Urban Currents

Data-driven investigation of ship fuel consumption integrating causal inference and hierarchical analysis

The study develops a data-driven framework combining multi-source data fusion, causal discovery via a direct linear non-gaussian acyclic model, and double machine learning with causal forests to analyze ship fuel consumption. Testing under an expanded directed acyclic graph specification revealed that daily sailing hours has the strongest positive conditional effect on fuel consumption with an Average Treatment Effect (ATE) of 2.058, followed by main engine RPM with an ATE of 0.268. Interpretive structural modelling was then used to decompose these directional dependencies into hierarchical transmission levels.

Why it matters — It establishes a causal, graph-informed method to isolate and rank the direct and indirect drivers of maritime fuel consumption, moving beyond simple predictive correlations to support structured energy management decisions.

Transportation Research Part D Transport and Environment · doi · code

A scalable reinforcement learning-based approach to dynamic airspace sectorization

This study develops a reinforcement learning framework for dynamic airspace sectorization by modeling the task as a Markov decision process and combining three-dimensional tiled Voronoi partitioning with an actor-critic algorithm. The framework was trained and validated using three weeks of historical ADS-B trajectory data from UK airspace, covering more than 200,000 flights. Compared to a genetic algorithm benchmark, the model reduced workload imbalance by an average of 50.59% and cut decision-making time by 99.98%, from over 82 minutes down to 1.31 seconds per interval.

Why it matters — It introduces the first reinforcement learning framework for dynamic airspace sectorization, proving that sector boundaries can be optimized in near-real-time to balance controller workloads under fluctuating traffic demands.

Transportation Research Part C Emerging Technologies · doi · code

KYA: Vision–Language Assistant for Emotional Reactions to Risky Driving

The Keep Yelling Assistant (KYA) is an in-vehicle vision-language pipeline that detects high-risk driving events, like sudden cut-ins, and generates emotionally tailored verbal responses. The system uses YOLOv8 variants to extract vehicle distance, speed, and time-to-collision metrics, which are then processed by large language models (including ChatGPT-4o, Claude 3, Gemini 2.5, and Copilot) to generate feedback in neutral, humorous, or analytical tones. An evaluation using dashcam footage and a user study with 108 participants found that the combination of YOLOv8s and ChatGPT-4o achieved the highest user preference rating of 4.29 out of 5.00.

Why it matters — It introduces emotional intelligence and driver preference customization to in-vehicle AI assistants, moving beyond purely functional driving alerts to address driver comfort and psychological safety during stressful road events.

Caveat: The system's performance and user acceptance were evaluated using pre-recorded dashcam videos rather than active, real-time deployment in physical vehicles.

Transportation Research Record Journal of the Transportation Research Board · doi

Shared autonomous vehicles and inclusive mobility: perceptions of older adults and people with reduced mobility in Athens, Greece

This study surveyed 267 older adults aged 60 and over and mobility-impaired individuals in the Athens metropolitan area using paper-based questionnaires to analyze their perceptions of shared autonomous vehicles (SAVs). Using descriptive statistics, chi-square tests, and ordinal logistic regression, the research identified that road safety, service reliability, and vehicle accessibility are the primary drivers of SAV acceptance, while the lack of human intervention during emergencies and accident risks represent the largest barriers.

Why it matters — It provides empirical evidence on how vulnerable populations in a southern European context perceive autonomous transit, demonstrating that their willingness to adopt SAVs hinges on tangible service reliability, safety, and physical accessibility rather than just technological novelty.

Caveat: The findings are based on a relatively small sample of 267 participants in a single metropolitan area, which may limit generalizability to other geographic or cultural contexts.

Transportation · pdf · doi

A community-centred polykatoikia scale framework for superblock transformation in Athens

This study adapts the superblock urban design model to Kypseli, the densest neighborhood in Athens, by focusing interventions on the local 'polykatoikia' residential building typology. Using GIS and drone mapping, street-level audits, expert interviews, and a survey of 98 residents, the researchers developed 'Demokatoikia', a five-pillar framework addressing green infrastructure, water reuse, energy retrofits, universal accessibility, and cultural activation. The framework incorporates co-design and citizen stewardship mechanisms to overcome the fragmented multi-ownership structures that typically stall building-scale climate action.

Why it matters — It expands the traditional superblock model—which primarily focuses on street-level traffic calming—by integrating building-scale retrofits and community governance, offering a pathway to bypass the administrative inertia of multi-owner residential buildings.

Caveat: The proposed framework is conceptual and based on a diagnostic study of a single neighborhood in Athens, meaning its participatory and physical strategies have not yet been deployed or tested at scale.

Discover Cities · pdf · doi

Landscape pedagogy for intergenerational urban learning a design based educational framework from Alexandria Egypt

This study implemented a four-phase landscape pedagogy framework within a third-year university course in Alexandria, Egypt, using tools like Survey123, Streetmix, and cognitive mapping to document and redesign local streets. Students analyzed urban spaces through visual memory mapping during Ramadan, street selection, and landscape narratives to shift their understanding of streets from physical infrastructure to cultural landscapes.

Why it matters — It establishes a transferable educational model that uses digital participation and participatory mapping to preserve intangible urban heritage and collective memory in rapidly redeveloping cities of the Global South.

Caveat: The framework's effectiveness is demonstrated through a single case study of a university course in Alexandria, which may limit immediate generalizability to other educational or cultural contexts.

Discover Cities · pdf · doi

Spatiotemporal analysis of urban sprawl and densification in Hyderabad India across five decades

This study maps five decades of urban land-use and land-cover changes in the Hyderabad Metropolitan Region from 1975 to 2024 using multi-temporal satellite data at resolutions ranging from 5.8 m to 80 m. The object-based classification achieved over 87% accuracy, revealing a 188% expansion in built-up area from 183.05 km² to 527.23 km², an 84% loss of farmland, and a 59% reduction in vegetation. Spatial metrics show simultaneous peripheral sprawl and core densification, with population-weighted density reaching approximately 19,000 persons/km² and per capita green space varying widely from 12.40 m² in the Central zone to 31.20 m² in the West zone.

Why it matters — It establishes a reproducible data-harmonization framework for tracking long-term urban growth in data-limited contexts, revealing that overall compliance with global green space standards can mask severe spatial inequities at the sub-city scale.

Discover Cities · pdf · doi

Analyzing the city of care and uncare - Ethics of care, care labor and caring relations

This paper introduces and synthesizes three analytical frameworks for studying caring cities: care ethics, care work, and caring relations. It places these perspectives in dialogue with urban sociology to establish a systematic approach for analyzing both supportive and neglectful urban environments during crises of urbanization.

Why it matters — It provides researchers with a structured, normative vocabulary to analyze how social reproduction, democratic crises, and bodily experiences intersect in urban spaces, moving beyond fragmented debates into a unified sociological framework.

City Culture and Society · doi

Observer-related factors and the perception of cultural ecosystem (dis)services

This study surveyed 684 visitors across seven ecologically and socially diverse green spaces in Poland's Małopolska Region to analyze how human capital, cultural capital, and nature interaction characteristics shape the perception of cultural ecosystem services (CES) and ecosystem disservices (EDS). The analysis reveals that while aesthetic appreciation, self-development, and belonging are universally perceived, other benefits like spirituality and physical regeneration depend heavily on observer-related factors. Additionally, negative perceptions vary by demographic and behavioral traits, with fear and discomfort linked to gender, and anger or irritation tied to visit frequency.

Why it matters — It demonstrates that public perceptions of green spaces are not uniform, mapping specific types of emotional and psychological benefits to distinct visitor profiles. This allows planners to move beyond one-size-fits-all designs and combine physical infrastructure with targeted educational programming to serve diverse community needs.

Caveat: The findings are based on self-reported survey data from visitors already present in the selected green spaces, which may exclude the perspectives of non-users.

Landscape and Urban Planning · doi

Mobilising nature-based solutions (NbS) for coastal protection: Public preferences for hard, soft and hybrid coastal protection measures in Bali, Indonesia

A discrete choice experiment surveyed 370 respondents in Bali, Indonesia, to evaluate public preferences and trade-offs regarding hard, soft, and hybrid coastal protection measures. Using a Mixed Logit model and Latent Class Analysis, the study assessed how choices are influenced by protection level, implementation urgency, lifetime, biodiversity impacts, and construction materials. The analysis revealed a primary preference for hybrid measures, followed by hard and then soft measures, with strong support for options offering full protection, rapid deployment, long durability, natural materials, and positive biodiversity outcomes.

Why it matters — It provides empirical evidence of public support for multifunctional and nature-based coastal protection in a rapidly developing, tourism-dependent Global South context, showing that hybrid infrastructure is favored over purely hard or soft engineering.

Caveat: The findings rely on a relatively small sample size of 370 respondents within a single island province, which may limit generalizability to other coastal regions with different socio-demographic or economic profiles.

Landscape and Urban Planning · doi

Who rides and who is left behind: a multilevel analysis of new cycling infrastructure use in three Canadian cities

Using an online travel survey of 2,690 respondents in Vancouver, Toronto, and Montréal, this study analyzed the use of 46 cycling facilities built between 2020 and 2022. Bayesian cross-classified multilevel models revealed that each additional kilometer of residential distance from a facility reduces the odds of using it by 24%. This distance decay is significantly steeper for women, gender-diverse individuals, and lower-income riders, while older adults and women also show lower overall odds of use.

Why it matters — It demonstrates that physical proximity to new cycling infrastructure is a more critical barrier for historically underrepresented groups than for higher-income men. This provides empirical evidence that simply building equal amounts of infrastructure does not guarantee equitable access or usage across different demographic groups.

Caveat: The study relies on self-reported data from an online travel survey, which may introduce selection bias regarding who reports their cycling habits.

Transportation Research Part A Policy and Practice · doi

The multi-commodity vehicle routing problem with temperature requirements

This study introduces the Multi-Commodity Vehicle Routing Problem with Temperature Requirements (MC-VRP-T) to optimize routing alongside vehicle temperature settings for products with diverse thermal constraints. The authors model and solve two operational variants: a constant-temperature model using branch-and-cut and branch-and-cut-and-price algorithms, and a temperature-adjustable model where vehicles can alter their target temperature after deliveries.

Why it matters — It provides logistics operators with mathematical formulations to simultaneously optimize delivery paths and refrigeration settings, quantifying the cost savings of dynamically adjusting vehicle temperatures versus maintaining a single constant temperature.

Caveat: The performance and economic benefits of the proposed formulations are demonstrated through computational experiments rather than a real-world pilot deployment.

Transportation Research Part C Emerging Technologies · doi

Fairness-aware strategic design of station-based electric car-sharing systems

The study develops a bi-objective trajectory-based optimization model that integrates long-term strategic decisions, such as station locations, charger capacities, and fleet size, with daily operations like vehicle routing and battery management. It incorporates two fairness paradigms, service-rate disparity and max-min fairness, measured through actual group service rates across a multi-day representative-demand setting. The framework uses a branch-and-price algorithm and a diving-heuristic-based approach, which was tested using a real-world case study of Vienna.

Why it matters — It shifts the design of electric car-sharing systems from purely maximizing economic efficiency to balancing revenue with social equity, using actual realized service rates rather than static spatial accessibility as the metric for fairness.

Caveat: The model's performance and trade-offs are demonstrated using a single-city case study in Vienna, which may limit the immediate generalizability of the specific revenue-equity trade-off curves to cities with different demand patterns.

Transportation Research Part C Emerging Technologies · doi

Resilience-oriented emergency resource site optimization for an urban multi-modal public transport network

The study develops an optimization approach for locating emergency resource sites in a multi-modal public transport network, accounting for different resource types and disruption severities. The model was tested using numerical experiments on the real-world transit network of Chengdu, China, analyzing how disruptions at rail stations compare to bus stops.

Why it matters — It demonstrates that optimizing emergency resource sites specifically for a multi-modal network yields different location configurations than single-modal planning, and quantifies how these optimized sites mitigate resilience losses, which otherwise drop by 62.22% for the multi-modal network and 86.96% for the rail network under disruption.

Caveat: The performance and findings of the optimization model are demonstrated using a single city's public transport network.

Transportation Research Part D Transport and Environment · doi

‘Maritime regulations crossroads: bridging IMO’s CII and EU ETS through the EUA efficiency indicator’

This study analyzes the relationship between the International Maritime Organisation's Carbon Intensity Indicator (CII) and the European Union Emission Trading System (ETS) using voyage data from selected vessels. It calculates the CII ratings and associated EU ETS costs for these voyages to develop a new metric, the EUA Efficiency Indicator, which measures the combined impact of both regulatory frameworks.

Why it matters — The new indicator provides a unified metric to evaluate how global and regional maritime decarbonization policies interact, helping ship operators navigate the financial and compliance overlaps of co-existing regulatory systems.

Caveat: The analysis and indicator development are based on a limited selection of vessel voyage data rather than a fleet-wide dataset.

Transportation Research Interdisciplinary Perspectives · doi

Golden Hour Divide: Trauma Care Accessibility and Resource Vulnerability in Sri Lanka

This study evaluated national emergency care resilience across all 25 districts of Sri Lanka by analyzing accessibility for seven critical conditions using terrain-aware H3 hexagonal modeling. It categorized districts into four policy archetypes using unsupervised K-Means clustering based on spatial gaps, clinical need, lethality, coverage, and resource availability. The analysis revealed that spatial gaps exceed 70% in the Northern and Eastern provinces, and modeled that a 25% accessibility improvement in high-priority clusters would reduce the national need-gap by 9.65%.

Why it matters — It demonstrates that specialist scarcity, rather than physical bed capacity, is the primary driver of systemic pressure in underserved regions, shifting the policy focus from building physical infrastructure to strategically redistributing medical personnel.

Caveat: The study relies on spatial and resource modeling rather than empirical patient transit times to evaluate Golden Hour accessibility.

arXiv · pdf

DSIP: A Dynamic Coordination Planner for Signal-Free Intersections using Diffusion-Model-Based Multi-Agent Motion Planning

The researchers developed DSIP, a multi-agent motion planning framework that uses a generative diffusion process to optimize continuous trajectories for connected and automated vehicles at signal-free intersections. Evaluated using the SUMO simulator across various four-leg intersection layouts, the model outperformed both fixed-time signal control and reinforcement-learning-based controllers by reducing average delays and maintaining higher average speeds under medium- to high-density traffic.

Why it matters — It demonstrates that shifting intersection management from discrete traffic light phases to continuous, software-defined trajectory coordination can unlock latent intersection capacity and improve traffic flow without physical infrastructure expansions.

Caveat: The evaluation measures the theoretical upper-bound performance under idealized communication and execution conditions, which may not reflect real-world deployment challenges.

arXiv · pdf

UrbanCDNet: Appearance-Robust and Boundary-Aware Bitemporal Change Detection for Korean Urban Building Monitoring

The researchers developed UrbanCDNet, a Siamese convolutional neural network designed for bitemporal building change detection in complex urban environments. Evaluated on a corrected Korean AIHub benchmark dataset of 5,000 image pairs, the model achieved an F1-score of 0.7511 and an IoU of 0.6014, outperforming the ChangeFormer-MIT-B0 baseline. The model's performance gains were most pronounced on challenging subsets, improving F1-score from 0.4765 to 0.6175 in sparse-change areas (under 5% change) and from 0.6349 to 0.7285 in scenes with high photometric differences.

Why it matters — The model provides a highly precise method for detecting building changes that preserves sharp structural boundaries rather than producing coarse blobs, even when aerial images are captured under vastly different lighting conditions or when changes are highly sparse.

Caveat: The model's performance was validated exclusively on a single Korean benchmark dataset, meaning its robustness to different international architectural styles and urban layouts remains untested.

arXiv · pdf

Optimizing Nursing Care Taxi Dispatch Leveraging Integer Linear Programming Solvers and Machine Learning

The study formulates the Nursing Care Taxi Dispatch problem to account for wheelchair compatibility, user-vehicle constraints, and specific pick-up and drop-off windows. It introduces a hybrid dispatch method that trains a Transformer-based machine learning model on high-quality solutions generated by an integer linear programming solver, followed by a post-processing step to guarantee zero constraint violations. Evaluated on real-world facility data, the approach reduced vehicle operating times by up to 8% for problem sizes with fewer than 30 users compared to existing baselines.

Why it matters — It bridges the gap between computationally expensive exact solvers and fast but constraint-violating neural routing models, providing a viable method for highly constrained specialized transit dispatching.

Caveat: The performance gains and execution times were evaluated on relatively small problem sizes of fewer than 30 users, and scalability to larger fleet operations remains unproven.

arXiv · pdf