Urban Currents

Toward Parking Spot Occupancy Recognition: A Self-Supervised Approach

The researchers developed a self-supervised transfer learning method for parking spot occupancy recognition using a SimCLR framework with a ResNet-50 encoder. The model was evaluated using a leave-one-out cross-environment protocol across three public datasets (PKLot, CNRPark-EXT, and PLds), achieving an average accuracy of 97.2% with a general model and 97.8% when incorporating unlabeled target-site images from the first N days of deployment.

Why it matters — It establishes a highly accurate parking monitoring system that can be deployed in new, unseen parking lots without requiring any manual labeling of images from the target site.

arXiv · pdf · code

User centric multimodal urban transportation network equilibrium including intermodality and shared mobility services

This study reviews multimodal transportation models incorporating shared mobility services and introduces a new multimodal traffic assignment model formulated as a Mixed-Integer Bilinear Programming problem. The model accommodates both continuous and integer settings, allowing commuters to combine public transit and shared mobility to optimize travel time and monetary costs. The framework was evaluated using two test cases to analyze user mode and path choices alongside the price of anarchy.

Why it matters — It provides a mathematical framework that explicitly captures the synergies and combinations between public transit and shared mobility services, correcting a common gap in traffic assignment models that treat these systems in isolation.

Caveat: The model's performance and scalability were only demonstrated on two simplified test cases rather than a real-world metropolitan network.

European Transport Research Review · pdf · doi

Systemic challenges and drivers of urban transformation in the oil city of Ghana

This study investigates the drivers of informal and fragmented urban development in Sekondi-Takoradi, Ghana's oil city, using a sequential mixed-methods approach. Quantitative analysis using binary logistic regression on survey data from planning committee members, customary authorities, assembly members, and property owners yielded low explanatory power, with Nagelkerke R² values between 0.011 and 0.037. Qualitative interviews resolved this weak statistical predictive power by identifying weak governance, tenure insecurity, public non-compliance, and procedural gaps as the primary forces shaping the physical landscape.

Why it matters — It demonstrates that formal spatial plans and regulatory frameworks in rapidly urbanizing Sub-Saharan cities fail because informal land commodification and institutional weaknesses create a disconnect between official planning maps and actual spatial development.

Caveat: The quantitative models suffered from very low explanatory power, meaning the statistical relationships analyzed cannot reliably predict spatial outcomes on their own.

Discover Cities · doi

Co-designing urban soundscapes with citizens and experts: Evidence from a field experiment

A field experiment in central London engaged citizens in co-designing public space soundscapes using aural-visual scenarios, including acoustic-metamaterial and acoustic-wall interventions, yielding 144 design proposals. These proposals were evaluated using the Soundscape Perception Index (SPI) against 'Vibrant' and 'Calm' targets to measure the impact of group deliberation and structured expert input. The analysis revealed that collaborative group designs consistently outperformed individual ones, and that expert guidance specifically enhanced group performance for the challenging 'Calm' target.

Why it matters — The study demonstrates that expert technical input does not have to suppress citizen agency; instead, when integrated into a collaborative group process, it directly improves the acoustic performance of community-designed spaces.

Caveat: The findings are based on a single co-design experiment in central London, which may limit generalizability to other cultural or urban contexts.

Landscape and Urban Planning · doi

Urban airflow field prediction using sparse sensors: A hybrid data-driven approach with reduced-order modeling

The study develops a three-stage surrogate modeling framework that reconstructs and predicts real-time urban wind velocity fields around a high-rise building configuration using sparse online sensor data. The method integrates a bidirectional long short-term memory (BILSTM) network for temporal dynamics, sparsity-promoting dynamic mode decomposition (SP-DMD) for spatial mode extraction, and a multi-task learning (MTL) network to map sensor signals. The framework's performance was validated against high-fidelity computational fluid dynamics (CFD) simulations under varying levels of sensor noise.

Why it matters — It provides a computationally efficient, real-time method to reconstruct complex urban airflow fields from minimal physical sensors, bypassing the prohibitive computational costs of traditional CFD simulations for applications like drone routing and pedestrian comfort.

Caveat: The framework was evaluated only on a simplified, single high-rise urban building configuration rather than a complex, real-world city neighborhood.

Sustainable Cities and Society · doi

Government holdings and performance in the airline industry: A focus on African and Middle East airlines

This study analyzed the impact of government ownership on airline operating profitability using a panel dataset of 162 airlines globally from 2007 to 2019. Employing a two-stage least squares (2SLS) regression to control for endogeneity, the analysis revealed a non-linear, quadratic relationship where moderate state ownership generally improves profitability, though airlines in Africa and the Middle East experience stronger positive profitability effects at higher levels of government ownership.

Why it matters — It establishes that the financial impact of state ownership is not uniform, demonstrating that regional market contexts in Africa and the Middle East alter the optimal balance of public and private equity in the aviation sector.

Transportation Research Part A Policy and Practice · doi

Predictive stochastic repurposable supply chain: an adaptive strategy toward supply chain plasticity

The study introduces a Repurposable Supply Chain (RSC) framework that adapts to long-term crises by shifting operations to emerging product networks, modeled using a two-stage stochastic mixed-integer linear programming (MILP) formulation. A machine-learning predictive algorithm prunes the scenario tree from 729 to 8 potential scenarios using real-time crisis-severity data, and a Generalized Benders Decomposition approach is implemented to solve the optimization at scale. A case study demonstrates that this adaptive model increases total income by 9.7% compared to a non-repurposable model and reduces sensitivity to purchasing cost increases.

Why it matters — It establishes a mathematical and computational method for supply chains to survive persistent, unpredictable crises by transitioning to new product lines during the disruption, rather than waiting for a return to a pre-crisis state.

Caveat: The performance and sensitivity findings are demonstrated on a single case study, which may limit immediate generalization to other supply chain structures.

Transportation Research Part C Emerging Technologies · doi

Joint optimization of electric bus scheduling and fast charging infrastructure location planning

The study presents an Integer Linear Programming formulation and four branch-and-price algorithms to jointly optimize electric bus scheduling and fast-charging station placement. The model minimizes total operational and infrastructure costs, and its performance was evaluated through computational experiments and a real-world transit case study.

Why it matters — It integrates two traditionally isolated planning problems—vehicle scheduling and charging infrastructure placement—allowing transit agencies to identify the most cost-effective electric bus types and charger locations while accounting for battery range limits and charging times.

Transportation Research Part C Emerging Technologies · doi

On-edge artificial intelligence technique for road pavement distress detection

The study developed a low-cost, non-intrusive road pavement distress detection system that runs on-board ordinary vehicles using an NVIDIA Jetson Orin Nano embedded platform. Researchers trained and compared three deep-learning architectures—YOLOv8n, RT-DETR, and Faster R-CNN with a ResNet50 backbone—on a custom dataset combining public and newly acquired images. The lightweight YOLOv8n model performed best for real-time deployment, processing 41 frames per second with an average F1 score of 0.54 across all distress classes.

Why it matters — It demonstrates that automated, real-time road damage assessment can be performed directly on edge devices inside standard vehicles, removing the dependency on expensive specialized sensors or continuous cloud connectivity for scalable infrastructure monitoring.

Caveat: The system's detection accuracy is relatively modest, as indicated by an average F1 score of 0.54 across the distress categories.

Transportation Research Interdisciplinary Perspectives · doi

The effects of split sleep on sleepiness on performance: A train simulator study

A simulator study evaluated the fatigue levels of 15 male participants during a 2.5-hour train-driving task under three sleep schedules: split sleep, consolidated daytime sleep, and baseline nighttime sleep. While EEG parameters and subjective sleepiness scales showed minimal differences, split sleep caused ocular and facial fatigue indicators to surge by 15% to 70% compared to baseline, whereas consolidated daytime sleep led to smaller increases of 9% to 31%.

Why it matters — The study demonstrates that splitting sleep into two blocks severely degrades alertness compared to a single block of sleep, even when the total hours slept are identical. This provides empirical evidence that transit scheduling must prioritize continuous rest periods rather than just cumulative sleep duration to prevent operator fatigue.

Caveat: The findings are based on a small, entirely male sample of 15 participants in a simulated environment, which may not fully capture the diverse demographic profile or real-world stressors of actual train operators.

Transportation Research Interdisciplinary Perspectives · doi

When the state doesn’t drive retirement: informal transport workers and the crisis of social protection in Ghana

Through qualitative interviews with 65 minibus and taxi drivers aged 30 to 65 across urban Ghana, this study analyzes how informal transport workers manage retirement without formal pensions. The research identifies that drivers rely on a hybrid strategy of liquidating vehicle assets, informal savings schemes, and family support, while exhibiting deep institutional distrust toward formal pension systems.

Why it matters — It conceptualizes 'negotiated social protection' to show how informal transport workers blend self-managed financial strategies, shifting the focus of transport research from physical infrastructure to the long-term welfare and life-course trajectories of the labor force sustaining Global South transit networks.

Caveat: The findings are based on a qualitative, purposively sampled group of 65 drivers, which may not represent the quantitative distribution of retirement strategies across the entire Ghanaian informal transport sector.

Transportation Research Interdisciplinary Perspectives · doi

Transit-oriented development as a catalyst for urban economic growth: evidence from the Chaoyang district of Beijing

This study analyzed survey responses from 2,346 small business owners in Beijing's Chaoyang District to evaluate how six transit-oriented development (TOD) dimensions relate to perceived property value, business growth, and employment opportunities. Using Partial Least Squares Structural Equation Modeling, the analysis found that density has the strongest association with perceived property values, while transit accessibility and proximity are most strongly linked to perceived business growth and employment opportunities.

Why it matters — It shifts the evaluation of TOD success from objective physical and economic metrics to the subjective perceptions of local business operators, demonstrating that high-density transit design aligns with positive economic expectations from the community.

Caveat: The study relies entirely on the subjective perceptions of business owners rather than objective financial or employment data.

Transportation Research Interdisciplinary Perspectives · doi

The importance of socio-economic indicators in predicting urban growth

A probabilistic hexagonal cellular automata model was developed to predict urban growth in five-year intervals, integrating satellite-derived land-cover data with regional socio-economic indicators. Calibrated and tested on the Houston urban area using data from 2001, 2006, 2011, and 2016, the model achieved an average prediction accuracy of 81% across a validation set of 200 cells.

Why it matters — The model demonstrates that the drivers of urban expansion shift based on existing density, revealing that neighborhood proximity dominates growth in low-density areas while employment and income levels are the primary drivers in highly urbanized zones.

Caveat: The model's predictive performance and the identified relationships between socio-economic indicators and urban growth were validated only within a single metropolitan area.

npj Urban Sustainability · pdf · doi

Sustainable Development of a Vertical Community in the Iao Hon District of Macau, China

This study proposes a planning framework for a vertical community in the historic Iao Hon district of Macau, China, using data gathered from field investigations, interviews, and questionnaires. The framework integrates the conservation and adaptive reuse of historic structures, public space enhancements, and resident participation to address local environmental, social, and economic challenges.

Why it matters — It provides a concrete planning model for balancing high-density vertical development with the preservation of cultural heritage and community well-being in a historic urban district.

Caveat: The study relies on qualitative planning frameworks and localized survey data without specifying the quantitative sample size, engineering feasibility, or specific metrics used to evaluate the proposed vertical community's performance.

Journal of Urban Planning and Development · pdf · doi

Carbon Storage Dynamics Driven by Land Use and Cover Change with Rapid Urbanization in the Chengdu–Chongqing Region in China

This study simulated land use and carbon storage dynamics in the Chengdu–Chongqing region of China from 2005 to 2020 and projected future changes under SSP126, SSP245, and SSP585 scenarios using the PLUS and InVEST models. The historical analysis revealed that built-up land expansion at the expense of cropland drove carbon losses, while grassland-to-forest transitions partially offset this by concentrating carbon storage in forested areas. Future projections indicate that carbon storage peaks under the SSP126 scenario, driven by an 11.97% expansion in forest cover.

Why it matters — It quantifies the specific carbon trade-offs between urban expansion and ecological restoration in a rapidly growing Chinese urban agglomeration, demonstrating that targeted forest restoration under a low-emission pathway can maximize regional carbon sequestration.

Caveat: The study relies on simulated future scenarios and simplified carbon storage coefficients in the InVEST model, which may not fully capture real-world microclimate and soil carbon dynamics.

Journal of Urban Planning and Development · doi

A hybrid machine learning and microsimulation framework for simulating metro station evacuation

This study develops a hybrid pedestrian evacuation microsimulation framework that models evacuation behaviors across low, medium, and high panic levels. The model was calibrated using Latin Hypercube Sampling, Random Forest modeling, and a Genetic Algorithm, and validated using Geoffrey E. Havers statistics against CCTV footage from a metro station. Under peak-hour conditions with 540 people, simulated clearance times ranged from 5.25–6.47 minutes under low panic, 5.12–5.25 minutes under medium panic, and 4.90–5.07 minutes under high panic.

Why it matters — It demonstrates that higher panic levels, while chaotic, can lead to faster overall station clearance times due to different temporal distributions of evacuees, providing transit operators with specific temporal evacuation profiles to design better emergency management strategies.

Caveat: The framework's behavioral parameters and validation are derived from a single metro station's CCTV data, which may limit its direct applicability to stations with different layouts or architectural constraints.

Transportation Planning and Technology · doi

The Moving Target of Urban Equity: Spatiotemporal Demand and Double Disadvantage in Hefei, China

This study introduces a temporally differentiated framework to measure urban equity using large-scale mobile phone GPS data in Hefei, China. By constructing dynamic hourly population exposure surfaces for both residential and workplace locations, the model evaluates accessibility to medical facilities and green spaces through network-based travel times and a per-capita provision metric. The analysis identifies areas of double disadvantage, where poor spatial access and low per-capita service availability overlap.

Why it matters — It demonstrates that urban service deficits are dynamic and concentrated in inner suburbs and daytime employment centers rather than static outer peripheries. This shifts the planning paradigm from static, home-based accessibility metrics to interventions that respond to hourly population fluctuations.

Caveat: The empirical findings and identified spatial patterns of double disadvantage are limited to the single case study of Hefei, China.

arXiv · pdf

Robusto-2: Benchmarking Humans & VLMs for Autonomous Driving in Lima & New York City

This study presents Robusto-2, a benchmark dataset comparing how human drivers from Lima and New York City, alongside Vision-Language Models (VLMs), answer visual question-answering prompts based on dashcam footage from both cities. The evaluation covers four question categories—factual, ratings, counterfactual, and reasoning—to test responses to out-of-distribution, edge-case driving scenarios in highly challenging urban environments.

Why it matters — It demonstrates that while human and VLM responses diverge depending on the question type, human drivers answer similarly regardless of their home city, and geography does not strongly modulate performance for either group in highly novel driving scenarios.

arXiv · pdf

The Algorithmic-Human Manager: AI, Apps, and Workers in the Indian Gig Economy

This study investigates the effects of algorithmic management on Indian ride-sharing and delivery workers using a mixed-methods approach that includes interviews with 16 gig workers and 21 key stakeholders. The analysis reveals that while automated systems improve operational efficiency and work access, they are intentionally opaque, distribute outcomes unfairly, and fail to scale compensation proportionally with increased worker effort.

Why it matters — It proposes a hybrid 'Algorithmic-Human Manager' governance framework, demonstrating how human accountability can be integrated with automated systems to protect worker dignity and fairness in the Global South's gig economy.

Caveat: The empirical findings are based on a relatively small qualitative sample of 37 total interviewees, which may not capture the full diversity of gig work experiences across India.

arXiv · pdf

Towards Graph-Based Deep Learning for Map Generalization: Insights from Building Footprints Simplification and Aggregation

This study reformulates map generalization tasks by applying graph-based deep learning to building footprints, modeling simplification as node movement prediction and aggregation as link prediction. The researchers evaluated GraphSAGE, GCN, and GAT architectures on multi-scale building datasets, finding that GraphSAGE performed best at link prediction, though precise node movement prediction remained difficult across all models.

Why it matters — It establishes the first unified graph learning framework for both simplification and aggregation, demonstrating that deep learning can capture complex spatial relationships for cartographic automation while identifying that aggregation is significantly more difficult to model than simplification.

Caveat: The models currently suffer from data imbalance and require manual or algorithmic post-processing to produce final cartographic outputs.

arXiv · pdf

Mechanism underlying the scaling law of home-return probability in human mobility

This study establishes a theoretical mechanism for the scaling law of human home-return probability, where the likelihood of returning home after visiting a set number of locations decays as a power law. By combining the principle of least effort, Zipf's law for activity priorities, and Luce's choice rule, the researchers analytically derive the exact exponent of this decay and validate the relationship using agent-based simulations.

Why it matters — It provides a microscopic, cognitive-based explanation for a widely observed empirical mobility pattern, replacing a previously assumed power-law input with a derived physical mechanism based on utility trade-offs.

arXiv · pdf