The research, published in the journal Discover Cities, was led by Zhangliang Deng of Heze University in China, together with Christopher Thomas Lloyd and Jim Wright of the University of Southampton, Lorna-Grace Okotto of Jaramogi Oginga Odinga University of Science and Technology, and Joseph Okotto-Okotto of VIRED International in Kenya. The team conducted one of the first empirical comparisons of road and water pipeline network structure across cities in the Global South and the Global North, using real pipeline data from four cities on three continents: Kigali in Rwanda and Kisumu in Kenya representing sub-Saharan Africa, and La Rochelle in France and Vancouver in Canada representing the Global North.

To make the comparison rigorous, the researchers converted both road and pipeline networks into dual graphs—a mathematical representation in which each street or pipe segment becomes a node, and connections are drawn between segments that share an endpoint. This representation allows centrality measures, the workhorses of network science, to be computed directly for the infrastructure segments themselves rather than for intersections. The team calculated three such measures: degree centrality, which counts how many connections a segment has; betweenness centrality, which quantifies how often a segment lies on shortest paths through the network and thus how critical it is to overall flow; and closeness centrality, which measures how efficiently a segment can reach all other parts of the network. Each metric captures a different dimension of structural importance—connectivity, flow loading, and accessibility, respectively.

The first major finding emerged from the D-measure, a graph-similarity metric developed by Schieber and colleagues and published in Nature Communications in 2017. The D-measure quantifies how dissimilar two networks are by comparing their node-to-node distance distributions, their connectivity heterogeneity, and their centrality profiles, returning zero for topologically equivalent graphs. Because it is largely insensitive to network size, it enabled fair comparison between networks of very different scales—from La Rochelle’s compact 28-square-kilometre footprint to Kigali’s sprawling 730 square kilometres. The results showed high structural similarity across all eight networks, with most similarity values between 0.8 and 0.9. But a clear regional split appeared: road networks from the two African cities were most similar to each other, and the two Northern road networks clustered together, while water pipeline networks remained remarkably similar across all four cities regardless of region. Road and water networks were also more similar to each other in the Northern cities than in the African ones, with Kigali’s road network showing the lowest similarity to water networks overall.

That regional divergence in road networks carried through to the centrality statistics. Mean degree centrality of road networks hovered near four, generally exceeding that of pipeline networks, and Vancouver’s networks showed the highest connectivity of the four cities. More tellingly, the pattern of betweenness flipped between regions: in Kigali and Kisumu, water pipelines carried higher betweenness than roads, whereas in La Rochelle and Vancouver the opposite held. The authors interpret this as evidence that the African case cities lack the primary high-capacity road corridors that dominate Northern networks—a finding consistent with prior work showing that informal settlements, which house roughly 79 percent of Kigali’s population and about 60 percent of Kisumu’s, tend to be topologically isolated, with poor connectivity to the surrounding urban fabric. Kigali’s road network, with an exceptionally narrow range of closeness values, may even approximate a tree-like structure resembling a minimum spanning tree, a signature of largely self-organised growth and, possibly, of the city’s hilly terrain.

The second analytical stage moved from pure topology into geometric space. The researchers overlaid pipeline and road layers and, to cope with positional uncertainty in the data, performed a sensitivity analysis: they drew buffers of increasing width around every street segment, in one-metre increments, and tracked how much pipeline length fell inside each buffer. Using segmented regression, they identified a breakpoint separating rapid from slower pipeline capture, which they took as the effective road width—11 metres for both Kigali and Kisumu, 7 metres for La Rochelle, and 10 metres for Vancouver. At these optimal widths, the proportion of pipeline lying co-located with roads ranged dramatically: 92.6 percent in Vancouver, 76.8 percent in Kigali, 74.2 percent in La Rochelle, and just 67.7 percent in Kisumu. The Northern cities, in other words, bury their pipes far more faithfully beneath their streets.

To quantify the coupling between the two systems, the team then extracted paired centrality values for co-located road–pipeline segments and fitted them with total least squares (TLS) regression. Unlike ordinary least squares, which assumes the explanatory variable is error-free, TLS minimises perpendicular distances to the fitted line, making it appropriate when no causal direction exists between the two variables—as is the case for roads and pipes, which influence each other rather than one determining the other. Only closeness centrality showed a sufficiently strong Pearson correlation to warrant this analysis. Across all four cities the slopes were positive: streets with high closeness tend to sit above pipelines that also have high closeness. But the slopes were steeper in the African cities, and the residual analysis revealed a systematic pattern in all four: from the urban core outward, pipeline closeness falls faster than road closeness. A Global Moran’s I test, validated with 999 Monte Carlo permutations, confirmed that these deviations were strongly and significantly spatially clustered in every city, with autocorrelation values between 0.928 and 0.957.

Mapping the outliers onto real neighbourhoods gave the pattern a human geography. Negative deviations—pipelines weaker than the roads above them—concentrated in peripheral areas such as Nyagahinga in Kigali, a low-density residential zone slated for future densification where roads exist but pipe infrastructure lags behind. Positive deviations clustered around major water facilities: Rugarama and Masaka in Kigali sit near the Nzove and Karenge water supply systems, Nyamasaria in Kisumu lies close to the private Nyamasaria Water Works, and Kibuye, Kisumu’s principal market district, hosts the city’s most important water tanks. At the upper tail, exceptionally high pipeline closeness appeared in historically integrated neighbourhoods such as La Genette in La Rochelle, a historic district outside the old city walls, and Shaughnessy in Vancouver, an affluent residential enclave. Pipeline structure, in short, is sculpted by the distribution of water facilities and urban communities, producing a heterogeneity that roads do not share.

The implications reach well beyond network theory. Contemporary planning practice often treats streets as the organising skeleton for all underground utilities, and street-led slum upgrading programmes assume that improving road connectivity will pull water and sanitation infrastructure along with it. These findings suggest that assumption holds better in the Global North than in sub-Saharan African cities, where pipelines obey organisational logics that diverge substantially from street patterns. Planners, the authors argue, should consider a settlement’s position within the wider urban network—its connectivity and its proximity to major water infrastructure—rather than relying on road layout alone as a proxy. The results also matter for asset management: because roughly two-thirds to nine-tenths of pipeline length shadows roads, excavation and renewal projects frequently interact with pipe segments of varying structural importance, and knowing where high-closeness road–pipeline pairs occur could help prioritise maintenance and reduce the risk of accidental damage where utility records are incomplete or outdated.

Why pipeline networks across all four cities converged so strongly—despite radically different histories, from La Rochelle’s medieval port morphology to Kisumu’s railway-borne origins—remains an open question, and one the authors suggest may reflect shared international norms in hydraulic engineering design. The study is explicitly exploratory, built on four cities and limited by data quality variations in OpenStreetMap and municipal pipeline records, and by the well-known caveat that topology alone captures only part of infrastructure performance. But as the first quantified, cross-regional comparison of its kind, it lays down a template: a workflow combining dual-graph topology, buffer-based co-location analysis, TLS regression, and spatial autocorrelation testing that can be scaled to many more cities as pipeline data becomes available. In a century when most urban growth will happen in the Global South, understanding that water does not simply follow the street may prove essential to building cities that work.

The research, published in the peer-reviewed journal Scientific Data, fills a significant gap left by traditional data sources which measure AC ownership and types only on broad scales or limited samples. Prior datasets like the American Housing Survey provide information at county or metropolitan levels and often rely on surveyed locations, while the Energy Information Administration’s data only sample subsets of the full population. Professor Ahn’s novel approach integrates spatial modeling, advanced statistical methods, and big data analytics to generate a nuanced picture of cooling infrastructure down to the census tract level.

Central to this research was the use of Dewey’s comprehensive real estate dataset, which offers household-level insights spanning the entire country. By combining this with known predictors such as housing type, building age, renovation dates, racial and ethnic demographics, historical housing policies, and prevailing climate conditions, the study constructed a highly granular framework to infer air conditioning presence and classification. To resolve data gaps, the research team employed sophisticated machine learning algorithms including random forest for imputing missing values and the XGBoost classifier for categorizing homes by AC type: central systems, window or portable units, evaporative coolers, or none.

The results revealed previously obscured spatial and sociodemographic patterns in cooling access. Urban and rural differences emerged sharply—rural areas of Oregon, for example, showed a prevalence of central air conditioning alongside evaporative coolers, while urban centers displayed greater diversity in AC types including a significant proportion without any form of air conditioning. Florida’s households demonstrated stark contrasts; around 20% relied on non-central air conditioners whereas in urban areas central AC dominated with over 95% ownership. Such detailed insights are invaluable for pinpointing vulnerable populations during heat events.

One of the most compelling aspects of this dataset is its revelation of how socioeconomic and demographic factors influence air conditioning ownership. Climate and heating types unsurprisingly act as the strongest predictors, but ethnicity also plays a substantive role. Households in regions with higher Hispanic populations, such as parts of California and New Mexico, are notably associated with higher use of evaporative coolers and other non-central AC types. These distinctions underscore the intersection of climate, economics, social equity, and infrastructure, raising important questions about energy justice and health disparities in a warming world.

Despite the dataset’s comprehensiveness, Professor Ahn acknowledges its limitations and the challenges inherent in compiling such nationwide data. High proportions of missing information in metropolitan areas like New York City add uncertainty, partly because local housing characteristics such as older building stock and income variability defy broad model assumptions. Additionally, the dataset captures contemporary conditions only, lacking historical depth that could shed light on longstanding trends and shifts in cooling technologies over time.

The researcher also highlights the inadequacies of self-reported data from traditional surveys, where discrepancies between reported and actual AC usage are common. Portable or swamp coolers may be overlooked or underreported despite their significant role in providing relief from heat, particularly in certain climates. By integrating multiple data streams and employing machine learning to correct these biases, this study sets a new standard for precision in environmental and public health data.

Beyond academic knowledge, this research provides tangible tools for diverse stakeholders. Public health officials can identify heat-vulnerable regions lacking effective cooling, enabling the design of targeted outreach and assistance programs. Urban and rural planners can tailor infrastructure investments to meet localized needs and optimize energy efficiency. Energy auditors and private industry players can promote affordable, climate-appropriate cooling solutions, mitigating unnecessary costs and emissions.

The dataset’s implications stretch into climate adaptation strategies as the U.S. confronts rising temperatures and more frequent, deadly heatwaves. Understanding exactly where and how people can cool themselves is crucial to reducing heat-related morbidity and mortality, especially among marginalized and low-income communities. As Professor Ahn’s lab at the University of Kansas continues to develop, forthcoming work aims to expand the dataset historically from 1980 onward and integrate new data sources to refine predictions and broaden impact.

Funded by the National Academy of Sciences Gulf Research Program and the University of Kansas General Research Fund, this research exemplifies how cutting-edge data science and machine learning can intersect with environmental justice and public health. The open access datasets are made publicly available, inviting further exploration and innovation by researchers and policymakers alike. This endeavor signals a critical step forward in our nation’s ability to adapt equitably and effectively to the mounting challenges posed by climate change.

In an era where living through a summer without reliable cooling is increasingly a public health crisis, comprehensive data on air conditioning use is not simply an academic triumph but a societal imperative. Yoonjung Ahn’s work illuminates the path for how data-driven interventions can protect the most vulnerable from the worst impacts of a warming world, ensuring that the promise of safety and comfort remains within reach for all Americans.

Subject of Research: Air conditioning usage patterns and vulnerability assessment in the United States with spatial and machine learning analysis
Article Title: Most comprehensive and detailed map of air conditioning usage in the United States
News Publication Date: 2024
Web References:

• https://www.nature.com/articles/s41597-025-06104-3
• https://dataverse.harvard.edu/dataverse/NRAC-US
  References: Scientific Data Journal, DOI: 10.1038/s41597-025-06104-3
  Image Credits: Not provided
  Keywords: air conditioning usage, climate change, heat vulnerability, spatial modeling, machine learning, environmental justice, public health, urban planning, energy efficiency

Researchers undertook an exhaustive survey of the DRC’s cities, identifying all urban centers officially designated as ‘cities’ and those with populations exceeding 80,000, thereby assembling a near-complete inventory of locations potentially affected by urban gullies. Leveraging ultra-high-resolution satellite imagery from Google Earth, with resolutions finer than one meter, the team meticulously identified gullies meeting stringent geomorphic criteria: elongated channels carved by concentrated runoff exhibiting distinct thalwegs, identifiable gully heads, and pronounced gully edges. A critical spatial constraint was that these gullies had to reside within 200 meters of built environments, underscoring their direct relevance to urban vulnerabilities.

The rigorous geospatial survey was buttressed by extensive ground-truthing campaigns in key cities such as Kinshasa, Kikwit, and Bukavu. Field teams inspected over 400 gullies, confirming their morphological classifications as genuine urban gullies rather than natural landforms. However, smaller gullies proved challenging to detect via satellite due to resolution limits, prompting a focus on gullies featuring a minimum thalweg length of 30 meters for analytical robustness. Historical aerial photographs from the 1950s were cross-examined to distinguish gullies naturally pre-existing before urban expansion from those emerging due to anthropogenic activities, eliminating natural gullies foreign to urbanization dynamics from further analysis.

Mapping the current extents and temporal dynamics of these urban gullies required a multi-temporal remote sensing approach. Utilizing a consistent reference dataset composed of recent, cloud-free high-res imagery from 2021 to 2023, the researchers digitized polygonal representations of gullies across affected cities. The mapping protocol recognized the networked nature of gullies, considering any branching features with discrete gully heads and lengths above the 30-meter threshold as individual entities. Notably, geophysical challenges such as persistent cloud cover and soil composition—exemplified by the clay-rich terrain of Bukavu—necessitated complementary handheld GPS fieldwork to accurately define gully boundaries.

By correlating imagery spanning two decades or more, from early-2000s satellite platforms to recent Pléiades acquisitions, the team quantified areal expansion rates for urban gullies. They delineated between new gully formation, upslope head retreat, and lateral sidewall widening, harnessing geospatial techniques to attribute expansion events precisely. This differentiation is crucial for understanding geomorphic processes and informing early-stage mitigation, given that gully heads pose particular risks through advancing upslope incision, often undermining critical infrastructure. Since satellite revisit intervals preclude pinpointing exact expansion dates, the team employed midpoints between imagery timestamps to approximate timing distributions.

Expanding the inquiry, the research scrutinized potential drivers shaping urban gully occurrence via bivariate and multivariate statistical models at a one-kilometer spatial resolution. Utilizing a sophisticated logistic regression framework refined through backwards stepwise selection, key predictors emerged: urban-built area density, proximity to roads, land cover characterized by tree canopy, soil type, and slope gradients. These variables encapsulate both natural terrain susceptibility and anthropogenic influences such as land cover change and infrastructural footprints. Importantly, reliable digital elevation datasets were a limiting factor, necessitating careful selection of coarser-scale proxies to maintain model validity.

The study’s socio-environmental dimension manifested in the estimation of human displacement induced by urban gullies. Integrating granular population density datasets from the Joint Research Centre’s Global Human Settlement layer with high-resolution gully mapping enabled calculation of populations within zones of recent gully expansion. This displacement metric accounted for variations in both gully area growth and changes in population density over time, interpolating between the five-year population census datasets to enhance temporal resolution. Disaggregating displacement into contributions from new gully initiation, lateral expansion, and head retreat informed differentiated risk assessments critical for urban planning.

In parallel, the team delineated hazard zones reflecting exposure risks to gully expansion. By defining buffer zones of varying radii around gully polygons—ranging from immediate 100-meter buffers to statistical estimates of maximum gully widths and retreat distances—researchers encapsulated both direct and potential future impacts. Intriguingly, gullies developing on sandy substrates demonstrated notably larger widths and faster retreat velocities than counterparts on non-sandy soils, highlighting substrate composition as a fundamental geomorphic control affecting urban gully dynamics and consequent hazard footprints.

Population exposure trends between 2010 and 2023 were dissected by overlaying hazard zones with temporal population distributions, clarifying drivers behind rising vulnerability. The analyses teased apart the effects of demographic growth within pre-existing hazard zones, spatial expansion of established gullies, and formation of new gullies, revealing complex interactions between demographic pressures and geomorphic evolution. This nuanced exposure mapping provides a valuable blueprint for targeted interventions.

Recognizing the importance of robustness, the researchers conducted uncertainty assessments contrasting estimates derived from JRC GHS population data with alternative datasets like WorldPop, alongside detailed local census data available for the city of Bukavu. The cross-validation underscored the likelihood that JRC GHS-based estimates, while slightly conservative, better capture urban population densities in high-risk zones than other global datasets prone to underestimations. Consequently, displacement and exposure figures may indeed represent lower-bound estimates, accentuating the urgency of addressing urban gully hazards.

This comprehensive investigation of urban gullies in the DRC reveals an underappreciated environmental threat intertwined with rapid urbanization and fragile geomorphological contexts. The implications stretch beyond immediate hazards: the socio-economic fabric of cities faces relentless pressures from land degradation, infrastructure loss, and involuntary displacement. As urban expansion accelerates in the developing world, these findings sound a stark call for integrating geomorphic hazard assessments within urban planning and development policies to safeguard vulnerable populations.

The multi-disciplinary methodology combining remote sensing, field measurements, statistical modeling, and population analytics offers a powerful template for similar investigations worldwide. It demonstrates how detailed spatial-temporal analyses can unravel emergent environmental crises masked by urban growth. Crucially, the study accentuates the need for finer resolution terrain and infrastructural data, localized soil characterizations, and nuanced population monitoring to enhance predictive capacity and establish early warning frameworks.

Looking forward, confronting the urban gully menace demands coordinated efforts spanning engineering solutions to stabilize susceptible terrains, participatory urban governance attuned to geomorphic hazards, and investment in resilient infrastructure design. Understanding the hydrological triggers and anthropogenic disturbances underlying gully initiation could foster preventative measures, while socio-economic support for displaced populations remains paramount. The DRC’s urban gullies embody the complex entanglement of natural processes and human development, serving as a cautionary exemplar as cities worldwide grapple with climate change-enhanced erosion and land degradation.

Ultimately, this landmark study dramatically elevates urban gullies from obscurity to a recognized urban hazard in the DRC, revealing their spatial extent, temporal dynamism, and deep societal ramifications. It beckons further interdisciplinary inquiry and policy mobilization to stem the tide of land loss and human displacement reshaping urban futures in developing nations facing rapid, often unplanned, urban growth.

Subject of Research: Urban gullies and their spatial and temporal dynamics in the Democratic Republic of the Congo, with emphasis on geomorphic processes, urban growth interactions, and population displacement.

Article Title: Mapping urban gullies in the Democratic Republic of the Congo

Article References:
Mawe, G.I., Landu, E.L., Dujardin, E. et al. Mapping urban gullies in the Democratic Republic of the Congo. Nature 644, 952–959 (2025). https://doi.org/10.1038/s41586-025-09371-7

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s41586-025-09371-7