to WGS84 and overlaid with the City Segments polygons after geometry checks. Blocks with \u003cem\u003ek\u0026nbsp;\u003c/em\u003e\u0026gt; 0.3 were treated as access-deficit, and for each city segment we calculated the proportion of intersecting blocks above this threshold using a centroid-majority rule with area-based fallbacks where needed (refer to Supplementary section 2.4.2). Segments were then classified as deprived by MN when this proportion exceeded a threshold \u0026tau;, taken as \u0026tau; = 0.1 for the comparative analyses. Sensitivity of the alignment metrics to alternative \u003cem\u003ek\u0026nbsp;\u003c/em\u003ecut-off and segment-level thresholds is reported in the Supplementary Material (Section 2.4 \u0026amp; Figure 7).\u003c/p\u003e\n\u003cp\u003eThe WRI Intra-Urban LULC dataset has coverage beyond sub-Saharan Africa (over 100 countries in 48 countries that overlap with City Segments) and followed Friesen et al. (2025) in treating the \u0026lsquo;informal subdvision\u0026rsquo; and \u0026lsquo;atomistic\u0026rsquo; classes as proxies for informal built areas. WRI rasters (5 m) were mosaicked and clipped to the city-segment boundaries for all cities where both datasets were available. For each segment, we computed the proportion of valid pixels belonging to these informal subdivision or atomistic classes and used this proportion as the WRI-based indicator of deprivation. Segments were classified as deprived by WRI when this proportion exceed \u0026tau; = 0.1, Detailed sensitivity analyses exploring alternative thresholds are provided in the Supplementary Material (Section 2.5 \u0026amp; Figure 8).\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eAcknowledgements\u003c/p\u003e\n\u003cp\u003eThis research was supported by funding from FORMAS (Swedish Research Council for Sustainable Development) under grant no. 2023-01210 (DEPRIMAP). The computation (model training) was partly enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by the Swedish Research Council through grant agreement no. 2022-06725. The IDEABench benchmark dataset is an output enabled by the foundational support of the IDEAtlas project (https://ideatlas.eu/). We formally acknowledge the essential contributions of the local co-anchors and collaborating institutions, who were responsible for the detailed data curation, collection, and validation necessary for the IDEABench work.\u003c/p\u003e\n\u003cp\u003eAuthor contributions\u003c/p\u003e\n\u003cp\u003eSGV, SG, DRT conceptualized the study. SGV led the data analysis and drafted the paper. SGV, SG, DRT, JF contributed to the methodological design and paper revisions. AMB, DRT contributed to City Segments v1 dataset. BT, MK, CP, RVM, AA contributed to IDEABench dataset. SGV, JF prepared the figures. DRT, JF, SG supported data interpretation. SG and DRT supervised the overall project. All authors contributed to discussions, provided critical feedback and approved the final version of the paper.\u003c/p\u003e\n\u003cp\u003eData availability\u003c/p\u003e\n\u003cp\u003eAll data and code used for preprocessing, modelling, and figure generation are publicly available in the project\u0026rsquo;s GitHub repository (https://github.com/saiga143/citysegmentdeprivation). A Zenodo archive linked to this repository (https://doi.org/10.5281/zenodo.17637298) provides all large files, including the trained Random Forest model and full prediction outputs that cannot be hosted on GitHub due to size limits. The City Segments v1 dataset is publicly available at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/XLRSF0. The IDEABench Reference dataset are available at https://phys-techsciences.datastations.nl/dataset.xhtml?persistentId=doi:10.17026/PT/X4NJII. External products used for comparative analysis were obtained from their respective sources: SSI from https://zenodo.org/records/14998570, MN from https://www.millionneighborhoods.africa/download and WRI Urban Land Use dataset via Google Earth Engine at https://code.earthengine.google.com/?asset=projects/wri-datalab/urban_land_use/V1.\u003c/p\u003e\n\u003cp\u003eCode availability\u003c/p\u003e\n\u003cp\u003eAll code is publicly available via GitHub (https://github.com/saiga143/citysegmentdeprivation) and Zenodo (https://doi.org/10.5281/zenodo.17637264).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eUN Habitat. \u003cem\u003eUN Habitat\u0026rsquo;s 2024 Annual Report, Adequate Housing for All\u003c/em\u003e. \u003cem\u003eUN-Habitat\u003c/em\u003e https://unhabitat.org/annual-report-2024 (2025).\u003c/li\u003e\n\u003cli\u003eUN-Habitat. \u003cem\u003eImplementation of the Strategic Plan for the Period 2020\u0026ndash;2025 \u003c/em\u003e. \u003cem\u003eUnited Nations\u003c/em\u003e https://unhabitat.org/sites/default/files/2025/04/2503673e.pdf (2025).\u003c/li\u003e\n\u003cli\u003eKundu, D. \u0026amp; Pandey, A. K. World Urbanisation: Trends and Patterns. in \u003cem\u003eDeveloping National Urban Policies\u003c/em\u003e 13\u0026ndash;49 (Springer Nature Singapore, Singapore, 2020). doi:10.1007/978-981-15-3738-7_2.\u003c/li\u003e\n\u003cli\u003eGrossmann, K. \u0026amp; Mallach, A. The small city in the urban system: complex pathways of growth and decline. \u003cem\u003eGeogr Ann Ser B\u003c/em\u003e \u003cstrong\u003e103\u003c/strong\u003e, 169\u0026ndash;175 (2021).\u003c/li\u003e\n\u003cli\u003eWagner, M. \u0026amp; Growe, A. Research on Small and Medium-Sized Towns: Framing a New Field of Inquiry. \u003cem\u003eWorld\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 105\u0026ndash;126 (2021).\u003c/li\u003e\n\u003cli\u003eAbascal, A. \u003cem\u003eet al.\u003c/em\u003e \u0026ldquo;Domains of deprivation framework\u0026rdquo; for mapping slums, informal settlements, and other deprived areas in LMICs to improve urban planning and policy: A scoping review. \u003cem\u003eComput Environ Urban Syst\u003c/em\u003e \u003cstrong\u003e93\u003c/strong\u003e, 101770 (2022).\u003c/li\u003e\n\u003cli\u003eBettencourt, L. M. A. \u0026amp; Marchio, N. Infrastructure deficits and informal settlements in sub-Saharan Africa. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e645\u003c/strong\u003e, 399\u0026ndash;406 (2025).\u003c/li\u003e\n\u003cli\u003eGeorganos, S., Vanhuysse, S., Abascal, A. \u0026amp; Kuffer, M. Extracting Urban Deprivation Indicators Using Superspectral Very-High-Resolution Satellite Imagery. in \u003cem\u003e2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS\u003c/em\u003e 2114\u0026ndash;2117 (IEEE, 2021). doi:10.1109/IGARSS47720.2021.9554849.\u003c/li\u003e\n\u003cli\u003eKuffer, M. \u003cem\u003eet al.\u003c/em\u003e Mapping the Morphology of Urban Deprivation. in \u003cem\u003eUrban Remote Sensing\u003c/em\u003e 305\u0026ndash;323 (Wiley, 2021). doi:10.1002/9781119625865.ch14.\u003c/li\u003e\n\u003cli\u003eWang, J. \u003cem\u003eet al.\u003c/em\u003e EO + Morphometrics: Understanding cities through urban morphology at large scale. \u003cem\u003eLandsc Urban Plan\u003c/em\u003e \u003cstrong\u003e233\u003c/strong\u003e, 104691 (2023).\u003c/li\u003e\n\u003cli\u003eOwusu, M., Engstrom, R., Thomson, D., Kuffer, M. \u0026amp; Mann, M. L. Mapping Deprived Urban Areas Using Open Geospatial Data and Machine Learning in Africa. \u003cem\u003eUrban Science\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 116 (2023).\u003c/li\u003e\n\u003cli\u003eUN Habitat. \u003cem\u003eThe Urban SDG Monitoring Series | Monitoring SDG Indicator 11.1.1\u003c/em\u003e. \u003cem\u003eUN-Habitat\u003c/em\u003e www.unhabitat.org. (2019).\u003c/li\u003e\n\u003cli\u003eKuffer, M., Pfeffer, K. \u0026amp; Sliuzas, R. Slums from Space\u0026mdash;15 Years of Slum Mapping Using Remote Sensing. \u003cem\u003eRemote Sens (Basel)\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 455 (2016).\u003c/li\u003e\n\u003cli\u003edo Nascimento, G. A., Giannotti, M., Regueira, T. A. \u0026amp; Tomasiello, D. B. Identifying slum areas: A multidimensional analysis leveraging with explanatory machine learning techniques. \u003cem\u003eSustain Cities Soc\u003c/em\u003e \u003cstrong\u003e131\u003c/strong\u003e, 106645 (2025).\u003c/li\u003e\n\u003cli\u003eWarren, J. L., Perez-Heydrich, C., Burgert, C. R. \u0026amp; Emch, M. E. Influence of Demographic and Health Survey Point Displacements on Point-in-Polygon Analyses. \u003cem\u003eSpat Demogr\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 117\u0026ndash;133 (2016).\u003c/li\u003e\n\u003cli\u003ePerez-Heydrich, C., Warren, J. L., Burgert, C. R. \u0026amp; Emch, M. E. Influence of Demographic and Health Survey Point Displacements on Raster-Based Analyses. \u003cem\u003eSpat Demogr\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 135\u0026ndash;153 (2016).\u003c/li\u003e\n\u003cli\u003ePettersson, M. B. \u0026amp; Daoud, A. \u003cem\u003eLeveraging Compact Satellite Embeddings and Graph Neural Networks for Large-Scale Poverty Mapping\u003c/em\u003e. \u003cem\u003earXiv\u003c/em\u003e http://arxiv.org/abs/2511.01408 (2025).\u003c/li\u003e\n\u003cli\u003eGeorganos, S., Hafner, S., Kuffer, M., Linard, C. \u0026amp; Ban, Y. A census from heaven: Unraveling the potential of deep learning and Earth Observation for intra-urban population mapping in data scarce environments. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e \u003cstrong\u003e114\u003c/strong\u003e, 103013 (2022).\u003c/li\u003e\n\u003cli\u003eTatem, A. \u0026amp; Espey, J. Global population data is in crisis \u0026ndash; here\u0026rsquo;s why that matters. \u003cem\u003eThe Conversation\u003c/em\u003e Preprint at https://doi.org/10.64628/AB.c64mdcqqs (2025).\u003c/li\u003e\n\u003cli\u003eLi, C. \u003cem\u003eet al.\u003c/em\u003e Mapping urban slums and their inequality in sub-Saharan Africa. \u003cem\u003eNature Cities\u003c/em\u003e 1\u0026ndash;12 (2025) doi:10.1038/s44284-025-00276-0.\u003c/li\u003e\n\u003cli\u003eGuzder-Williams, B., Mackres, E., Angel, S., Blei, A. M. \u0026amp; Lamson-Hall, P. Intra-urban land use maps for a global sample of cities from Sentinel-2 satellite imagery and computer vision. \u003cem\u003eComput Environ Urban Syst\u003c/em\u003e \u003cstrong\u003e100\u003c/strong\u003e, 101917 (2023).\u003c/li\u003e\n\u003cli\u003eMari Rivero, I. \u003cem\u003eet al.\u003c/em\u003e GHS-UCDB R2024A - GHS Urban Centre Database 2025. \u003cem\u003eEuropean Commission, Joint Research Centre (JRC)\u003c/em\u003e Preprint at https://doi.org/10.2905/1a338be6-7eaf-480c-9664-3a8ade88cbcd (2024).\u003c/li\u003e\n\u003cli\u003eThomson, D. R. \u0026amp; Blei, A. \u003cem\u003eCity Segments Layer V1\u003c/em\u003e. \u003cem\u003eHarvard Dataverse\u003c/em\u003e https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/XLRSF0 (2025) doi:10.7910/DVN/XLRSF0.\u003c/li\u003e\n\u003cli\u003eTereke, B. W. \u003cem\u003eet al.\u003c/em\u003e \u003cem\u003eIDEABench: Benchmark Dataset for Mapping Deprived Urban Areas\u003c/em\u003e. \u003cem\u003eDANS Data Station Physical and Technical Sciences\u003c/em\u003e https://phys-techsciences.datastations.nl/dataset.xhtml?persistentId=doi:10.17026/PT/X4NJII (2025) doi:10.17026/PT/X4NJII.\u003c/li\u003e\n\u003cli\u003eKochupillai, M., Kahl, M., Schmitt, M., Taubenbock, H. \u0026amp; Zhu, X. X. Earth Observation and Artificial Intelligence: Understanding emerging ethical issues and opportunities. \u003cem\u003eIEEE Geosci Remote Sens Mag\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 90\u0026ndash;124 (2022).\u003c/li\u003e\n\u003cli\u003eUnited Nations. \u003cem\u003eWorld Urbanization Prospects 2018 Highlights\u003c/em\u003e. \u003cem\u003eDepartment of Economic and Social Affairs, Population Division\u003c/em\u003e (2019).\u003c/li\u003e\n\u003cli\u003eMahabir, R., Croitoru, A., Crooks, A., Agouris, P. \u0026amp; Stefanidis, A. A Critical Review of High and Very High-Resolution Remote Sensing Approaches for Detecting and Mapping Slums: Trends, Challenges and Emerging Opportunities. \u003cem\u003eUrban Science\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 8 (2018).\u003c/li\u003e\n\u003cli\u003eVeeravalli, S. G. \u003cem\u003eet al.\u003c/em\u003e Towards a Spatial Measure of SDG 11.1.1: Open Data for Urban Deprivation Mapping. in \u003cem\u003e2025 Joint Urban Remote Sensing Event (JURSE)\u003c/em\u003e 1\u0026ndash;4 (IEEE, 2025). doi:10.1109/JURSE60372.2025.11076033.\u003c/li\u003e\n\u003cli\u003eVeeravalli, S. G., Haas, J., Friesen, J. \u0026amp; Georganos, S. Understanding Informal Settlement Transformation through Google\u0026rsquo;s 2.5D Dataset and Street View based Validation. \u003cem\u003eThe International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences\u003c/em\u003e \u003cstrong\u003eXLVIII-M-7\u0026ndash;2025\u003c/strong\u003e, 245\u0026ndash;251 (2025).\u003c/li\u003e\n\u003cli\u003eThomson, D. R. \u003cem\u003eet al.\u003c/em\u003e Need for an Integrated Deprived Area \u0026ldquo;Slum\u0026rdquo; Mapping System (IDEAMAPS) in Low- and Middle-Income Countries (LMICs). \u003cem\u003eSoc Sci\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 80 (2020).\u003c/li\u003e\n\u003cli\u003eSamper, J. \u003cem\u003eet al.\u003c/em\u003e \u003cem\u003eSpatiotemporal Dynamics of Informal Settlements Across Argentine Cities: A National-Scale Analysis\u003c/em\u003e. https://www.ssrn.com/abstract=5588589 (2025) doi:10.2139/ssrn.5588589.\u003c/li\u003e\n\u003cli\u003eGithira, D. \u003cem\u003eet al.\u003c/em\u003e \u003cem\u003eAnalysis of Multiple Deprivations in Secondary Cities in Sub-Saharan Africa\u003c/em\u003e. \u003cem\u003eUN-Habitat\u003c/em\u003e https://unhabitat.org/sites/default/files/2021/04/analysis_of_multiple_deprivations_in_secondary_cities_-_analysis_report.pdf (2020).\u003c/li\u003e\n\u003cli\u003eFriesen, J., Georganos, S. \u0026amp; Haas, J. Differences in walking access to healthcare facilities between formal and informal areas in 19 sub-Saharan African cities. \u003cem\u003eCommunications Medicine\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, 41 (2025).\u003c/li\u003e\n\u003cli\u003eKaza, S., Yao, L., Bhada-Tata, P. \u0026amp; Woerden, F. Van. \u003cem\u003eWhat a Waste 2.0: A Global Snapshot of Solid Waste Management to 2050. Urban Development Series\u003c/em\u003e. \u003cem\u003eWorld Bank\u003c/em\u003e https://openknowledge.worldbank.org/handle/10986/2174. (2018).\u003c/li\u003e\n\u003cli\u003eHerfort, B., Lautenbach, S., Porto de Albuquerque, J., Anderson, J. \u0026amp; Zipf, A. A spatio-temporal analysis investigating completeness and inequalities of global urban building data in OpenStreetMap. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 3985 (2023).\u003c/li\u003e\n\u003cli\u003eBreuer, J. H. P., Friesen, J., Taubenb\u0026ouml;ck, H., Wurm, M. \u0026amp; Pelz, P. F. The unseen population: Do we underestimate slum dwellers in cities of the Global South? \u003cem\u003eHabitat Int\u003c/em\u003e \u003cstrong\u003e148\u003c/strong\u003e, 103056 (2024).\u003c/li\u003e\n\u003cli\u003eThomson, D. R. \u003cem\u003eet al.\u003c/em\u003e Evaluating the Accuracy of Gridded Population Estimates in Slums: A Case Study in Nigeria and Kenya. \u003cem\u003eUrban Science\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, 48 (2021).\u003c/li\u003e\n\u003cli\u003eGenuer, R., Poggi, J.-M., Tuleau-Malot, C. \u0026amp; Tuleau, C. VSURF: An R Package for Variable Selection Using Random Forests. \u003cem\u003eR J\u003c/em\u003e\u003cstrong\u003e7\u003c/strong\u003e, 19\u0026ndash;33 (2015).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8189204/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8189204/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"In growing cities, deprived neighborhoods house large numbers of residents, yet their extent and distribution remain poorly quantified, complicating implementation of SDG 11.1.1. We present the first global, neighborhood-scale spatial estimates of morphological deprivation, covering 5,132 cities in 103 countries across Africa, Asia, and Latin America \u0026 the Caribbean (LAC) home to 3.2 billion people. Neighborhood units and built-environment indicators from the City Segments v1 dataset were combined with segment-level labels from the eight-city IDEABench benchmark to train a supervised model, which was then applied to classify each segment as morphologically deprived or non-deprived. The mapped cities contained 1.96 billion residents, of whom 349 million (17.8%) lived in deprived segments, with the highest regional shares in Africa and substantial burdens in Asia and LAC. Morphologically deprived populations spanned the urban hierarchy, with about one-third living in small and medium cities, revealing important gaps in current deprivation monitoring.","manuscriptTitle":"The Hidden Burden of Morphological Deprivation in Small and Medium Cities","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-11 06:29:44","doi":"10.21203/rs.3.rs-8189204/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-cities","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"natcities","sideBox":"Learn more about [Nature Cities](https://www.springer.com/journal/44284)","snPcode":"44284","submissionUrl":"","title":"Nature Cities","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"983909ba-aab9-465b-b95f-d4a4b502140c","owner":[],"postedDate":"December 11th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":59422380,"name":"Earth and environmental sciences/Environmental social sciences/Sustainability"},{"id":59422381,"name":"Social science/Geography"},{"id":59422382,"name":"Social science/Environmental studies"},{"id":59422383,"name":"Scientific community and society/Geography"},{"id":59422384,"name":"Scientific community and society/Developing world"}],"tags":[],"updatedAt":"2026-04-25T08:21:59+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-11 06:29:44","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8189204","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8189204","identity":"rs-8189204","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}