Urban Currents

City2Graph: A Python library for Heterogeneous Graph Neural Networks and spatial analysis in urban systems

The researchers developed City2Graph, an open-source Python library that standardises the construction of heterogeneous Graph Neural Networks (GNNs) across urban morphology, transportation, mobility, and proximity domains. The library converts spatial geometries and network topologies into GNN-ready tensors and supports metapath construction to capture higher-order connections. Its utility was demonstrated in Liverpool, UK, where a heterogeneous Graph Autoencoder model using spatial contiguity, walking accessibility, and multimodal transit relations clustered urban functions more coherently than a homogeneous baseline.

Why it matters — It bridges the gap between spatial data science and deep graph learning by providing a unified, reproducible pipeline to build and train heterogeneous GNNs, which previously suffered from fragmented workflows across different data domains.

Caveat: The library's performance advantages were demonstrated on a single case study of urban function clustering in Liverpool, meaning its comparative benefits may vary for other spatial prediction or classification tasks.

Computers Environment and Urban Systems · doi

Monitoring pedestrian walking dynamics under environmental and policy influences in urban spaces

A computer-vision framework combining real-time object detection and multi-object tracking was developed to extract pedestrian trajectories from a year-long video dataset in a busy shopping district in Fukuoka, Japan. The system analyzed walking speeds across four daily time slots under varying temperatures and during three COVID-19 State of Emergency periods, revealing that morning commuters walked faster and were more sensitive to temperature, while nighttime pedestrians responded more strongly to policy restrictions.

Why it matters — It demonstrates that the influence of environmental and policy factors on pedestrian behavior is highly dependent on time of day and walking purpose, capturing critical behavioral variations that are typically lost in conventional daily-aggregate analyses.

Caveat: The empirical findings are derived from a single camera system monitoring a single shopping district, which may limit direct generalization to other urban contexts.

Computational Urban Science · pdf · doi

Socioeconomic determinants of per capita household carbon dioxide emissions: evidence from Lucknow India

This study estimated per capita household carbon dioxide emissions from electricity, LPG cooking, and private vehicles for 270 households in Lucknow, India, using the emission coefficient method and double-log OLS regression. A 1% increase in per capita household income was associated with a 0.50% increase in electricity emissions and a 1.07% increase in transport emissions, while LPG emissions showed no significant change. Higher emissions were also linked to households in developed urban areas and those from advantaged socio-economic and caste groups.

Why it matters — It provides empirical evidence on how intra-urban spatial disparities and India-specific socio-cultural factors, such as caste, shape household carbon footprints in a rapidly urbanizing tier-2 Indian city.

Caveat: The findings are purely associational and may only generalize to similar rapidly urbanizing tier-2 cities in India.

Discover Cities · pdf · doi

The association pathways between rapid urban growth and urban governance practices in Hawassa city, Ethiopia

This study analyzed how population growth, settlement density, and horizontal expansion affect five core urban governance principles in Hawassa, Ethiopia, using survey data from 424 households alongside qualitative interviews. Path analysis revealed that while population growth negatively impacts all governance principles, especially equity (beta = -0.425), and horizontal expansion reduces participation and transparency, higher population density consistently improves the implementation of all governance principles, particularly participation (beta = 0.362) and transparency (beta = 0.304).

Why it matters — The findings demonstrate that the governance challenges of rapid urbanization are not uniform, showing that compact urban forms can actually facilitate better governance practices rather than hindering them.

Caveat: The empirical findings are based on a single-city case study of Hawassa, Ethiopia, which may limit generalizability to other rapidly growing cities with different institutional structures.

Discover Cities · pdf · doi

Smarter and healthier cities in the era of AI: A state-of-the-art review

This study synthesizes the intersection of smart and healthy city research through a systematic literature review following the PRISMA protocol. It analyzes academic work across governance, environment, and modeling domains to define the concept of 'Smarter and Healthier' cities, mapping how artificial intelligence drives shifts toward participatory governance, dense IoT sensing, and edge intelligence.

Why it matters — It provides a unified conceptual framework that bridges technological innovation with human-centered goals, establishing explicit principles for ethical safeguards, environmental justice, and trustworthy AI in urban systems.

Caveat: As a conceptual review, the paper outlines frameworks and principles rather than testing these AI-enabled governance or environmental models in a real-world deployment.

Cities · pdf · doi

Integrating AI into urban vitality promotion: The impact of building morphology on daily consumption quality in Hong Kong

This study developed a Generative Adversarial Network (GAN) model to predict daily consumption quality, measured by user ratings of commercial points of interest, based on building morphology variables like height, age, and architectural features like podiums with atriums. Using open-source spatial datasets from Hong Kong, the model establishes quantitative correlations to simulate how specific urban design alterations affect local consumption environments.

Why it matters — It replaces qualitative or non-predictive urban design heuristics with a spatially explicit, predictive tool, allowing planners to quantitatively test how changes in building height and age will impact local commercial vitality before construction begins.

Caveat: The predictive model is trained and validated solely on the high-density urban context of Hong Kong, which may limit its applicability to low-density or structurally different cities.

Cities · doi

Perceiving “city of proximity” beyond spatial closeness: Explaining travel and consumption behavior through perceptions and preferences

This study analyzes how individual preferences and perceptions of urban environments influence travel and consumption behaviors using structural equation modeling. Based on online survey data from the Brussels-Capital Region, Belgium, the model evaluates the mediating role of perceived proximity alongside objective physical access.

Why it matters — It demonstrates that personal preferences for mobility and urban facilities shape behavior more strongly than the physical environment itself, showing that objective proximity metrics alone are insufficient for predicting how residents use 15-minute city designs.

Caveat: The findings rely on self-reported travel and consumption behaviors from a single metropolitan region rather than direct GPS or transaction measurements.

Cities · doi

How can land development intensity and land use benefit be coordinated in high-speed railway station areas? Evidence from 1570 station areas in China

This study evaluated the coordination between land development inputs and land use outputs across 1570 high-speed railway station areas in China using multi-source spatial data. The researchers applied entropy-weighted TOPSIS, coupling coordination degree (CCD) models, XGBoost-SHAP machine learning, and PLS-SEM structural path analysis to classify and analyze these areas. The analysis revealed that 61.8% (970 stations) operate as low-efficiency development areas, while 38.2% (600 stations) are high-efficiency areas, with local 'place-dimensional' factors like bus stop density, road network length, and land use diversity serving as the primary predictors of coordination quality.

Why it matters — It identifies specific, actionable local infrastructure thresholds and land-use diversity metrics that planners can target to resolve the common mismatch between high-density physical construction and low economic or functional output around transit hubs.

Cities · doi

Pollinator habitat enhancement through ornamental plants, floral traits and urban green area features for planning biodiversity-friendly cities

Researchers surveyed plant-pollinator interactions across seven Italian cities during May to rank 43 ornamental plant species based on pollinator abundance and species richness. The study analyzed how specific floral traits and environmental contexts, including parks, historic villas, and botanic gardens, influenced insect visitation rates.

Why it matters — It establishes a ranked baseline of ornamental plants and floral traits that support urban pollinators, demonstrating that habitat heterogeneity and trait diversity are critical design factors for urban green infrastructure.

Caveat: The field surveys were conducted exclusively during the month of May, which limits the findings to late-spring pollinator dynamics.

Landscape and Urban Planning · pdf · doi

Expectations and land-value premiums: The metro effect on residential land in a consolidated city

This study measures the land-value premiums generated by the announcement and early construction of Bogota's first two metro lines: a 23.96 km elevated line (L1) and a planned underground line (L2). Using Coarsened Exact Matching and weighted log-linear regressions on cadastral data, the analysis finds that residential land near L1 appreciated by 11.7% in 2019 upon announcement and retained an 8.8% premium by 2023 when construction reached 20% completion, while the underground L2 announcement generated an 11.8% premium. The appreciation is highly dependent on vertical morphology, with blocks averaging fewer than four stories experiencing significantly smaller gains than taller blocks.

Why it matters — It provides the first empirical quantification of anticipation-driven land-value changes in a consolidated Global South city, demonstrating that both elevated and underground alignments generate nearly identical initial price premiums, though the actual capture of these gains is constrained by existing building heights.

Caveat: The study relies on cadastral data to estimate premiums from project announcements and early-stage construction, rather than measuring the long-term impacts of an operational transit system.

Research in Transportation Economics · pdf · doi

Between education and employment: How apprentices navigate commuting mobility challenges across spatial contexts in Austria

This study analyzes the commuting challenges of apprentices in Austria's dual education system using qualitative content analysis of 12 semi-structured interviews with youth aged 16 to 23. The research covers urban, peri-urban, and rural settings across the Vienna metropolitan area and Upper Austria to examine how these students manage travel between vocational schools and dispersed training companies.

Why it matters — It identifies a distinct, structurally vulnerable group of young commuters who face heightened risks of transport and time poverty due to early working hours and poor public transit outside urban centers. The findings demonstrate that while car dependency and long commutes are highly normalized among these youth, actual coping resources like household support remain highly unequal.

Caveat: The findings are based on a small qualitative sample of twelve interviews, which limits statistical generalizability across the broader Austrian apprentice population.

Journal of Urban Mobility · pdf · doi

Transport disadvantage and suppressed travel among low-income older adults: Lived experiences of social housing tenants in Ghent

Through semi-structured interviews with 16 social housing tenants aged 54 to 83 in Ghent, Belgium, this study examines how financial, health, cognitive, and emotional barriers lead to suppressed travel. The research documents how declining public transport services, driving stress, and planning anxiety cause vulnerable older adults to forgo desired trips, particularly for leisure and family visits outside the city.

Why it matters — The study demonstrates that high urban density and physical proximity to services are insufficient to prevent unmet travel needs, challenging the assumption that dense urban environments automatically resolve transport disadvantage for vulnerable populations.

Caveat: The findings are based on a small, qualitative sample of 16 individuals in a single European city, which limits direct statistical generalization to other urban contexts.

Journal of Urban Mobility · pdf · doi

Smart cities fall short: why ecoadaptive urban intelligence is essential for climate resilience

This perspective introduces the Ecoadaptive Intelligence Theory, a conceptual framework that reframes urban systems as socio-ecological-computational assemblages. The theory shifts the focus of urban intelligence from prediction, optimisation, and control toward co-learning and adaptive responses driven by spatial form and environmental feedback.

Why it matters — It challenges the dominant smart city paradigm by showing that traditional optimization models are insufficient for the deep uncertainty of climate change, offering a new theoretical pathway to integrate ecological feedback directly into urban computational systems.

Caveat: The paper is a theoretical perspective and does not present empirical data, quantitative modeling, or a concrete implementation of the proposed framework.

npj Urban Sustainability · pdf · doi

Deep Learning–Based Speed Reduction Modeling for Independent Tangent–Curve Transitions on Four-Lane Divided Rural Highways

This study established a threshold tangent length to distinguish between short and long tangents using acceleration and deceleration data, then modeled the 85th percentile speed reduction across 32 independent tangent-to-curve transitions on flat, four-lane divided rural highways. Using instrumented vehicle data, the researchers compared multiple linear regression to a deep neural network model, finding that curve radius and deflection angle are the primary predictors of speed reduction.

Why it matters — The deep neural network model predicts speed reductions with much higher accuracy (R-squared of 0.957 versus 0.563 for regression), providing a more reliable tool for highway engineers to evaluate design consistency, identify hazardous transitions, and set appropriate speed limits.

Caveat: The predictive models are based on a small sample of 32 transition zones restricted entirely to flat terrain.

Journal of Transportation Engineering Part A Systems · doi

MDWD: A Street-Level Dataset for Municipal Solid Waste Detection in Dense Urban Environments

The Maltese Domestic Waste Dataset (MDWD) is a new street-level benchmark containing 3,697 high-resolution images and 11,461 manually annotated instances of domestic waste across five categories. Evaluation of multiple object detection models on the dataset shows that the RF-DETR-M transformer-based detector achieves the highest performance with an mAP50 of 94.49% and an F1-score of 93.56%, while smaller YOLO variants remain highly competitive.

Why it matters — It provides the first street-level, instance-level dataset specifically categorized around domestic waste streams in a structured municipal collection context, enabling the development of automated visual monitoring systems for dense urban environments.

Caveat: The dataset is tailored to the specific municipal collection system and urban context of Malta, which may limit direct generalization to cities with different waste disposal practices.

arXiv · pdf

Ring-based Spatial Transformer: Learning Non-linear Spatial Interactions between Building Distribution and Pedestrian Flow

The study developed a ring-based Spatial Transformer model to analyze how building uses across concentric 100-meter ring buffers, up to an 800-meter radius, influence pedestrian flow around 100 randomly selected railway stations in Tokyo. Using GPS-derived walking trip counts as the target variable, the model outperformed Geographically Weighted Regression across 30 independent trials. SHAP and attention matrix analyses revealed that mid-to-outer distance zones (beyond 100 meters) dominate pedestrian flow predictions and interact strongly with other distant zones.

Why it matters — The findings challenge the conventional compact city planning assumption that concentrating development immediately adjacent to transit stations maximizes pedestrian activity, demonstrating instead that pedestrian flow is driven by land-use interactions across the entire walkable catchment area.

Caveat: The model's findings are based on a sample of 100 stations within a single, highly transit-oriented metropolitan area, which may limit generalizability to cities with different urban forms or transit habits.

arXiv · pdf

Poverty Mapping: Data, Models and Applications

This review synthesizes recent methodological advances in poverty mapping that utilize nontraditional data sources, including satellite imagery, mobile phone records, social media activity, and multisource data fusion. It outlines the core concepts and measurement frameworks of poverty, while evaluating how computational methods from statistical physics, complex systems science, and data science enable finer spatial and temporal resolutions than traditional census surveys.

Why it matters — It establishes a comprehensive framework for understanding the analytical capabilities and practical boundaries of emerging digital data sources, shifting the field from ad-hoc case studies toward a systematic evaluation of model transferability, data representativeness, and uncertainty quantification.

arXiv · pdf