Urban Currents

Exploring street-level landscape patterns around illegal dumping monitoring locations using unsupervised computer vision

The study developed an unsupervised computer vision pipeline to classify street-level landscapes around frequent illegal dumping sites in Seoul, South Korea. Using feature extraction, dimensionality reduction, and clustering on street-level images, followed by Grad-CAM for visual interpretation, the method identified three distinct landscape typologies: low-rise mixed-use areas with commercial signs, dense aged low-rise residential zones, and poorly managed, vegetation-dominated spaces. A manual validation experiment using 200 randomly sampled images confirmed a moderate correspondence between the unsupervised clusters and human annotations.

Why it matters — It establishes a scalable, automated screening method that can complement or replace resource-intensive field audits, allowing municipal authorities to proactively identify and prioritize potential illegal dumping hotspots based on visual environmental cues.

Caveat: The classification performance and identified typologies are validated only on a sample of 200 images from a single city, and the correspondence with human annotations is moderate rather than strong.

Computational Urban Science · pdf · doi

A Fuller City: Rekindling the work of Buckminster Fuller in the age of planetary urban crisis

The paper analyzes three designs by Buckminster Fuller—the Geodesic Dome, the Tetrahedronal City for Tokyo Bay, and Skyrise for Harlem—through the respective urban lenses of architecture, infrastructure, and utopia. It critically evaluates how these mid-twentieth-century concepts intersect with contemporary urban crises and the politics of technological solutions.

Why it matters — It introduces Fuller's design history into modern urban studies, offering a framework to use his speculative designs as provocations for imagining alternative urban futures rather than treating them as literal technological blueprints.

Caveat: The analysis is theoretical and historical, focusing on conceptual design imaginaries rather than empirical urban data or contemporary case studies.

Cities · doi

Which conditions make the train a convenient modal alternative in interurban commuting for different commuting profiles? A comprehensive analysis based on mobile phone data

This study analyzes regional interurban train commuting in Italy's Friuli-Venezia Giulia region using an origin-destination matrix derived from mobile phone data. Using Latent Class Analysis, the researchers identified four distinct commuter profiles and applied Generalized Linear Mixed Models to evaluate how first- and last-mile factors influence their travel choices.

Why it matters — It demonstrates how mobile phone big data can segment regional commuters into behavioral profiles, revealing how specific first- and last-mile conditions influence the viability of rail transit for different types of travelers.

Caveat: The findings are based on a single region in Italy, which may limit their direct applicability to regions with different transit infrastructure or spatial layouts.

Journal of Transport Geography · pdf · doi

Mapping urban energy transitions: Decoding solar pv potential, energy burden, and socio-demographic dimensions in the athens metropolitan area

This study models rooftop solar photovoltaic (PV) potential, electricity demand, and energy burden across all municipalities in the Athens Metropolitan Area using high-resolution 3D radiative modeling and Earth observation data. It finds that 24% of households (affecting 1.2 million residents) exceed a 10% energy burden threshold, with vulnerability concentrated in compact, high-density zones. The model estimates that full rooftop PV deployment could save 1103 kt of CO2 emissions, while even a 10% deployment rate would halve the number of energy-burdened municipalities.

Why it matters — It establishes a spatially explicit link between physical solar potential and socio-demographic vulnerability, showing that the areas with the highest energy burdens are also those with the lowest physical capacity for rooftop solar self-sufficiency.

Caveat: The findings rely on a simulated deep building renovation scenario to estimate electricity demand rather than direct, empirical consumption data.

Sustainable Cities and Society · doi

Smart point cloud–guided 3D gaussian splatting for urban modeling and data-driven analysis

The researchers developed Smart 3D Gaussian Splatting (S3DGS), a framework that fuses laser-scanning point clouds with multi-view aerial and street-level imagery into a single coordinate system. Tested on the Aula-library block in Delft, the workflow propagates semantic labels from a CityGML-inspired point cloud to Gaussian primitives, achieving an alignment root mean square error of 0.0439 meters and a k-nearest neighbors label consistency of 92.24%. This co-referenced model enables simultaneous analysis of volumetric spatial presence, visual exposure, and surface composition within the same urban unit.

Why it matters — It bridges the gap between separate 3D modeling, geospatial mapping, and street-level image workflows, allowing planners to detect complex spatial conditions like building masses that are physically present but visually hidden from pedestrian viewpoints.

Caveat: The framework's performance and alignment stability were evaluated on only a single urban block in Delft.

Sustainable Cities and Society · doi

A New Approach to collinearity and elasticity problems in value of travel time savings estimation with multinomial logit models

The study introduces an elasticity-constrained estimator for multinomial logit models to address imprecise Value of Travel Time Savings (VTTS) calculations caused by collinearity and low price-elasticity. The estimator was evaluated using simulated datasets and the Swissmetro mode choice dataset, comparing its performance against maximum likelihood and ridge estimation.

Why it matters — It provides a method to stabilize VTTS estimation in revealed-preference datasets where travel time and cost are highly collinear, reducing mean squared error and estimator variance by anchoring the cost coefficient to external elasticity evidence.

Caveat: The proposed estimator introduces bias if the actual demand elasticity differs significantly from the external restriction applied.

Transportation Research Part A Policy and Practice · pdf · doi

How is online shopping reshaping activity patterns? Modeling activity behavior using discrete continuous models

This study analyzed changes in daily time-use patterns across six activity categories using the 2013 and 2023 waves of the American Time Use Survey. Employing a multiple discrete continuous framework, the model captures how online shopping interacts with other activities, revealing that virtual shopping complements in-store shopping and at-home leisure but acts as a substitute for outside-of-home leisure and maintenance tasks.

Why it matters — It quantifies how virtual retail reshapes broader daily schedules, demonstrating that while baseline shopping preferences held steady over ten years, at-home leisure and maintenance increased while out-of-home activities declined.

Transportation Research Part A Policy and Practice · doi

Effects of off-street parking and car sharing on car use and car ownership in low-parking developments

This study analyzed survey data from residents of 24 multi-family housing developments in Gothenburg and Malmö, Sweden, to evaluate how parking supply and car-sharing options relate to vehicle ownership and usage. Properties with higher parking ratios showed a positive correlation with household car ownership, and residents in these high-parking buildings spent approximately 30% of their travel time driving compared to just 10% for those in low-parking properties. Active car-sharing users maintained significantly lower car ownership rates than non-users, while property-level parking supply frequently exceeded actual household car ownership.

Why it matters — The findings demonstrate that municipal minimum parking requirements are often set higher than actual demand, and they provide empirical evidence that combining reduced parking minimums with car-sharing services is associated with both lower vehicle ownership and reduced driving time.

Caveat: The study relies on correlational survey data from self-selected housing developments, meaning it cannot definitively establish a causal link between parking reductions and behavioral changes.

Transport Policy · doi

Port environmental health Risks: Insights from the Tarragona Port-Industrial complex and planning implications

This review synthesizes literature on the health risks of living near ports using a source-to-exposure-to-impact framework, and applies these insights to the Tarragona Port-Industrial complex. The synthesis links port emissions to cardiovascular and respiratory mortality, lung cancer, asthma, adverse pregnancy outcomes, and neurodevelopmental issues, while highlighting the local burden of preventable deaths and disability-adjusted life years in Tarragona.

Why it matters — It demonstrates the challenge of isolating port-specific health impacts from co-located petrochemical and urban pollution sources, providing a justification for planning interventions like mandatory Health Impact Assessments and health-based buffer zones.

Caveat: The specific fraction of health burden caused by port activities alone could not be isolated from the emissions of the adjacent petrochemical complex and other urban sources.

Transportation Research Part D Transport and Environment · pdf · doi

A scientometric review of unmanned aerial vehicles (UAVs) and transportation science

This paper maps the intersection of unmanned aerial vehicles and transportation research using a suite of scientometric approaches. The analysis identifies hybrid truck-drone logistics operations as the dominant research focus, while highlighting growth in urban infrastructure inspection and surveillance applications.

Why it matters — It reveals that emerging UAV applications in transportation, such as infrastructure monitoring, are highly dependent on localization technology developments occurring in allied fields, exposing critical cross-disciplinary dependencies that are often overlooked.

Transportation Research Interdisciplinary Perspectives · doi

Prioritization and Implementation of Safe Routes to a School in Austin, Texas

This study analyzed 4,654 Safe Routes to School (SRTS) recommendations across 137 schools in Austin, Texas, spanning 2016 to 2023, to understand how project types influence prioritization and actual implementation. Using text network analysis and mixed-effects logit models, the researchers found that while bike facilities and trail connections have the highest perceived safety benefits (odds ratios of 13.6 and 6.57), they have low cost-benefit rankings and are 53% and 65% less likely to be implemented. Conversely, speed-related measures and crosswalk improvements are highly favored for implementation due to superior cost-benefit scores, with speed-related projects being 2.21 times more likely to be built.

Why it matters — It reveals a systematic disconnect in municipal planning where infrastructure projects with the highest safety benefits are routinely passed over for implementation in favor of cheaper, more economically efficient alternatives.

Caveat: The findings are based on a single city's SRTS program and administrative decisions, which may reflect local funding structures and political priorities unique to Austin, Texas.

Journal of Transportation Engineering Part A Systems · doi

Racing with the Concepts: Using F1 as Way to Explore Economic Geography

This case report details an undergraduate course project that uses Formula One racing as a pedagogical tool to teach core economic geography concepts and global economic dynamics. Students engaged with the sport's structure to conduct location analyses and examine the spatial movement of goods, services, and capital.

Why it matters — It demonstrates a concrete, student-centered curriculum design that translates abstract economic geography theories into tangible spatial relationships using a highly globalized, real-world industry.

Caveat: The assessment of the project's success is based on qualitative pedagogical results rather than a quantitative evaluation of student learning outcomes.

Journal of Geography · doi

System optimality versus self-organization in metro networks: An optimal transport analysis

This study introduces a data-driven framework using optimal transport theory and geometric comparisons to evaluate metro networks against three benchmarks: a self-organized network based on local rules, a system-optimal network based on global optimization, and the actual empirical network. Applying this framework to Singapore's Mass Rapid Transit system, the analysis reveals that the real-world network aligns much more closely with the self-organized benchmark than the system-optimal one, with the remaining differences primarily driven by geographic constraints.

Why it matters — It demonstrates that real-world transit networks evolve closer to local, self-organized demand patterns than to theoretical global optimums, suggesting that planners should design infrastructure around local service needs rather than relying solely on top-down, system-wide optimization.

Caveat: The empirical findings and validation of the framework are demonstrated using only a single transit system, Singapore's Mass Rapid Transit.

Computers Environment and Urban Systems · doi

Integrating Structure and Attributes for Transportation Network Partitioning via Optimal Transport

The study introduces a flexible framework for partitioning heterogeneous transportation networks by representing them as attributed graphs and applying an optimal transport formulation based on the semi-relaxed Fused Gromov-Wasserstein discrepancy. This approach allows researchers to explicitly control the trade-off between network structure and diverse operational attributes, such as scalar indicators or temporal profiles. The methodology was validated on two distinct systems: an urban road network for traffic-oriented partitioning and a bicycle-sharing system for identifying usage-based communities.

Why it matters — It replaces rigid, predefined integration formulas with a tunable framework, allowing planners to customize network partitions based on whether they want to prioritize physical connectivity or operational behavior.

arXiv · pdf

Economic Complexity as a Determinant of Regional Human Development in Brazil: Evidence across Aggregation Scales

This study adapted the Economic Complexity Index to Brazil and evaluated how local productive sophistication and road network topology predict the Municipal Human Development Index (IDHM). Using linear models (Ridge, LASSO, Elastic Net) and nonlinear models (Decision Trees, Explainable Boosting Machine) across two spatial scales, the researchers found that regional aggregation reduces statistical noise. The Explainable Boosting Machine achieved its highest predictive performance of R^2 = 0.8196 at the Immediate Geographic Region level when incorporating road network metrics.

Why it matters — It demonstrates that regional productive sophistication is a dominant structural driver of human development in Brazil, and that integrating physical infrastructure connectivity into economic complexity models significantly improves their predictive power.

Caveat: The predictive performance and the relative importance of economic complexity depend heavily on the spatial scale of analysis, with municipal-level models exhibiting higher statistical noise and lower stability than regionally aggregated ones.

arXiv · pdf

Unequal Trips, Unequal Places: Diagnosing and Mitigating Delay Inequity in Autonomous Vehicle Fleet Coordination

The study audits trip-length and spatial delay inequities in autonomous vehicle fleet coordinators using taxi-demand and road-network datasets from Manhattan, Chicago, and San Francisco. It introduces SPatially Aware RErouting (SPARE), an online coordination framework that dynamically redirects delayed vehicles based on localized waiting pressure. Evaluated against six baseline models, SPARE improves both travel efficiency and fairness across all three cities without requiring full-fleet replanning.

Why it matters — It demonstrates that optimizing for aggregate fleet travel times systematically disadvantages specific neighborhoods and trip lengths, and proves that targeted, bounded rerouting can mitigate these spatial inequities without sacrificing overall city-scale scalability.

Caveat: The findings and framework performance are evaluated using historical taxi-demand datasets as a proxy for autonomous vehicle fleet demand.

arXiv · pdf

Eliminating Propagation Delay: Attention-Based Spatial-Temporal Fusion Graph Convolution Network for Traffic Flow Prediction

The researchers developed the Attention-Based Spatial-Temporal Fusion Graph Convolution Network (A-STFGCN) to predict traffic flow. The model uses a spatial-temporal fusion block to eliminate information propagation delays between adjacent nodes and integrates a masked multi-head self-attention mechanism to capture both long-term and short-term temporal patterns. The network was evaluated against eight baseline methods using five real-world datasets.

Why it matters — It introduces a way to account for varying propagation delays between neighboring sensors in traffic networks, while reducing the computational training time typically associated with highly stacked, complex deep learning architectures.

arXiv · pdf

Forecasting the Emergence and Evolution of Crash Hotspots: A Unified Deep Learning Framework for Proactive Traffic Safety

The HERALD deep learning framework combines a CNN-Transformer with a mixture-of-experts architecture to forecast weekly crash risks, detect emerging hotspots, and track their life cycles. Evaluated across six diverse counties in Wisconsin, the model uses recent crash histories and long-run spatial patterns to predict next-week crash locations, outperforming five baseline models in both spatial precision and early risk detection.

Why it matters — It shifts traffic safety enforcement from a reactive posture based on historical crash maps to a proactive system capable of identifying and tracking transient, short-term hotspots as they form and decay.

Caveat: The framework's performance and applicability are demonstrated using data from only six counties within a single state.

arXiv · pdf

Quantum-Inspired Evolutionary Neighborhood Search for Arrival-Departure Track Utilization Adjustment under Short-Term Disturbances

The study develops a quantum-inspired evolutionary algorithm combined with neighborhood search (QEA-NS) to adjust arrival-departure track allocation and train timing during short-term station disturbances. Using GTFS timetable data from Frankfurt Hauptbahnhof, Germany, to construct 10 random perturbation instances, the model represents station operations as zone-level resource-occupation intervals. Compared to the CP-SAT solver, the QEA-NS method reduced total train delay by 25.2% from 519 to 388 minutes, and lowered the mean delay of affected trains from 4.99 to 3.73 minutes.

Why it matters — It demonstrates a scheduling approach that consistently achieves lower total delays and higher stability under random perturbations than standard constraint programming solvers, though at the cost of longer computation times.

Caveat: The proposed algorithm requires longer solution times than the baseline CP-SAT solver, which may limit its immediate deployment for real-time dispatching until computational efficiency is improved.

arXiv · pdf

Monitoring Post-Disaster Urban Recovery Using High-Resolution SAR Time Series and Unsupervised Learning: Evidence from the 2023 Türkiye-Syria Earthquake

An unsupervised deep-learning framework was developed to monitor post-disaster urban reconstruction using multi-temporal COSMO-SkyMed synthetic aperture radar (SAR) time series. Tested across four cities impacted by the 2023 Türkiye-Syria earthquake, the model maps persistent temporal anomalies to identify rebuilding activities, temporary container settlements, and new residential districts. These SAR-derived structural changes were compared with nighttime-light recovery indicators from SDGSAT-1, which track electricity restoration and socioeconomic activity.

Why it matters — The framework establishes a scalable method to track physical reconstruction dynamics in real time without requiring labeled ground-truth datasets. By combining SAR and nighttime-light data, researchers can now distinguish between the physical rebuilding of structures and the functional return of electricity and human activity.

arXiv · pdf