Urban Currents

DSETA: A Dual-Stage Continual Learning Framework for Travel Time Prediction in Dynamic Traffic Environments

The researchers developed DSETA, a dual-stage continual learning framework for travel time prediction that splits model updates into real-time intra-day learning and short-window inter-day learning. The framework incorporates a knowledge consolidation module to prevent catastrophic forgetting of regular traffic patterns. Tested in online A/B experiments on the DiDi ride-hailing platform across Beijing, Wuhan, and Xi'an, the system reduced Mean Absolute Error by 6.62%, 0.73%, and 2.40% respectively, and now processes hundreds of millions of daily requests in production.

Why it matters — It establishes a method for updating travel time models that adapts to sudden, short-term disruptions like accidents or holidays without losing track of long-term seasonal trends and network evolution.

arXiv · pdf

Green Infrastructure and Housing Prices: Heterogeneous Effects and Sensitivity to Modeling Choices

This study compares traditional OLS difference-in-differences (DID) models with Double Machine Learning (DML) to evaluate the property value impacts of four green infrastructure projects in Cleveland. The analysis reveals highly divergent outcomes across the sites: one trail showed a 25 percent price increase, a nature preserve showed no effect, a park near low-income households led to a localized 12 percent decline, and a suburban reservation saw a sign flip from a positive 4 percent OLS estimate to a negative 18 percent DML estimate.

Why it matters — It demonstrates that the capitalization effects of green infrastructure are highly sensitive to statistical modeling choices and are not universally positive, challenging the common assumption that green infrastructure inevitably triggers green gentrification.

Caveat: The findings are based on a single shrinking, post-industrial legacy city, which may limit generalizability to high-demand housing markets.

Urban forestry & urban greening · doi

What counts as urban smartness? A longitudinal thematic analysis of smart city indexes

This study tracks the conceptual evolution of smart city evaluation frameworks between 2017 and 2022. The researchers conducted a longitudinal thematic analysis of six prominent smart city indexes, selected from an initial inventory of sixteen frameworks.

Why it matters — It documents a structural shift in how urban smartness is defined and measured, showing that evaluation frameworks have transitioned from purely technocentric metrics toward integrated logics that incorporate governance capacity, economic conditions, and societal dimensions.

Caveat: The analysis is limited to a qualitative thematic tracking of six selected indexes, which may not capture the full diversity of localized or informal smart city metrics used by individual municipalities.

Journal of Urban Management · pdf · doi

Advancing smart school initiatives: A case study of Zahedan city

This mixed-methods case study evaluated Iran's Smart Schools Project in Zahedan using surveys from 150 participants, 10 policymaker interviews, and classroom observations across urban, semi-urban, and rural schools. It identified severe infrastructure gaps, such as a 1:175 device-sharing ratio in rural areas, and deployed low-cost interventions including solar microgrids, offline learning-management systems, and Raspberry Pi devices. These targeted interventions increased rural literacy by 25% and STEM engagement by 18%, while community engagement with religious leaders raised parental approval of the technology from 20% to 65%.

Why it matters — It demonstrates that low-cost, culturally attuned interventions and decentralized teacher training can successfully bridge the urban-rural digital divide in highly resource-constrained educational settings without requiring massive national infrastructure overhauls.

Caveat: The findings and specific intervention outcomes are based on a single, socioeconomically heterogeneous case study city in Iran, which may limit direct generalizability to other national contexts.

Journal of Urban Management · doi

From design attributes to social life: Modelling conviviality in urban underground spaces

This study models how seven spatial-perceptual design dimensions influence perceived conviviality in urban underground spaces, using Partial Least Squares Structural Equation Modeling (PLS-SEM). Analyzing structured survey data from the Theatre-Shahr underground complex in Tehran, the model tests how physical attributes affect social interaction and willingness to use, which in turn shape conviviality. The results show that social interaction is the primary driver of conviviality, with accessibility, environmental comfort, and safety exerting the strongest indirect effects, while identity, meaning, structure, and form had no statistically significant impact.

Why it matters — It demonstrates that physical design alone does not make underground spaces convivial; instead, design elements must actively facilitate social interaction to successfully transform these subterranean transit and commercial hubs into inclusive public spaces.

Caveat: The findings are based on a single case study of a cultural and transit underground complex in Tehran, which may limit their generalizability to other underground settings with different cultural or functional profiles.

Journal of Urban Management · pdf · doi

A Freight Transit Network Model to Optimize Seaport Access for the Landlocked Nation Nepal

The study develops a weighted-sum scalarised mixed-integer linear programming model to evaluate road-only and rail-road freight transit routes connecting Kathmandu, Nepal, to seaports in Kolkata (India), Chittagong (Bangladesh), and Tianjin (China). The framework evaluates these corridors across multiple policy scenarios prioritizing transport cost, carbon emissions, transshipment steps, and delivery lead times. The analysis identifies the Kolkata-Kathmandu corridor as the most efficient route due to its shorter inland distance and favorable rail-road configuration, while the Tianjin route is structurally disadvantaged by its extreme distance despite superior logistics infrastructure.

Why it matters — It provides a corridor-level decision-support framework tailored to the specific geopolitical and logistical constraints of a landlocked nation relying on transit countries, shifting planning focus from generic route optimization to targeted cross-border coordination and rail utilization.

Transport Policy · pdf · doi

Flight Frequency, Schedule Differentiation, and Route Revenue: Evidence for Competition and Scheduling Policy

This study analyzes how flight frequency, between-carrier schedule differentiation, and within-carrier schedule dispersion jointly affect passenger volume, airfares, and route revenue. Using a monthly panel dataset of 349 carrier-route units across 102 oligopolistic domestic routes in Australia from January 2015 to June 2025 (excluding the COVID-19 period), the analysis reveals that higher flight frequency consistently boosts passengers and revenue. It also finds that spreading schedules apart from competitors increases demand and airfares, while clustering a single carrier's own flights increases passenger volume without significantly affecting fares.

Why it matters — The findings demonstrate that temporal scheduling strategies directly influence market power and service convenience, proving that airlines can maximize revenue by combining high flight frequencies with schedules that are clustered internally but differentiated from competitors.

Caveat: The analysis is restricted to oligopolistic domestic routes within a single national market, which may limit how well these scheduling dynamics generalize to highly competitive international markets or different regulatory environments.

Transport Policy · doi

Spatial heterogeneity in the distance-decay pattern of migration: facts and explanation from intercity migration network in China

Using intercity migration data from the 2005 and 2015 Chinese bi-censuses, this study applies an origin-constrained competing destination model and multiple linear regressions to analyze distance sensitivity across different cities. The analysis reveals significant spatial heterogeneity in how distance deters migrants, showing that underdevelopment at the origin restricts migration to shorter distances, while advanced regional development reduces the need for long-distance moves.

Why it matters — It demonstrates that the gap in distance sensitivity between different origin cities is narrowing over time, suggesting that local urban development can reduce geographic and administrative inequalities in migration accessibility.

Caveat: The findings are based on macro-level census data from 2005 and 2015, which may not capture more recent shifts in migration dynamics or individual-level decision-making processes.

npj Urban Sustainability · pdf · doi

Urban high-rise compact form amplifies humid heat risk from vegetation transpiration

This study analyzed 251 rapidly developing Chinese cities, categorized into four background climate zones, to measure how three-dimensional urban form influences the coupled cooling and humidifying effects of vegetation transpiration on wet-bulb temperature. The analysis revealed that in high-rise compact urban forms, transpiration raises wet-bulb temperatures by increasing local humidity without providing significant temperature cooling, resulting in a population-weighted exposure intensity 23 times higher than in low-rise dispersed forms.

Why it matters — It demonstrates that urban greening in hot, humid climates can counterproductively increase human heat stress depending on the surrounding physical architecture, challenging the universal assumption that urban vegetation always mitigates heat risk.

Caveat: The study relies on wet-bulb temperature as the primary metric for heat stress, which may not capture the full suite of human thermal comfort variables.

npj Urban Sustainability · pdf · doi

What is enough? Rethinking sufficiency in transport through relational egalitarianism

The paper develops a theoretical framework for transport justice by integrating relational egalitarianism into the concept of accessibility sufficiency. It introduces a relational sufficiency threshold alongside a limitarian principle to constrain how socio-economic privilege is converted into durable accessibility advantages.

Why it matters — It shifts the focus of transport justice from merely preventing basic participation poverty to addressing how accessibility disparities above the poverty line reinforce social hierarchies and second-class status.

Caveat: The paper is entirely theoretical and does not provide empirical measurements, spatial data, or quantitative thresholds for the proposed principles.

Transport Reviews · doi

Bisexual geographies: mapping bisexual experiences and imagining non-monosexual spaces in Catalonia

This study maps the spatial experiences of 140 self-identified bisexual individuals out of 1,009 total survey participants in Catalonia, Spain. Using the ReliefMaps model to integrate quantitative and qualitative data, the research measures spatial discomfort across five key domains: home, public space, leisure, virtual spaces, and healthcare.

Why it matters — It demonstrates that bisexual spatial discomfort stems from a lack of social intelligibility and monosexual assumptions rather than explicit biphobia, offering a new framework for how geographies of sexuality conceptualize non-monosexual identities.

Caveat: The findings rely on a self-selected sample within a single region, which may limit generalizability to other cultural or geographic contexts.

Social & Cultural Geography · doi

Politics of mobility and the urban segregation in Shanghai, China

This study analyzes urban segregation in Shanghai using high-resolution data on housing tenure, housing costs, real-time commuting times, and job opportunities. It maps a socio-spatial division of mobility across three distinct zones: a privileged central zone, a comfortable suburban zone, and a marginalized outer zone.

Why it matters — It demonstrates that urban segregation is not merely a product of individual choices or static residential locations, but is actively produced and maintained by historically institutionalized urban governance that prioritizes strategic central zones over marginalized populations.

Caveat: The findings are based on the specific monocentric structure and historical institutional context of Shanghai, which may limit direct generalizability to cities with different governance models.

Mobilities · doi

Paris as a 15-Minute City: An Explainable AI Perspective

This study analyzed approximately 70,000 trip segments from the NetMob 2025 Data Challenge in the Paris metropolitan area, combining them with sociodemographic data and OpenStreetMap points of interest. Using gradient-boosted tree models and explainable AI methods, it measured how walking and cycling accessibility to local services relates to trip duration, transport mode, and short-trip car use. The models revealed that while high service density correlates with increased active mobility and reduced private vehicle use, this relationship weakens significantly in outer suburban areas.

Why it matters — It provides empirical, model-based evidence validating the core assumptions of the 15-minute city concept, while demonstrating how explainable machine learning can map the spatial and demographic limits of these planning policies.

Caveat: The findings rely on mobility trajectories from a specific data challenge and may reflect the unique spatial structure and transit network of the Paris metropolitan area.

arXiv · pdf

In-Context Learning to Assess Built Environment Impacts on Perceived Neighborhood Walkability Among Mobility-impaired Older Adults

This study applied the TabPFN foundation model using in-context learning to predict perceived neighborhood walkability among 257 older adults with knee osteoarthritis or fall histories. Using survey data from the NEWS-A instrument, the model achieved a macro F1 score of 54.89% in classifying walkability perceptions into three categories, outperforming XGBoost and Random Forest baselines. Shapley Interaction Quantification revealed that higher-order feature interactions, such as street circuity combined with drivable road ratios, drove the model's predictions.

Why it matters — It demonstrates that transformer-based in-context learning can outperform traditional machine learning baselines on small-scale, highly specialized urban health datasets. Additionally, it reveals that environmental features like neighborhood greenery do not influence walkability perceptions in isolation, but rather in combination with individual psychological factors like a fear of falling.

Caveat: The findings are based on a small, highly specific demographic sample of 257 older adults with mobility impairments, which may limit generalizability to the broader aging population.

arXiv · pdf

Beyond Lanes: Traffic Flow Dynamics in Disordered Conditions Based on High-Resolution Trajectory Data

This study analyzes disordered traffic flow lacking lane discipline using high-resolution UAV trajectory data collected from an urban arterial. By applying a two-dimensional extension of Edie's framework, the research quantifies aggregate traffic variables to construct a 2D fundamental diagram and tracks congestion propagation through spatiotemporal speed fields. At the microscopic level, it identifies steady-state follower-leader pairs to measure desired time gaps, minimum lateral spacing, and vehicle dimension distributions across heterogeneous vehicle classes.

Why it matters — It demonstrates that disordered traffic states cannot be accurately represented by traditional one-dimensional, lane-based formulations due to continuous lateral redistribution, yet shows that macroscopic stop-and-go waves propagate similarly to conventional lane-based systems despite highly heterogeneous vehicle interactions.

Caveat: The empirical findings and framework are derived from a single urban arterial corridor, which may limit immediate generalization to other disordered traffic environments.

arXiv · pdf